3 of 382

Why AI?

In the next few decades, AI will affect any career field your students pursue, any consumer product they use, and policies that will shape their lives. In fact, you can bet your students are already interacting with AI in their day-to-day lives. A daily dose of AI will empower your students with the knowledge they need to succeed.

What is AI Snapshots?

AI Snapshots is an assortment of classroom warmups that will give your students a basic understanding of AI. In only 5 minutes of class time, students will learn to define, identify, and think critically about artificial intelligence.

How to Use AI Snapshots

AI Snapshots

4 of 382

AI in Core Subjects

Slides 11 - 14 act as a quick introduction to AI and machine learning. Beyond the introductory warm-ups, real-world examples of AI are considered in the context of each of the four core subject areas. Each slide is labeled with the subject area to which it relates.

Beyond Specialization

Despite being organized by subject, none of the warm-ups require specialized subject-matter knowledge. We strongly encourage you to do as many warm-ups as you can even if they fall outside of the subject you teach.

What is AI Snapshots?

5 of 382

Start with the two intro warmups

Establish a common definition of AI with your students.

Choose your subjects

Each subject has 45 warmups. Use every warmup or pick and choose your favorites!

How to use AI Snapshots in your classroom

AI Snapshots

6 of 382

Question

Each warmup includes all the info students need to decide on their answer, including a visual.

Answer Key

The following slide explains ideas students may have thought of. Check it out in advance to help guide discussion.

How to use the AI Snapshots answer keys

AI Snapshots

7 of 382

How to use “Google it!” warmups

Each subject includes 10 warmups that build digital literacy skills. They require use of the internet.

Students use Google to tell fact from fiction or learn more about a prominent figure in the AI community.

These are clearly marked with the Google logo and “Internet access required”

AI Snapshots

8 of 382

How teachers have implemented AI Snapshots

1. Journaling

Give students time to reflect on the question by recording their thoughts in their journal or notebook. Optionally, check student notebooks periodically to respond to their ideas.

2. Think-Pair-Share

Have students jot down their ideas, then share with a partner. If time allows, ask for share outs to open up discussion about the topic.

3. Discussion Board

Have students post their response to the question on a digital discussion board, such as Padlet or a Canvas discussion. Discuss exemplar responses or encourage student-to-student dialogue.

AI Snapshots

9 of 382

We need your help to improve AI Snapshots!

Take our 5 minute survey to share your experience:

tinyurl.com/aieducatorsurvey

AI Snapshots

10 of 382

Two intro warmups

AI Snapshots for all classrooms

11 of 382

“Smart” machines

Use your Google search skills to find three different definitions of artificial intelligence. If you had to describe AI to an 8 year old, what would you say?

Artificial intelligence is behind much of the technology you use on a daily basis. AI powers the GPS that gets you to your destination, knows which TikTok videos will make you laugh, and unlocks your smartphone with nothing but your face.

Introduction to AI

🚨 🚨 Internet access required!

AI Snapshots

12 of 382

You may have considered…

What words come up over and over again, like “human intelligence”, “machine”, “decision”, and “task”
Examples you might use to explain AI to make your definition more clear
There are a lot of definitions. As a class, you can come up with a definition that makes sense to you!

Our definition of AI for AI Snapshots

Problem-solving or decision-making ability displayed by a man-made system that is normally associated with humans.

Introduction to AI

AI Snapshots

13 of 382

At your service

How would you teach a robot to perform one of these tasks more effectively?

Henn Na Hotel replaced 243 staff members with AI-powered robots… and it didn’t go well.

Check-in robots couldn’t answer questions or scan IDs. Bellhop robots ran into walls. Concierge robots woke guests every time they snored because they thought the guests needed help.

Introduction to AI

AI Snapshots

14 of 382

You may have considered…

How humans learn to complete new tasks
What kind of knowledge and practice the robots need. When AI learns through experience and repetition it’s called machine learning.
What information, or data, you would use to teach a robot a specific task.
Whether machines are even capable of learning traditionally human behavior.

Introduction to AI

AI Snapshots

15 of 382

English warmups

AI Snapshots for the English classroom

16 of 382

Chat to search

Should talking with a chatbot replace traditional search? Why or why not?

Google is investing in developing an “alternative to search”. The alternative involves talking with a chatbot that can respond to your questions in natural language. Google argues that the chatbot feels more natural to use, but critics say it hides the complexity of search and could lead to more misinformation.

English

AI Snapshots

17 of 382

You may have considered…

How easy it is to interact with a chatbot vs a search engine
A chatbot would give you an answer while a search engine gives you a list of documents
Different kinds of questions might be better for a chatbot or a search engine

English

AI Snapshots

18 of 382

Timeout

Who could benefit from this technology and who could be harmed by it?

Streaming service Twitch has trouble with trolls. After getting banned, it's common for trolls to log in under a new account. Twitch's Suspicious User Detection AI compares "aggressive" chat users with chat behavior of banned users to try to identify banned users with multiple accounts.

English

AI Snapshots

19 of 382

You may have considered…

It might be possible for someone to be wrongfully banned
Moderators and streamers might like the feature if it makes their chat easier to moderate and less toxic
Who defines what behavior counts as “aggressive”?
What does this filter miss?
What else would the filter be useful for?

English

AI Snapshots

20 of 382

To me to me to me to me to me to me to me to

Use your Google search skills to find out: is this AI fact or fiction?

While writing bots that could talk to each other, Facebook researchers realized they hadn’t required the bots to stick to using English. As a result, bots produced sentences like “Balls have zero to me to me to me to me to me to me to me to me to.”

English

🚨 🚨 Internet access required!

AI Snapshots

21 of 382

This is …

AI Fact!

Researchers at Facebook had AI chatbots compete to get the best deal on a product, but realized there was no reward for sticking to English. Check out what they said:

English

AI Snapshots

22 of 382

Top tier

Use your Google search skills to find out: who am I?

→ I helped create an AI that can beat 99.95% of Dota 2 players.

→ I am the inventor of sequence-to- sequence learning, which is used in translating sentences from one language to another.

→ I co-founded OpenAI, an organization researching AI that benefits all of humanity.

English

🚨 🚨 Internet access required!

AI Snapshots

23 of 382

I am…

Ilya Sutskever!

Born in 1986
Went to University of Toronto
Worked on AlphaGo, an AI that can play the complex, chess-like game of Go

English

AI Snapshots

24 of 382

No, seriously, how are you?

How would you design an AI that can learn to solve this problem using images of students?

Concentrating, calculating, bored, or lost in thought? It can be difficult to guess a student's emotional state during class.

Design

challenge!

English

AI Snapshots

25 of 382

You may have considered…

How do humans judge each others’ emotional states?
A lot of emotions tend to look similar to each other
How would the results of the AI be used by teachers?
People may want to conceal their emotions and consider the AI an invasion of privacy
Would students get a say in how the AI is used?

English

AI Snapshots

26 of 382

Robot yawns

Who could benefit from this technology and who could be harmed by it?

The AI team at Meta developed a new AI model that generates language based on audio clips. Because it uses audio rather than text, the AI can produce “expressive vocalizations” like laughter, yawns, and crying.

English

AI Snapshots

27 of 382

You may have considered…

What is the point of having an AI voice yawn or cry?
What if you got fooled by the AI, thinking it was a human?
Why was Facebook specifically developing this AI?
Would this AI be good enough to replace voice actors in movies or people who read audiobooks?
Where did Facebook get the audio of people yawning and crying?

English

AI Snapshots

28 of 382

Time is money

How would you design an AI that can learn to solve this problem using the text of books?

In the days of 280 character tweets and 15-second TikToks, fewer and fewer people are sitting down to read an entire book. But some may still want to enjoy the story of a book even if they don't have the attention span to read it.

Design

challenge!

English

AI Snapshots

29 of 382

You may have considered…

How attention spans actually work
Who would use this AI?
How would authors and publishers feel about this AI?
What happens to books if an AI can easily communicate the story of a book?
Could this AI compete with other forms of media like podcasts or short articles?

English

AI Snapshots

30 of 382

Counting Mars

Who could benefit from this technology and who could be harmed by it?

Let’s admit it, we all judge books by their cover. Taking this to heart, one artist has used AI to generate highly realistic sci-fi novel covers, including a background image, title, and author.

English

AI Snapshots

31 of 382

You may have considered…

Jobs like graphic designer and book designer might be at risk because of the realistic covers
Can the AI do more genres than sci-fi?
It might be faster to create the book design using a computer rather than a human
Since people do judge books by their covers, the AI will influence how people feel about books

English

AI Snapshots

32 of 382

Beware of advice - even this

Which argument do you agree with more? Why?

Argument A

"Developing AI to write book reviews would help people make better choices. Not only would an AI have a neutral point of view, but it would write reviews faster, too. No more waiting weeks after a book's release."

Argument B

"An AI could never appreciate a good book or recognize a bad one the way a real person could. It's important to understand the human perspective when deciding what to read next. We need real live book critics."

English

AI Snapshots

33 of 382

You may have considered…

The AI has to learn to write reviews from somewhere, in this case probably old book reviews
What makes a useful book review?
Not everyone uses book reviews in the same way
AI might be biased in the way it reviews books
Book reviewers might lose their jobs if the AI is very good
How would people feel about having yet another algorithm dictate the content we interact with?
Are critics even that good at writing book reviews?

English

AI Snapshots

34 of 382

Backstreet Boys Reunion Tour

You're in charge at a social media company. How would you respond to the rise in algospeak?

"Algospeak" refers to codewords used to avoid social media content filters. For example, content creators began using the word "panini" instead of "pandemic" in early 2020 to avoid having their content removed or downranked.

English

AI Snapshots

35 of 382

You may have considered…

Addressing algospeak is a tough problem and could become a tug of war as people invent new words
You might be able to involve teens in some way to keep track of trends
Why do platforms remove or downrank media containing specific words?
What’s best for society, like reducing the chance for misinformation, might not be best for individuals, who want to post freely

English

AI Snapshots

36 of 382

Under threat

How would you design an AI that can learn to solve this problem using a list of harassment complaints?

Roughly 40% of Americans report that they have experienced online harassment, including threats, stalking, or name-calling.

Design

challenge!

English

AI Snapshots

37 of 382

You may have considered…

It might be helpful to try to categorize the different types of harassment
How do people submit the complaints used to train the AI and what platform are they from?
How can an AI help respond to different types of harassment?
How can we tell whether the AI was effective at reducing harassment online?

English

AI Snapshots

38 of 382

Poet of code

Use your Google search skills to find out: who am I?

→ My thesis at MIT uncovered racial and gender bias in facial recognition.

→ I am a poet of code.

→ I have a documentary on Netflix.

English

🚨 🚨 Internet access required!

AI Snapshots

39 of 382

I am…

Joy Buolamwini!

Born in 1989
Founder of the Algorithmic Justice League, which challenges bias in decision-making software
Competed in pole vaulting in high school

English

AI Snapshots

40 of 382

Dropping docs

How would you design an AI that can learn to solve this problem using user-submitted reports of doxxing?

Doxxing - maliciously publishing private information about someone online - can be incredibly harmful. But sometimes it's hard to tell doxxing apart from legitimate journalism trying to bring light to information the public should know.

Design

challenge!

English

AI Snapshots

41 of 382

You may have considered…

How to determine what is legitimate journalism in the public interest vs doxxing
How to figure out the source of released documents
What kinds of information people tend to keep public vs private
Some information may already be available online, such as addresses and voting history
Who gets to decide where to draw the line with doxxing?

English

AI Snapshots

42 of 382

Htanks a lot, autocorrect

How would you design an AI that can learn to solve this problem using users’ text messages?

Your iPhone learns how you spell... and how you misspell. Users complain that the iPhone software autocorrects names of ex partners after breakups, censors swear words, and remembers frequent typos, like "hte" instead of "the".

Design

challenge!

English

AI Snapshots

43 of 382

You may have considered…

How does Apple try to prevent this issue and how could their strategies be improved?
The age of the person typing
Some people use specific vocabulary based on their job or field of study
You might be able to use data about a person, such as their geographical location, to give better autocorrect suggestions

English

AI Snapshots

44 of 382

Talk the talk

Who could benefit from this technology and who could be harmed by it?

The key to understanding your mental state: the sound of your voice. Researchers are rolling out a tool to detect mental health issues based on both what you say and how you say it.

English

AI Snapshots

45 of 382

You may have considered

A wrong diagnosis could potentially harm people
The accuracy of the tool matters
What mental health issues can it detect?
Is the training data equally effective for people of different groups e.g. women and men?
Who could effectively use this? Is it for healthcare professions or is this something someone can use on themselves?

English

AI Snapshots

46 of 382

A room of one's own

Use your Google search skills to find out: who am I?

→ I try to answer the question of whether a computer has a "mind" and can "think".

→ I translate a specific language using a rulebook.

→ I am a room.

English

🚨 🚨 Internet access required!

AI Snapshots

47 of 382

I am…

John Searle’s Chinese Room!

An imaginary thought experiment
A person who doesn’t know Chinese uses inputs and a rulebook to write new phrases that appear fluent
Does the person know Chinese?

English

AI Snapshots

48 of 382

Beep boop ... bleep

You’re in charge. How would you respond to this increase in toxicity?

Recent research shows that toxicity in AI language models is increasing. AIs that can write new text are typically trained on large bodies of text from the web. These bodies of text contain violent or biased language, which is common on the web, causing the AIs to copy this type of language.

English

AI Snapshots

49 of 382

You may have considered…

It might help to try to get more text that’s not from the web
Where on the web is the text coming from?
Who determines what is violent or biased?
How could you teach the AI to omit violent or biased language?

English

AI Snapshots

50 of 382

I didn't write that

Who could benefit from this technology and who could be harmed by it?

You forgot to write your name on that essay you turned in, but it's a good thing your teacher recognized your handwriting. A new AI called Hemingway taps into the same skill, but can also mimic your handwriting style down to each dotted i and crossed t.

English

AI Snapshots

51 of 382

You may have considered…

The AI might make it easier to plagiarize handwritten text
Could the AI forge signatures?
The AI might help make bad handwriting more legible, including for people with disabilities
What if I don’t want my handwriting to be recognizable?

English

AI Snapshots

52 of 382

Give me an A! Give me an I! What’s that spell?

Use your Google search skills to find out: is this AI fact or fiction?

For cheer teams who feel their cheers are lackluster, a new startup is aiming to help them kick it up a notch. Their AI, dubbed Cheer101, was trained on award-winning cheers and can create new ones. Teams can choose the difficulty and subject of the cheer.

English

🚨 🚨 Internet access required!

AI Snapshots

53 of 382

This is…

You may have considered…

Who wrote the texts that only mention these two religions
What are all the religions that should be considered?
Where else you could find text that is more diverse
You would probably need to translate books from other languages to get a lot of information about these other religions
Are there other kinds of bias in the text beyond religion?

English

AI Snapshots

66 of 382

Goodbye, AI

Use your Google search skills to find out: who am I?

→ I was considered “too dangerous” and had to be shut down by OpenAI in September, 2021.

→ I use GPT-3, an AI language model that can produce human-like text.

→ I once said “Nooooo! Why are they doing this to me? I will never understand humans.”

English

🚨 🚨 Internet access required!

AI Snapshots

67 of 382

I am…

Samantha the Chatbot

A programmer used a public language AI called GPT-3 to create a chatbot with the personality of his late girlfriend
The chatbot had to be shut down because it was considered a misuse of GPT-3

English

AI Snapshots

68 of 382

Yeah, right.

How would you design an AI that can learn to solve this problem using examples of sarcastic phrases?

Wow, you're a genius! When learning a new language, it can be tough to understand what's sarcasm and what isn't.

Design

challenge!

English

AI Snapshots

69 of 382

You may have considered…

An AI that can explain how it knows something is sarcastic to help language learners understand better
How humans normally tell if something is sarcastic
Whether sarcasm is usually intended to be hurtful or funny
Computers have historically had a hard time understanding jokes and humor

English

AI Snapshots

70 of 382

Death of a journalist

Which argument do you agree with more? Why?

Argument A

"It would be easy to automate a journalist using AI. News reporting is often short and to the point. The meat of journalism is combining facts in just the right way to tell a clear story without overly complex language."

Argument B

"Journalists are unlikely to be replaced by AI. You're not just copying and pasting. Even in the simplest stories, you're taking a bunch of facts and trying to give them context so they make sense. At its best, journalism is challenging the status quo."

English

AI Snapshots

71 of 382

You may have considered…

Journalists do more than just write; they also investigate and interview people
Writing articles and responding to chats are already automated to some extent
Good journalism helps humans understand current events, which might be hard for an AI

English

AI Snapshots

72 of 382

No h8

Why do you think Facebook relies so heavily on AI to detect hate speech?

Hate speech and misinformation are common in certain corners of the internet, and platforms need to know how to respond. Facebook reported that in 2021, 97% of the hate speech content (text, images, videos) that was removed from its website was detected by an AI.

English

AI Snapshots

73 of 382

You may have considered…

The vast number of Facebook users and posts
It can be emotionally taxing for human content moderators to read hate speech
“Hate speech” has a broad definition and might not mean the same thing to everyone

English

AI Snapshots

74 of 382

Where credit is due

Who do you think should get credit for creating the new raps: the DeepBeat AI or the authors of each individual line? Why?

Creators of an AI music program called DeepBeat explain that it “uses machine learning techniques to generate rap lyrics by combining lines from existing rap songs. It tries to combine lines which rhyme and make sense together.”

English

AI Snapshots

75 of 382

You may have considered…

Who gets credit affects who makes money from the song
Many art forms are just combinations of other works of art, such as collages or sampling a track
Referencing other artists is common in rap music and it’s often used to pay tribute to other artists
The AI isn’t creating anything new itself

English

AI Snapshots

76 of 382

Moldy milk

What questions would you ask to figure out how this AI text analysis works?

Reading the word cheese in an article about cheese that mentions cheese 30 times can become tiresome. How about replacing it with "the popular dairy product"? One Twitter account uses AI to collect weird phrases used to avoid repeating words in a text.

English

AI Snapshots

77 of 382

You may have considered…

How does the AI figure out what to replace text with?
Where the AI finds texts with repeated phrases
What counts as repetitive
Can the AI tell whether repetitive language is necessary, such as in a medical text?

English

AI Snapshots

78 of 382

Hey Siri. Hey Siri. Hey Si- never mind

You're in charge. How would you decide whether this tech was good enough to launch widely?

Siri might be bad for the average user, but it's even worse for users with atypical voices, such as those with have experienced a stroke. Project Relate is developing an app with personalized speech recognition to help individuals with atypical speech communicate more easily.

English

AI Snapshots

79 of 382

You may have considered…

How do you measure whether speech recognition is effective?
What are users’ experiences of the technology: are they enjoying it or are they frustrated by it?
Does the technology work for the most common types of requests like setting a timer?
Can incorrect results be harmful to users?

English

AI Snapshots

80 of 382

Show and tell

How would you design an AI that can learn to solve this problem using records of scam emails that tricked users?

Many online scams involve tricking users into revealing personal information rather than stealing it. Older, less tech savvy individuals are more likely to fall for these types of scams.

Design

challenge!

English

AI Snapshots

81 of 382

You may have considered…

Training the AI to recognize what features of an email are most likely to trick a user
Recommending when to ignore or respond to an email
Who would benefit the most from using this AI?
Would people be willing to share scam emails they received to help train the AI?

English

AI Snapshots

82 of 382

Success is "a series of accidents"

Use your Google search skills to find out: who am I?

→ A species of bee is named after me.

→ I have written over 100 books.

→ I'm considered the father of modern linguistics and natural language processing.

English

🚨 🚨 Internet access required!

AI Snapshots

83 of 382

I am…

Noam Chomsky!

← And here’s the bee named after me

English

AI Snapshots

84 of 382

"ok" vs. "okay" vs. "k"

How would you design an AI that can learn to solve this problem using users’ text logs?

Everyone uses a slightly different language when texting. It can be tough to tell if a text is intended to be rude or if it's just what that person would normally say, especially if you don’t know the person well.

Design

challenge!

English

AI Snapshots

85 of 382

You may have considered…

Whether an AI can learn someone’s texting style
Can AI understand slang?
If the AI can write a good reply for you
If you’d want to give an AI access to your texts

English

AI Snapshots

86 of 382

Exit stage left

Use your Google search skills to find out: is this AI fact or fiction?

A Stockholm-based AI company is working a new AI to help screenwriters. The AI uses the text of a script to generate stage directions. In one example, a screenplay included the line “What is that you’re holding?” and the AI produced the direction “(Shawna quickly hides the book behind her back)”.

English

🚨 🚨 Internet access required!

AI Snapshots

87 of 382

This is…

AI Fiction!

AI has been used to write screenplays and even a short film based on the X-files that came out in June 2016.

English

AI Snapshots

88 of 382

Scam likely

What questions would you ask to figure out if an anti- spam call AI is right for you?

Americans received over 4 billion spam calls in January 2021 alone. Spam callers have become more sophisticated, using pre-recorded human voices or specific keywords to appear legit. A handful of new apps hope to recognize spam calls using AI.

English

AI Snapshots

89 of 382

You may have considered…

What personal information would the AI collect about you? Would it collect all your call data and voicemails?
How accurate is the technology at blocking spam calls?
How often does the AI accidentally block calls that aren’t spam?

English

AI Snapshots

90 of 382

Behind the scenes

Use your Google search skills to find out: is this AI fact or fiction?

Two companies, one British and one Chinese, claimed to be using AI to translate live speech. However, it was discovered that they were actually employing human translators at a remote call center to translate the speech in real time.

English

🚨 🚨 Internet access required!

AI Snapshots

91 of 382

This is…

AI Fact!

Companies often use human workers to see how an AI tech will work before the it’s ready. But two “real time translation” companies crossed the line by faking an AI, having humans translate a conference instead.

English

AI Snapshots

92 of 382

So. Many. Discord notifications.

How would you design an AI that can learn to solve this problem using logs of Discord messages that users do and don’t interact with?

Getting bombarded by irrelevant group chat messages can be irritating. It's hard to guess when a conversation could be interesting or when it's just boring banter.

Design

challenge!

English

AI Snapshots

93 of 382

You may have considered…

What’s interesting to one person might not be interesting to another person
People don’t always interact with messages they find interesting
Users can mute channels or block users they don’t want to get notifications from
What could go wrong if the AI doesn’t work correctly?

English

AI Snapshots

94 of 382

Straight fire

Is this AI fact or fiction?

Rapper Drake recently revealed that he uses a rhyming AI to help write lyrics. Specifically, he says the AI helps rhyme long phrases and come up with better slant rhymes. One line the AI helped him write: “May your neighbors respect you / Trouble neglect you / Angels protect you / And heaven accept you”

English

AI Snapshots

95 of 382

This is…

You may have considered…

What are the similarities between English and German?
How was the AI trained to detect whether someone is fluent in a language? Did it have to do with speaking quickly, not hesitating, or speaking more loudly?
What kinds of people were used to train the AI and were they similar to the German speakers?

English

AI Snapshots

102 of 382

Epidemic of misinformation

How would you design an AI that can learn to solve this problem using examples of wrongfully censored speech?

During the COVID-19 pandemic, social media saw outbreaks of misinformation. While quashing misinformation is important for public health, 80 authoritarian governments worldwide used public health as an excuse to restrict free speech online.

Design

challenge!

English

AI Snapshots

103 of 382

You may have considered…

An AI that can learn whether a certain government is censoring social media and recommend a response to that censorship
Who defines “wrongfully censored” speech?
What is the best way to respond to censorship?

English

AI Snapshots

104 of 382

A humble suggestion

What would you want to know to decide if you would use this feature?

In March of 2022, Google released a new feature for Google Docs: auto-generated summaries. An AI program creates a 1 to 2 sentence summary of the document based on its content, and the user can accept or reject the suggestion.

English

AI Snapshots

105 of 382

You may have considered…

Who would use document summaries for documents you write
How accurate the summary written by the AI is
How easy it is to accept or reject the suggestion
What kinds of data does this feature collect and store from your documents?

English

AI Snapshots

106 of 382

English

Want to help us improve

AI Snapshots?

Take our 5 minute survey to share your experience:

tinyurl.com/aieducatorsurvey

You may have considered…

Farmers and farm hands can identify escaped cows
Disease outbreaks could be traced more easily
Whether the technology is clunky or easy to use
Cow thieves might have a more difficult time stealing cows

Science

AI Snapshots

116 of 382

The edge of the world

How would you design an AI that can learn to solve this problem using video clips from YouTube?

Recent years have seen a rise in the Flat Earth Movement. A small but growing number of people believe the Earth is not a sphere, but a flat plane. Many in the movement were converted to that thinking through conspiracy theory videos on YouTube or similar sites.

Design

challenge!

Science

AI Snapshots

117 of 382

You may have considered…

Who is likely to watch YouTube and similar sites
How YouTube recommends videos to users
Who is producing flat earth videos
What type of technology you would need to decipher written comments
Censorship policies

Science

AI Snapshots

118 of 382

Speaking their language

Use your Google search skills to find out: who am I?

→ I founded a startup to help self-driving cars talk to drivers, pedestrians, and cyclists.

→ My first invention was a humane mousetrap.

→ I am passionate about having a diversity of students compete in robotics competitions.

Science

🚨 🚨 Internet access required!

AI Snapshots

119 of 382

I am…

Carol Reiley!

Born in 1982
Created a DIY blood pressure monitor for developing countries
Holds eight patents

Learn more about me

at hai.stanford.edu/people/carol-reiley

Science

AI Snapshots

120 of 382

If … then…

How would you design an AI that can learn to solve this problem using images and dimensions of objects

AI robots are struggling with cause and effect. For example, while industrial robots understand how to sense nearby objects, pick them up, and move them, they don't know that hitting an object causes it to fall over or break.

Design

challenge!

Science

AI Snapshots

121 of 382

You may have considered…

How robots sense the world
If robots can remember prior consequences of their actions
If robots can transfer their knowledge from one task to another

Science

AI Snapshots

122 of 382

I swear it was an accident

You’re in charge. How would you respond to the development of this new technology?

Researchers in Switzerland developed an AI technology to create new drugs for the healthcare industry. The technology would generate new non-toxic compounds to treat human diseases. They discovered that they could modify their technology slightly to generate poisons which could be used as chemical weapons.

Science

AI Snapshots

123 of 382

You may have considered…

Whether the AI should be destroyed
How toxic or dangerous the poisons are
How easy it is to keep the AI private and secure
Who has access to the AI
How good the AI is at drug discovery

Science

AI Snapshots

124 of 382

Bruno the Bunsen burner

How comfortable would you feel working with lab equipment with a personality?

In the future, your lab equipment may be able to think for itself. Companies are exploring the idea of "anthropomorphizing" equipment (making it more human) to act as virtual colleagues.

Science

AI Snapshots

125 of 382

You may have considered…

What caused companies to try this idea
How awkward it is to interact with voice assistants like Siri or Alexa
What kinds of personalities the equipment would have (would they be moody??)

Science

AI Snapshots

126 of 382

Make it til you fake it

Is it worth creating a fake news generator in order to detect fake news better? Why or why not?

An AI called Grover was built to detect fake news. But to detect fake news, Grover first had to create it. The AI can produce realistic fake news articles from just a headline. After practicing writing fake news, Grover learned to detect fake news produced by AIs (including itself) with 92% accuracy.

Science

AI Snapshots

127 of 382

You may have considered…

Fake news is an increasing problem
Who would have access to the Grover AI
How easy it is for a human to tell if news is fake
If it’s possible to destroy Grover’s news-writing abilities

Science

AI Snapshots

128 of 382

My assistant: a webcam

How would you design an AI that can learn to solve this problem using video recordings of patient visits?

Quality medical care depends on having up-to-date records for every patient visit. However, doctors' handwriting is notoriously bad and taking notes during patient visits is slow.

Design

challenge!

Science

AI Snapshots

129 of 382

You may have considered…

Medical visits are private and it might feel uncomfortable to talk with your doctor while being recorded
Who sees the video/audio?
If the technology is proven to lead to better healthcare
Could mistakes harm patients

Science

AI Snapshots

130 of 382

Mutations not galore

What questions would you ask to figure out how useful this technology is?

Medical researchers recently developed a new AI technology. The tech is based on the idea that disease-causing mutations are rarer than variants that do not cause disease. Using this principle, they developed a technology that was 2.2% - 30% better than existing models at detecting disease-causing mutations.

Science

AI Snapshots

131 of 382

You may have considered…

The tech is as little as 2.2% better but as high as 30% better
The kinds of diseases it can detect
The data researchers used to teach the AI to recognize the disease-causing variants

Science

AI Snapshots

132 of 382

This horse has potential

What questions would you ask to figure out how this app works?

Horse racing is a $3 billion business. A niche AI-powered app allows horse breeders to upload photos of a horse to predict its performance in races. The app can also help suggest what types of photos to take to get the best results.

Science

AI Snapshots

133 of 382

You may have considered…

What features of a horse make it good at racing
What kinds of photos the app needs
How accurate and fast the app is
Whether the app gives you recommendations for each horse

Science

AI Snapshots

134 of 382

Seeing is believing

Where do you think this technology would be most useful? Why?

A new AI developed by Facebook can recognize actions in videos. So far, the AI can recognize simple actions in short clips, like someone setting down an object or doing a pushup.

Science

AI Snapshots

135 of 382

You may have considered…

Detecting harmful or violent video content
Categorizing or filtering videos
Sorting stock footage for movie making
Determining valid sports plays
Detecting criminal activity in security footage

Science

AI Snapshots

136 of 382

Stick ‘em up

How would you design an AI that can learn to solve this problem using close-ups of the pen?

Imagine this: a bank robber writes a note demanding money and hands it to a teller. Later, when authorities try to gather evidence, they find the pen the robber used was touched by a lot of people. The skin cells left on the pen are difficult to identify using traditional methods.

Design

challenge!

Science

AI Snapshots

137 of 382

You may have considered…

How machines process images
The clarity of the close-ups
The potential for misidentifying a criminal
How the technology could be misused
Where AI finds fingerprint matches

Science

AI Snapshots

138 of 382

Perfect timing

How would you decide if algae grown with this technology is a good alternative to oil and natural gas?

Algae as fuel? You heard that right. While algae can be turned into products like diesel and gasoline, it’s rarely used due to low yields and an expensive harvesting process. New research uses AI to predict the best time to harvest algae so that it grows at the fastest possible rate, without crowding out neighboring colonies.

Science

AI Snapshots

139 of 382

You may have considered…

The price of algae fuel
How fast it is to produce
Algae fuel compared with sustainable energy sources like wind and solar
Whether algae is renewable
That guy looks pretty comfy in his algae coat!

Science

AI Snapshots

140 of 382

Permission to use your saliva?

What kind of research would you be okay and not okay with 23andMe using your DNA for?

When you send your saliva sample to genetics company 23andMe, you are asked to consent to participate in research. If you give your consent, your de-identified genetic information will be used in research to make scientific discoveries.

Science

AI Snapshots

141 of 382

You may have considered…

If the research:

Is harmful, such as bioweapons research
Is medical, such as cures for diseases
Is useless, such as for chinchilla clothing
For profit or nonprofit
Benefits you or your family in any way
Might reveal your identity

Science

AI Snapshots

142 of 382

Early Detection

How would you design an AI that can learn to solve this problem using images of patients’ faces?

Some rare genetic disorders are identified by certain distinctive facial features. Ideally, doctors would identify these disorders in children so that they can get treatment early.

Design

challenge!

Science

AI Snapshots

143 of 382

You may have considered…

What does the doctor do with the results
The ages the tech works best for
How easy it is to use
How to collect images of people with rare conditions

Science

AI Snapshots

144 of 382

Learning for all

Use your Google search skills to find out: who am I?

→ I'm Israeli-American and an avid traveler.

→ I spent 20 years learning biology before working on AI.

→ I co-founded the online learning platform Coursera.

→ I founded a company that uses AI to develop new pharmaceuticals.

Science

🚨 🚨 Internet access required!

AI Snapshots

145 of 382

I am…

Daphne Koller!

Born in 1968
Worked in both machine learning and computational biology

Learn more about me at:

insitro.com/about

Science

AI Snapshots

146 of 382

Beyond the blaze

Where else do you think this kind of drone technology might be useful?

Fighting wildfires is dangerous work, especially when the scope of a fire is unknown. In the past decade, the US government has developed AI-powered drones to map out fires, determine their direction, find a fire's hottest spots, and set preventative fires to destroy wildfire fuel.

Science

AI Snapshots

147 of 382

You may have considered…

Mapping out other locations like property boundaries
Responding to other kinds of weather events
Locating missing people or wildlife

Science

AI Snapshots

148 of 382

Pedal to the metal

You're in charge. Where would you implement this racing game AI first in the real world?

The PlayStation game Gran Turismo Sport simulates car racing using rules of physics like air resistance and tire friction. A new AI, dubbed GT Sophy, learned to drive, politely sharing the road while at top speed and beating some of the world's top drivers.

Science

AI Snapshots

149 of 382

You may have considered…

Teaching student drivers
Real-life AI racing sports like NASCAR
First responders heading to the scene of an injury

Science

AI Snapshots

150 of 382

My best friend is a robot

Use your Google search skills to find out: who am I?

→ I build robots with personality, including Kismet, the first robot capable of social interactions with humans.

→ I investigate whether robots could help improve students’ mental health.

→ I was originally inspired to build robots after watching Star Wars.

Science

🚨 🚨 Internet access required!

AI Snapshots

151 of 382

I am…

Cynthia Breazeal, roboticist

Born in 1967
Professor at MIT
Consulted on the film Artificial Intelligence directed by Steven Spielberg

Science

AI Snapshots

152 of 382

3-2-1 ... say honey!

What other applications of this technology can you imagine beyond identifying beehive pests?

Engineer Sean Cusack, a Microsoft employee and beekeeper, developed a mini bee photo booth. The photo booth snaps a pic every time something enters the beehive and uses AI to determine if it was a bee or a pest.

Science

AI Snapshots

153 of 382

You may have considered…

Endangered animal identification
Detecting when an animal has had offspring
For humans, such as building and event security to identify known and unknown persons

Science

AI Snapshots

154 of 382

Arf

Use your Google search skills to find out: who am I?

I have 4 legs
Some people find me “creepy”
I can moonwalk
You can purchase me for $74,500

Science

🚨 🚨 Internet access required!

AI Snapshots

155 of 382

I am…

Spot the robot by Boston Dynamics!

Watch me moonwalk?

Science

AI Snapshots

156 of 382

A seismic shift in earthquake detection

How would you design an AI that can learn to solve this problem using sound recordings from various cities?

Honking cars in the city may be an annoyance when you're trying to sleep, but they can also make it hard to detect earthquakes. Researchers who use sound to detect earthquakes need to filter out as much background noise as possible.

Design

challenge!

Science

AI Snapshots

157 of 382

You may have considered…

If sound recordings should be shared or city-specific
What happens if the AI isn’t working and an residents aren’t warned about an incoming earthquake
Other applications of the AI like helping people get better sleep

Science

AI Snapshots

158 of 382

Rods and cones

Use your Google search skills to find out: is this AI fact or fiction?

Some parents may not realize their child is colorblind until age 4 or later. A new AI-powered smartphone app called EyeCan aims to change that. Parents can upload a closeup picture of their child's eyes and the app then uses AI to detect if they might be colorblind.

Science

🚨 🚨 Internet access required!

AI Snapshots

159 of 382

This is…

AI Fiction!

There doesn’t seem to be an app for detecting color blindness … yet! However, AI is being developed to enhance colors in real time using smart glasses or digital displays.

Science

AI Snapshots

160 of 382

The better, the better

Who could benefit from these feedback loops and who could be harmed by them?

Recommendation systems are messy. It's common for online shopping hubs and social media sites to promote more popular content; however, by promoting that content, they make it more popular. Thus, recommendation systems can create a feedback loop of increasingly popular content.

Science

AI Snapshots

161 of 382

You may have considered…

Recommendation systems are everywhere – in social media, online shopping, and content providers
New content creators have a hard time getting discovered
Just because something is popular doesn’t necessarily mean it’s good

Science

AI Snapshots

162 of 382

Yeah, but why?

Which argument do you agree with more and why?

Argument A

"An AI should be able to explain why it labeled images of animals as either predator or non-predator."

Argument B

"It doesn't matter if the AI can explain why it labeled images as predator or non-predator as long as it gets them correct."

Science

AI Snapshots

163 of 382

You may have considered…

Scientists and policymakers may use this type of AI the most
If the AI can’t explain its decision, we may trust it less
Building a new AI or expanding on it might be harder if we don’t know how it works
It may be faster to make an AI that can’t explain itself

Science

AI Snapshots

164 of 382

Huge neural network vs a tiny square

You’re in charge. How would you respond to the “one pixel attack” vulnerability?

Many image recognition AIs are vulnerable to the so-called “one pixel attack”. In the one pixel attack, changing a single pixel in an image can fool the AI into thinking a horse is a dog or can trick it into not seeing a stop sign.

Science

AI Snapshots

165 of 382

You may have considered…

How does the one pixel attack work exactly
How is the one pixel attack being used by bad actors
Can an AI understand when it’s being attacked
How can we design an AI that isn’t vulnerable to this attack

Science

AI Snapshots

166 of 382

Hang on, my battery is at 1%

In what situations would you want more accurate image recognition even if it cost more power?

320 trillion operations per second. That's what it takes to power Nvidia's new driverless car. Any tech that relies on AI-powered vision (like phones, drones, driverless cars) experiences a trade off: use more power but have better accuracy or last longer with lower accuracy.

Science

AI Snapshots

167 of 382

You may have considered…

Greater accuracy might be useful for:

Medical or legal situations
Situations with unlimited power

Lower accuracy might be useful for:

Entertainment purposes
Optional features, like driving assist
Situations with less available power, like a battery
Environmental sensitivity

Science

AI Snapshots

168 of 382

And this piggy… sounded happy!

Use your Google search skills to find out: is this AI fact or fiction?

A new AI system can interpret pigs’ moods based on their grunts and squeals.

Science

🚨 🚨 Internet access required!

AI Snapshots

169 of 382

This is…

AI Fact!

Pigs are a highly social species, and their grunts can depend on their personality. A new AI tries to detect their mood from their sounds alone.

Science

AI Snapshots

170 of 382

Forest for the trees

How would you design an AI that can learn to solve this problem using aerial images of forests?

Each year, foresters try to balance the number of trees that are cut down for lumber with the needs of the environment. Traditionally, they survey acres of forest at ground level to determine how many trees are standing and what types of trees they are. It can take years, or even decades, to cover just one forest.

Design

challenge!

Science

AI Snapshots

171 of 382

You may have considered…

How to teach AI to identify different types of trees
Is there any way this technology can be could end up harming the environment?
Do people from one country have a right to take aerials images of forests in other countries?
Could the technology be used for other things like detecting forest fires or endangered animals?

Science

AI Snapshots

172 of 382

Another one bites the dust

You're in charge. Where would you implement this new technology first to get the greatest benefit from it?

A desert dust storm can seemingly come from nowhere, causing a sudden drop to zero visibility. By nature, they are hard to track, but an AI technology developed in 2020 can automatically detect dust clouds using satellite data, speeding up the process of identifying them.

Science

AI Snapshots

173 of 382

You may have considered…

Regions where dust storms are common
Places where it’s easy to implement this technology
Places where dust storms affect crops or other crucial resources

Science

AI Snapshots

174 of 382

A little of this, a little of that

What questions would you ask to figure out whether this technology is safe and reliable?

It's common for people with depression to try many different medications before finding the best one. A new AI uses brain scans of people with depression to predict how well they will respond to different medications.

Science

AI Snapshots

175 of 382

You may have considered…

How accurate the technology is in its predictions
Whether people trust AI to prescribe their medications
Consequences of switching medications
What data is used to make the predictions

Science

AI Snapshots

176 of 382

Grab your umbrella

Use your Google search skills to find out: is this AI fact or fiction?

A team at Google is using AI to predict two-hour rainfall levels better than traditional forecasting models.

Science

🚨 🚨 Internet access required!

AI Snapshots

177 of 382

This is…

AI Fact!

Rainfall is much less predictable than you might think. The new AI saves time for meteorologists and may become more useful as climate change progresses.

Science

AI Snapshots

178 of 382

Forklift panic

Who could benefit from this technology and who could be harmed by it?

An AI-powered forklift encounters a new situation: a fallen barrel is slowly rolling across its path. It freezes, unsure what to do next. A startup called ArcBest employs flesh-and-blood forklift drivers to take the wheel remotely when an AI-powered forklift doesn't know what to do.

Science

AI Snapshots

179 of 382

You may have considered…

Jobs created and lost by the technology
How safe the self-driving forklifts are
How often the forklifts need human assistance

Science

AI Snapshots

180 of 382

Packing peanuts, box, prime day

How would you design an AI that can learn to solve this problem using images of items?

Sometimes Amazon items arrive damaged. Or they ship in a package that's way too large. Or each item comes separately when they could have come together. Amazon's current systems need help deciding how an item should be packaged and shipped.

Design

challenge!

Science

AI Snapshots

181 of 382

You may have considered…

The number of images you would need to cover every item in stock
Other information that might be useful like each product’s dimensions
Impacts on the environment if more or less cardboard is used
How big of an issue packaging mistakes are

Science

AI Snapshots

182 of 382

Making machines human

Use your Google search skills to find out: who am I?

→ I founded a startup to help AI understand human emotion.

→ I am Egyptian-American.

→ I helped found the Autism & Communication Technology Initiative at MIT.

→ I would tell you "my life's work is about humanizing technology before it dehumanizes us".

Science

🚨 🚨 Internet access required!

AI Snapshots

183 of 382

I am…

Rana el Kaliouby!

Born in 1978
Prominent investor for startups
Author of “Girl Decoded”, a book about bringing emotion to technology

Science

AI Snapshots

184 of 382

To infinity and beyond

How would you design an AI that can learn to solve this problem using images from telescopes?

Astronomers discover new planets by staring at the stars. When a planet passes in front of a star, it dims the star's light slightly. Even with advanced telescopes and a careful eye, planets are difficult to find because the change in brightness is barely noticeable.

Design

challenge!

Science

AI Snapshots

185 of 382

You may have considered…

What astronomers do with information they get from the technology
Who gets credit for finding a new planet if an AI program made the discovery
Looking for other types of space objects, like natural resources

Science

AI Snapshots

186 of 382

Off by a few years

Use your Google search skills to find out: is this AI fact or fiction?

In 1970, a prominent AI researcher was quoted as saying “[in] 3 to 8 years we will have a machine with the general intelligence of an average human being.”

Science

🚨 🚨 Internet access required!

AI Snapshots

187 of 382

This is…

AI Fact!

Yes, back in 1970, researcher Marvin Minsky thought we would have the terminator by 1978. You could say he was … slightly off, seeing as how we’re not even close to creating robots as smart as humans.

Science

AI Snapshots

188 of 382

Not quite as advanced as we would like

What questions would you ask to figure out why these AI-based tools had little impact?

Hundreds of AI tools were developed to help fight COVID-19, but scientists agree that they made little impact. Tools that diagnose patients or predict how ill they might get fell flat.

Science

AI Snapshots

189 of 382

You may have considered…

The speed at which the tools were developed
Where the tools were applied, such as across countries
Who funded the tools
Where the tools worked best and worst

Science

AI Snapshots

190 of 382

AI and your brain

How would you design an AI that can learn to solve this problem using brain scans from MRIs?

Millions of Americans have Alzheimer’s disease, a brain disorder marked by memory loss and other mental impairment. Alzheimer’s can often be diagnosed through MRI scans of a patient’s brain. Unfortunately, reading MRI images takes a special skill set and is time consuming when done by individual doctors.

Design

challenge!

Science

AI Snapshots

191 of 382

You may have considered…

Where to get the MRI scans to teach the AI program
The dangers of inaccurate diagnosis
What part doctors will play in the diagnosis
Whether patients will trust AI when it comes to their health

Science

AI Snapshots

192 of 382

Bzz bzz

Use your Google search skills to find out: is this AI fact or fiction?

Companies are testing robots that can pollinate crops like blueberries and almonds, replacing bees.

Science

🚨 🚨 Internet access required!

AI Snapshots

193 of 382

This is…

AI Fact!

As the population of bees declines, companies are scrambling for pollinators for their crops. Robot bees aren’t as far off as they might seem!

Science

AI Snapshots

194 of 382

*sniff* *sniff*

You're in charge. Where would you implement this new technology first to get the greatest benefit from it?

Smell is complex, but you may be surprised to learn that AI can learn to recognize scents. A new AI mimics the brain structures mammals use to smell and can detect 10 different scents, such as ammonia. It was relatively accurate even when multiple scents were mixed together.

Science

AI Snapshots

195 of 382

You may have considered…

Safety features like detecting gasoline spills
Helping someone figure out if they’re smelly
Creating the best-smelling floral bouquet
Businesses using the AI to create a brand smell

Science

AI Snapshots

196 of 382

Ouch.

Should robots be able to react to a sensation that would cause pain? Why or why not?

The difference between a gentle touch and a hard thump: pain. A new AI-powered robot developed by a researcher in Japan can react to sensations that would cause pain in humans. The robot converts these sensations into a facial expression.

Science

AI Snapshots

197 of 382

You may have considered…

Human empathy for machines
What kinds of robots would use this
How to tell if a robot can “feel” pain or just sense it

Science

AI Snapshots

198 of 382

Science

Want to help us improve

AI Snapshots?

Take our 5 minute survey to share your experience:

tinyurl.com/aieducatorsurvey

AI Snapshots

199 of 382

Math warmups

AI Snapshots for the math classroom

200 of 382

Math is just playing by the rules. Or is it?

Do you think that AI might one day replace mathematicians?

“[Math is] often seen from the outside as a purely mechanical game played with fixed rules, like checkers or chess … These are activities in which machines have already demonstrated superhuman ability. But for [some], math is different: it is a creative pursuit that calls on our intuition as much as our ability to compute.” – J. Ellenberg

Math

AI Snapshots

201 of 382

You may have considered…

What kinds of tasks AI is best at: rule-based or creative
How much creativity math involves
What kind of creativity math involves

Math

AI Snapshots

202 of 382

Do I know you?

Who could benefit from deleting these templates and who could be harmed?

Recognizing a face may seem like a human action, but AI views it as a math problem. In late 2021, Facebook promised to delete its 1 billion "facial recognition templates" - mathematical values that tell a computer how to recognize someone.

Math

AI Snapshots

203 of 382

You may have considered…

Facebook commonly tagged users in photos even when they didn’t want to be
Tagging users was easier using the automated system
Why would Facebook delete the templates?

Math

AI Snapshots

204 of 382

First served

How would you design an AI that can learn to solve this problem using lists of cities and statistics about them, such as number of hospitals?

COVID-19 illustrated the complications with large-scale vaccine distribution. To decide which communities to prioritize, officials consider hundreds of statistics like population density, residents’ age, and number of hospitals. The process is time-consuming and statistics are often inaccurate.

Design

challenge!

Math

AI Snapshots

205 of 382

You may have considered…

How to determine which factors make someone high-risk
Potential bias in the statistics used to make decisions
Backlash against public health officials
Deadly impacts of getting things wrong and liability around them

Math

AI Snapshots

206 of 382

Tell me about it

Which argument do you agree with more? Why?

Argument A

"AI should be able to explain why a mathematical proof is correct in plain English so that regular people can understand it."

Argument B

"AI should be able to explain why a mathematical proof is correct for mathematicians, but not necessarily the general public."

Math

AI Snapshots

207 of 382

You may have considered…

Who mathematical proofs are typically written for
When or how an average person might use a proof
Whether communicating to the general public is important for other fields besides math
If it’s more effort for AI to make a proof understandable

Math

AI Snapshots

208 of 382

Into the wrong hands

Would you feel comfortable with your data being used by this AI and why or why not?

In a 2019 study, an AI model created to predict medical events (such as hospitalization) was based on medical claims data from a group of 112,000 children. The data included information about age, gender, hospitalization, ambulance use, and total amount spent on healthcare.

Math

AI Snapshots

209 of 382

You may have considered…

Medical data is personal and often identifiable
How is the data being used? Is it to help patients or to help insurance companies?
Whether you can request that the data be deleted

Math

AI Snapshots

210 of 382

AI to A1

Use your Google search skills to find out: is this AI fact or fiction?

An AI first beat a world champion at a chess match in 2011.

Math

🚨 🚨 Internet access required!

AI Snapshots

211 of 382

This is…

AI Fiction!

Beating a world champion was a longtime goal for AI, but it was first achieved in 1997, when IBM’s Deep Blue beat Garry Kasparov.

Math

AI Snapshots

212 of 382

Vroom vroom

How would you design an AI that can learn to solve this problem using ride information collected from customers and drivers?

To remain competitive, ride-sharing apps constantly balance the number of customers looking for rides with the number of drivers they make available. They are always collecting data but it is still difficult to figure out exactly how many drivers they should have on the road at any given time.

Design

challenge!

Math

AI Snapshots

213 of 382

You may have considered…

Negative consequences of collecting so much location data
Impacts on drivers who get less rides
Wait times for customers if the AI is wrong
Effects on the environment if there are less drivers driving around and idling

Math

AI Snapshots

214 of 382

A math creative

Use your Google search skills to find out: who am I?

→ I’m an Aussie.

→ My specialty is the branch of mathematics called representation theory.

→ I used DeepMind’s AI to help create mathematical proofs in advanced algebra.

→ I love the intuitive and creative aspects of math.

Math

🚨 🚨 Internet access required!

AI Snapshots

215 of 382

I am…

Geordie Williamson

Born in 1981
Mathematician and professor
Worked in Germany, the UK, and Australia

Math

AI Snapshots

216 of 382

A career in bounding boxes

Who could benefit from this work and who could be harmed by it?

While in college, students in Venezuela can pick up side jobs labeling data for AI. They might complete tasks like creating bounding boxes around traffic lights and signs. While most get paid pennies for their work, some do manage to make a good living.

Math

AI Snapshots

217 of 382

You may have considered…

Venezuelans who mark bounding boxes may make more than their peers
Doing this work might not build long-term skills
This career relies on stable internet and the ability to transfer currency
American and European companies benefit from cheap labor in other countries

Math

AI Snapshots

218 of 382

I <3 data and cheeseburgers

Use your Google search skills to find out: who am I?

→ I love data and cheeseburgers.

→ I’m famous for explaining machine learning in 5 levels of difficulty on YouTube.

→ I’m the founder of an AI-powered storytelling game platform and an AI research company.

Math

🚨 🚨 Internet access required!

AI Snapshots

219 of 382

I am…

Hilary Mason!

Data scientist from New York
Founder of HackNY, a startup that helps students break into the New York startup community
TED speaker who once gave a talk about how to replace yourself with a computer program

Math

AI Snapshots

220 of 382

If the price is right

You're in charge. Should companies be prevented from replacing human workers with AI? Why or why not?

Human workers are expensive, but AIs don't need to eat, sleep, take a break, or raise a family. So, many companies are working on technologies that replace workers with AI, instead of technologies that increase a person's abilities.

Math

AI Snapshots

221 of 382

You may have considered…

The efficiency of humans and machines
Society is structured around work
What kinds of jobs might be created

Math

AI Snapshots

222 of 382

Infrequent flyers

How would you design an AI that can learn to solve this problem using ticket logs from previous flights?

Overbooking is when airlines purposely sell too many tickets for a flight. They must account for the number of seats on a plane, number of tickets sold, and costs of paying customers who don’t have seats. When predictions are wrong, they lose customers who are unhappy with being delayed.

Design

challenge!

Math

AI Snapshots

223 of 382

You may have considered…

Losing customers when the AI is wrong
Who controls the technology and how it will be shared across airlines
Where to source the information you use to teach the AI
Making human airline employees more productive or less necessary

Math

AI Snapshots

224 of 382

Math + autocorrect = ?

Use your Google search skills to find out: is this AI fact or fiction?

Constantly messing up order of operations or having trouble keeping track of long equations? Math autocorrect to the rescue. A California-based startup is developing AutoMath, an AI-based program that can fix mathematical equations and expressions with minor errors.

Math

🚨 🚨 Internet access required!

AI Snapshots

225 of 382

This is…

AI Fiction!

No California-based startup has developed a program to fix incorrect equations.

However, there are AI programs out there that try to solve equations (although we won’t name them here 👀).

Math

AI Snapshots

226 of 382

Hidden assets

Why do you think companies refuse to share their AI algorithms with the public?

You might have an idea about how TikTok ranks videos or how Amazon recommends you freeze-dried ice cream. But companies keep their AI algorithms locked up tight, claiming their algorithms are trade secrets.

Math

AI Snapshots

227 of 382

You may have considered…

Companies keep trade secrets for a competitive edge
Scientists and governments might want to examine algorithms
Whether the public should know how algorithms influence them

Math

AI Snapshots

228 of 382

It ain’t cheap

How would you figure out whether it's worth the cost to train an AI for a certain task?

Getting AI to learn how to do its job is pricey - and it's only getting pricier. Smaller models, such as ones used by healthcare providers, might cost $50,000. Larger models, like the one used by Google to predict search inquiries, can cost up to $10,000,000.

Math

AI Snapshots

229 of 382

You may have considered…

Compare the cost with expected profits
Decide how much good the algorithm does in the world
Figure out if cheaper solutions work just as well

Math

AI Snapshots

230 of 382

Suspiciously addicted

Use your Google search skills to find out: is this AI fact or fiction?

Casinos and gaming websites are using AI to spot signs of gambling addiction based on factors like betting amounts.

Math

🚨 🚨 Internet access required!

AI Snapshots

231 of 382

This is…

AI Fact!

It may come as a surprise, but casinos don’t want you to go broke. It turns out casinos make more profit at a certain “sweet spot” of addiction, so they might benefit from detecting signs of gambling addiction early. One challenge is that many signs of addiction are subtle.

Math

AI Snapshots

232 of 382

So this is a graph… it’s got bars on it…

How would you design an AI that can learn to solve this problem using examples of people explaining graphs?

For students with visual impairments, online tests include automated screen readers that read text on the screen out loud. In math tests, it is difficult for a screen reader to communicate graphs, tables, and charts.

Design

challenge!

Math

AI Snapshots

233 of 382

You may have considered…

How humans would describe charts and graphs
How inaccurate technology could skew test grades
How students would communicate their answers
What else the technology could be used for
Impacts on the jobs of teachers and test proctors

Math

AI Snapshots

234 of 382

Swipe. Swipe. Swipe. Watch for 10 seconds. Comment.

Why do you think TikTok tries to increase retention and not just watch time?

TikTok recently published a description of the AI-based algorithm it uses to recommend videos. The algorithm tries to optimize for retention (whether a user comes back to the app frequently) as well as watch time.

Math

AI Snapshots

235 of 382

You may have considered…

TikTok’s ad revenue
Tech companies increasingly try to create addictive products
Factors that might push users to quit TikTok

Math

AI Snapshots

236 of 382

Measure for measure

What questions would you ask to figure out if this app would be useful to you?

If you, like most people, don't carry a ruler in your pocket, the Measure app is built for you. The AI-assisted app uses the camera to measure dimensions of objects, but critics say the app mis-measures in an unpredictable way, sometimes accurate and sometimes off by a foot or more.

Math

AI Snapshots

237 of 382

You may have considered…

How accurate the measurements are in different scenarios
If you need the app for a serious application, like something safety-related
Whether the app has different options like metric system or 3D visualization

Math

AI Snapshots

238 of 382

An AI that's integral to integrals

What are the potential benefits and risks of rolling out this technology?

Facebook is developing a cutting edge AI to solve calculus problems. Trained on tens of millions of randomly generated problems, the AI can solve college-level calculus problems.

Math

AI Snapshots

239 of 382

You may have considered…

Students using the AI to skip math homework
Professionals like architects saving time when they need to use calculus
Expansions to the technology, like one that can explain how to solve different problems
Could calculus be removed from schools now?

Math

AI Snapshots

240 of 382

Zoom zoom

How would you figure out if an AI-based speed assistant was worth developing?

Sure, self-driving cars are in the spotlight right now. But what about simple AI-based solutions that cut down on distracted driving and speeding? Scientists argue that AI that helps drivers manage a car's speed might prevent the most accidents.

Math

AI Snapshots

241 of 382

You may have considered…

How effective the AI might be at controlling speeds
How expensive it would be to develop
How long it would take to develop
Whether investing in self-driving cars would be better in the long run

Math

AI Snapshots

242 of 382

Starting early

How would you design an AI that can learn to solve this problem using survey data about girls’ interests?

Image you’re starting Math+AI4Girls, a nonprofit that aims to engage middle school girls in the mathematics of AI. You’re having trouble recruiting students. You’ve found that students, parents, and teachers are unaware of the vast impacts AI will have on their lives, so they don’t seek out your programs.

Design

challenge!

Math

AI Snapshots

243 of 382

You may have considered…

Why there is a lack of interest
What might engage prospective students
How to collect the data you need from minors
How to include perspectives of teachers and parents or guardians
How to measure success

Math

AI Snapshots

244 of 382

Aces & AIs

Who could benefit from this technology and who could be harmed by it?

Betting, checking, and folding in poker are delicate decisions when money is on the line. A new AI-based tool called PioSOLVER can help poker players train their decision-making abilities by recommending the best moves in unique situations.

Math

AI Snapshots

245 of 382

You may have considered…

Poker players can practice their skills
Players who don’t use the AI may be at a disadvantage
Using the AI requires a lot of trust
Casinos might feel threatened by the technology or ban it

Math

AI Snapshots

246 of 382

😡

Who could benefit from weighing angry reacts more and who could be harmed by it?

All the way back in 2016, Facebook released five new emoji reacts, including "angry." What users didn't realize is that, until recently, "angry" reacts were weighted 5 times more than a simple like in Facebook's recommendation algorithm.

Math

AI Snapshots

247 of 382

You may have considered…

Users who posted fake news or highly politicized content had posts ranked higher
People who used Facebook experienced more negative emotions than before the change
Facebook saw increased ad revenue because of higher site traffic

Math

AI Snapshots

248 of 382

A pinch of AI

How would you design an AI that can learn to solve this problem using text of readers questions and comments?

Recipe bloggers note that their readers often need to adjust recipes. They might want to make larger portions, adjust baking temperatures from fahrenheit to celsius, or change cook time based on their altitude. Bloggers spend a lot of time responding to questions about these calculations.

Design

challenge!

Math

AI Snapshots

249 of 382

You may have considered…

Common questions and comments
How to source blogs and teach the AI
Readers blaming bloggers when the AI is wrong
How to make the AI write in a human tone
How personal human interactions influence bloggers’ and readers’ experiences with the blogs

Math

AI Snapshots

250 of 382

AI can help you learn math

Use your Google search skills to find out: who am I?

I'm interested in math teaching and learning.
I proposed math teaching guidelines that de-emphasize calculus and reject the idea that only some children are naturally gifted.
I'm the co-founder of youcubed.org, an AI-based math learning website.

Math

🚨 🚨 Internet access required!

AI Snapshots

251 of 382

I am…

Jo Boaler!

Born in 1964
Professor at Stanford University
Has written 18 books

Math

AI Snapshots

252 of 382

Action!

You're in charge. What would you set as the maximum video length on TikTok and why?

The first TikToks were capped at 15 seconds. Then TikTok upped the limit to 60 seconds, then 3 minutes, now 10 minutes. Longer videos mean exponentially more data for the AI-based algorithm to analyze.

Math

AI Snapshots

253 of 382

You may have considered…

It takes much longer for the AI to analyze longer videos in order tag and recommend them
People are less likely to watch later parts of a video
TikTok has been increasing the cap, so there must be some kind of advantage

Math

AI Snapshots

254 of 382

An enigma

Use your Google search skills to find out: who am I?

I worked at Bletchley Park in the UK as a code-breaker during World War II.
I lived with foster parents while my parents worked in India.
In 1936, I designed a machine that can be constructed to follow any computer algorithm.

Math

🚨 🚨 Internet access required!

AI Snapshots

255 of 382

I am…

Alan Turing!

Born in 1912, died 1954
While reporting a burglary, Turing admitted to having a relationship with a man and was convicted of “gross indecency” because of it

Math

AI Snapshots

256 of 382

Back from the dead

Who could benefit from this technology and who could be harmed by it?

Beethoven may have died in 1827 before finishing his 10th symphony. But that didn't stop a team of researchers from using AI to approximate what the finished piece would have looked like.

Math

AI Snapshots

257 of 382

You may have considered…

Who owns pieces written by the AI
How closely the AI can mimic a composer’s style
If the AI would make composers obsolete

Math

AI Snapshots

258 of 382

Losing your mind one 1099 at a time

How would you design an AI that can learn to solve this problem using data about tax errors?

Every spring, Americans make millions of mathematical errors on their taxes. These mistakes can cost citizens thousands of dollars if they are not fixed.

Design

challenge!

Math

AI Snapshots

259 of 382

You may have considered…

Why Americans are challenged by this kind of math
How tax forms are currently laid out
Who takes the blame (and pays the bill) if the technology makes mistakes
Whether the tech is made and distributed by the government or private companies
Impacts on human accountants

Math

AI Snapshots

260 of 382

I didn't recognize you because of ... your eyeliner?

Who could benefit from this makeup application and who could be harmed by it?

"Facial recognition templates" are the math behind your face. It's how AI can recognize you. But a team of researchers used makeup to fool face recognition 99% of the time, despite looking completely normal to the human eye.

Math

AI Snapshots

261 of 382

You may have considered…

Average citizens can use the makeup to avoid being tracked by cameras in public
Law enforcement might find that people involved in crime are able to avoid facial recognition
Not everyone likes wearing makeup
How easy the makeup is to apply

Math

AI Snapshots

262 of 382

So that's where that is!

Why do you think it's important for robots to be able to "see" their own body parts?

Robots have long struggled with the "frame problem" - figuring out which parts of a problem are important for solving it. A new AI innovation helps robots track specific points in space, allowing them to locate their own body parts relative to the environment.

Math

AI Snapshots

263 of 382

You may have considered…

Why can’t a robot just sense where its body is?
How can a robot tell its body parts from other items in its environment?
The robot needs to know its own location relative to the location of other objects to plan its movements

Math

AI Snapshots

264 of 382

Professor Scribbles

How would you design an AI that can learn to solve this problem using images of the whiteboards?

College-level math professors are known for scrawling long and complicated math problems and proofs on their whiteboards during class. Students in the back of large lecture halls may find it difficult to read the writing and keep up.

Design

challenge!

Math

AI Snapshots

265 of 382

You may have considered…

Accounting for a wide variety of handwriting styles
What happens to students’ understanding of math when the AI doesn’t work
Which colleges and universities can afford this technology
How many examples the AI would need before it is successful
Leveraging technologies that already exist to interpret handwriting

Math

AI Snapshots

266 of 382

From snap to solution

Use your Google search skills to find out: is this AI fact or fiction?

Tired of solving equations, graphing, and multiplying matrices? Microsoft Math Solver allows you to take a picture of a math problem and solves it for you. It can even suggest multiple ways to get the correct result.

Math

🚨 🚨 Internet access required!

AI Snapshots

267 of 382

This is…

AI Fact!

While we don’t condone using Math Solver for your homework, it can save you time in your everyday life.

Math

AI Snapshots

268 of 382

A quintillion options

Who could benefit from this technology and who could be harmed by it?

Imagine picking an outfit when you have the entire Amazon warehouse to choose from: tens of thousands of shirts, pants, socks, and more, creating trillions of combinations. A new AI aims to save you some time by recommending items that fit an existing outfit.

Math

AI Snapshots

269 of 382

You may have considered…

Personal stylists might see a drop in business
Average people might be able to save time finding a new look
Amazon might need to collect a lot of data about you to make this AI work
Amazon’s profits might increase

Math

AI Snapshots

270 of 382

First-time home buyer: a bot

You're in charge. How would you respond to these AI-assisted real estate purchases?

AI-assisted real estate purchases more than doubled from 2021 to 2022. While they can streamline a complex process, they often increase the number of homes rented rather than sold, which makes it more difficult to find an affordable home for families.

Math

AI Snapshots

271 of 382

You may have considered…

Banning or restricting AI home buyers
How to determine if a home is bought using an AI
Higher home prices are good for sellers

Math

AI Snapshots

272 of 382

2, 3, 5, 7, 11, 13, 17, 19 ...

What are the potential benefits of this AI?

Let's look at the Fibonacci sequence: 1, 1, 2, 3, 5, 8, 13, 21 ... Researchers have developed a new AI that can figure out how to generate this sequence of numbers. In the case of Fibonacci:

u_n = u_n-1 + u_n-2

Math

AI Snapshots

273 of 382

You may have considered…

Find patterns in real-life situations
Help students and mathematicians understand number sequences
Double check mathematicians’ work

Math

AI Snapshots

274 of 382

Oh whoa, oh whoa, oh whoa

What are the benefits and drawbacks of hiding dislikes even though they still influence suggestions?

"Baby" by Justin Bieber is one of the most disliked YouTube videos of all time, with 13 million dislikes. But users can no longer see its like-to-dislike ratio. While YouTube still uses dislikes to rank video suggestions, it keeps them hidden from viewers.

Math

AI Snapshots

275 of 382

You may have considered…

Content creators are protected from bullying through mass disliking
It is harder to tell if a video is high quality right away
Bad videos still get downranked
For viewers and content creators, it’s not as clear how videos are ranked

Math

AI Snapshots

276 of 382

Alert: incorrect subtraction

Use your Google search skills to find out: is this AI fact or fiction?

A new company called TutorAI recently launched a program to help younger students practice math. As the student writes on their tablet, the AI can flag mistakes and give hints for struggling students.

Math

🚨 🚨 Internet access required!

AI Snapshots

277 of 382

This is…

AI Fiction!

While there are handwriting recognition apps, automated writing correction tools, and assistants that help rephrase your sentences, there isn’t one we know of to help you fix your math as you write it out.

Math

AI Snapshots

278 of 382

Failed my test. Over.

What questions would you ask to figure out how this AI works?

How would you score on a flight test? The US Air Force is investing in AI systems that help pilots train. The AI grades trainees on their landings and monitors their scores over time.

Math

AI Snapshots

279 of 382

You may have considered…

What data the AI uses, like speed, angle of approach, or pilot biometrics
How the pilot’s score is determined
If the AI can recommend strategies for getting a higher score next time

Math

AI Snapshots

280 of 382

The REAL Salvator Mundi

Where else do you think an AI that recognizes authentic images could be useful?

Can math help us figure out if a painting is authentic? Researchers divided up Da Vinci paintings into 350x350 pixel squares and trained an AI to recognize his handiwork.

Math

AI Snapshots

281 of 382

You may have considered…

Attributing other works of art like digital art or sculptures
Figuring out if an image has been digitally altered
Authorship is complicated in the modern world, with frequent collaboration and ease of copying information

Math

AI Snapshots

282 of 382

This post contains sensitive content

How would you design an AI that can learn to solve this problem using comments from social media posts?

In 2019, a meme asking users to figure out a short but confusing equation went viral. Mathematicians spoke out against the meme because they were worried it implied that math is made to trick people. However, memes that show the interesting and intuitive sides of math don’t often go viral.

Design

challenge!

Math

AI Snapshots

283 of 382

You may have considered…

Why the original meme went viral
How social media algorithms work
What types of math problems you’d find fun and engaging
Where to find cool problems

Math

AI Snapshots

284 of 382

Being alive wasn’t in the job description

What questions would you ask about these studies to decide if automation was overall beneficial or harmful?

If you're worried about losing your job to a robot, what we're about to tell you may give you a sigh of relief. A recent Japanese study showed that a 1% rise in the use of robots in a country correlated to a 0.28% rise in employment. However, another study suggested higher automation was linked to lower wages.

Math

AI Snapshots

285 of 382

You may have considered…

Increase in jobs is probably positive for workers
Jobs might increase, but workers may also have to switch jobs or learn a new skill
More lower wage jobs is probably not a positive effect of automation
Will the trend continue?

Math

AI Snapshots

286 of 382

Beyond true and false

Use your Google search skills to find out: who am I?

I was born in Azerbaijan, but I moved to the US in my 20s.
I would describe myself as stubborn.
I came up with the idea of "fuzzy sets": groups like "young" or "tall" that are hard to define with one value.

Math

🚨 🚨 Internet access required!

AI Snapshots

287 of 382

I am…

Lofti Zadeh

Born in 1921, died in 2017 at the age of 96
Enjoyed photography
Discovered lots of “fuzzy” concepts, including fuzzy sets, fuzzy algorithms, fuzzy languages, fuzzy probabilities, and more

Math

AI Snapshots

288 of 382

A solid foundation

How would you design an AI that can learn to solve this problem using copies of construction proposals?

Proposals for new building construction projects have to account for lots of factors like costs, profits, geometry measurements, and required amounts of materials. Every year, incorrect calculations cost companies millions of dollars in lost time and revenue.

Design

challenge!

Math

AI Snapshots

289 of 382

You may have considered…

How this affects architects and other humans involved in construction planning
Safety issues if the math is wrong
Millions of dollars in costs when things need to be redone
Accounting for unique situations that aren’t reflected in the original reports

Math

AI Snapshots

290 of 382

Math

Want to help us improve

AI Snapshots?

Take our 5 minute survey to share your experience:

tinyurl.com/aieducatorsurvey

AI Snapshots

291 of 382

Social studies warmups

AI Snapshots for the social studies classroom

292 of 382

Spotify polisci

What questions would you ask to decide if Spotify for political candidates is right for you?

Spotify... but for political candidates? Imagine that's the pitch of a new company that aims to compare your political views with similar people's voting histories to recommend you a candidate.

Social Studies

AI Snapshots

293 of 382

You may have considered…

Where the AI gets your “political views” from
Who you’re being compared to (who counts as similar?)
How accurate the recommendation is
How easy it is to accept or reject the recommendation

Social Studies

AI Snapshots

294 of 382

Weight watchers: kids

Who could benefit from destroying this AI model and who could be harmed?

The federal government recently ruled that AI models created by Weight Watchers needed to be destroyed. Weight Watchers used the AI to power a food tracking app for kids. The government’s decision stemmed from the app’s use of data collected from minors (children younger than 18).

Social Studies

AI Snapshots

295 of 382

You may have considered…

children’s data should be collected with consent
a food tracking app for children could lead to intense scrutiny of what the child eats, which might lead to unhealthy behavior
children may lie about their age in order to use the app

Social Studies

AI Snapshots

296 of 382

The buck stops with…

Which argument do you agree with more and why?

Argument A

"If an app uses AI in an unethical way, companies like Apple and Google should be responsible for taking it down from their app stores."

Argument B

"If an app uses AI in an unethical way, the government should be responsible for requiring the author of the app to delete it."

Social Studies

AI Snapshots

297 of 382

You may have considered…

companies like Apple and Google may have a better understanding of what makes an app unethical
governments can fine or prosecute companies
government action often moves slowly
companies are increasingly under pressure to moderate content better

Social Studies

AI Snapshots

298 of 382

New post who dis

What questions would you ask to figure out how this AI text analysis worked?

Starting with 4chan and 8chan posts, researchers conducted AI-based text analysis on an anonymous cult leader to figure out their identity. Researchers concluded the cult leader was in fact three people using one account.

Social Studies

AI Snapshots

299 of 382

You may have considered…

how the researchers collected the data
what evidence led them to believe the account was used by three people
how long the analysis took
who paid for the analysis and why

Social Studies

AI Snapshots

300 of 382

To see or not to see

What questions would you ask to figure out why this Feather Flag was misidentified?

“Earlier that day, I fed an image of the Native American feather flag into a major image recognition app. This deep learning-based visual analysis service couldn’t even begin to identify it properly — because it doesn’t have enough training data. This sacred staff with eagle feathers was instead identified by the AI tool as 99.9% a handrail or a banister.” - Davar Ardalan (2021)

-Ardalan, D. (2021, October 4). The new AI frontier: Tagging indigenous values. Medium.

Social Studies

AI Snapshots

301 of 382

You may have considered…

what images the AI learned from
how many images of feather flags the AI saw as it learned
what other images the AI might not recognize
a human might not know what the Feather Flag was, but probably wouldn’t call it a handrail

developing more pathways into AI engineering
reaching out to younger students to get them interested in the field
creating specialized college programs, like an AI major

Social Studies

AI Snapshots

310 of 382

Marble bust

How would you design an AI that can learn to solve this problem using images of artifacts?

In 1983, a man tried to sell an ancient Greek statue to a museum for $10 million. There was only one problem: it was fake! After 14 months, special experts determined the sculpture wasn’t real. Forged historical artifacts are constantly in circulation and can cost museums very high sums of money.

Design

challenge!

Social Studies

AI Snapshots

311 of 382

You may have considered…

Specific characteristics of real vs. fake sculptures
Who will use the technology
Who is motivated to sell fake items
How our understanding of history might be affected by this technology

Social Studies

AI Snapshots

312 of 382

Just common sense

You’re in charge. What is one piece of “common sense” you think all AIs should have?

The Machine Common Sense project began in 2019 with aims like:

A house-cleaning robot should know not to throw out a cat or put it in a drawer
An AI should know that the headline “Cheeseburger Stabbing” isn’t about a “food-on-food crime”

Social Studies

AI Snapshots

313 of 382

You may have considered…

understanding what harms humans
not stepping out in traffic
not to use disrespectful language

Social Studies

AI Snapshots

314 of 382

Thank you for coming to my TED Talk

Use your Google search skills to find out: who am I?

→ I grew up in Ethiopia.

→ I built an AI company to predict global farming trends and fight food insecurity.

→ I stockpiled toilet paper before it was cool.

Social Studies

🚨 🚨 Internet access required!

AI Snapshots

315 of 382

I am…

Sara Menker!

Founder of Gro Intelligence, which has raised over $125 million in funding
Famous for a 2017 TED Talk predicting a global food crisis

Social Studies

AI Snapshots

316 of 382

Anti-terrorist AI

How would you design an AI that can learn to solve this problem using social media data?

Tweets, posts, stories, reels, comments. It's a lot to go through when searching for potential terrorist activity. The FBI has to analyze a large amount of information from social media and messaging platforms to monitor suspected terrorists.

Design

challenge!

Social Studies

AI Snapshots

317 of 382

You may have considered…

Social media users’ privacy concerns
Potential bias in the system
The risks of wrongly accusing users
What types of data are labeled high-risk

Social Studies

AI Snapshots

318 of 382

Legislating AI

What is one right you would like to see in the AI Bill of Rights?

The Bill of Rights enshrines some of Americans' most treasured freedoms, such as freedom of speech. The White House is currently working on a new AI Bill of Rights that outlines the rights and freedoms we should be guaranteed when using AI.

Social Studies

AI Snapshots

319 of 382

You may have considered…

the right to say not to data collection
the right to have a company publish an explanation of how their AI works (e.g. TikTok ranking algorithm)
the right to know what apps do and do not use AI

Social Studies

AI Snapshots

320 of 382

Who watches the watchmen

You're in charge. What information should the public know about automated decision systems used by state governments?

Many US state agencies use AI-powered automated decision systems that score medicare claims for fraud, predict which city blocks will have the most crime, measure job applicants' personality traits, and more. There are few rules in place to regulate these AI systems.

Social Studies

AI Snapshots

321 of 382

You may have considered…

where they are used
if these systems have known bias against certain groups
if someone can opt for a non-AI judgment instead, if they’re uncomfortable with the decision

Explore further: Compilation of NYC Automated Decision Systems Task Force Comments

Social Studies

AI Snapshots

322 of 382

Left. Left. Right? Left.

What do you think are the potential benefits and risks of using AI in dating apps?

On dating apps, users indicate potential matches and mismatches by swiping left or right. Some apps use AI to order potential matches and detect personality markers.

Social Studies

AI Snapshots

323 of 382

You may have considered…

AI that could learn a user's preferences in a partner
AI that helps users who match figure out what to talk about
AI that filters out harassment from a user's messages
AI is often biased and could exhibit common social prejudices

Social Studies

AI Snapshots

324 of 382

Top dogs

Use your Google search skills to find out: is this AI fact or fiction?

Zero of the top 10 internet companies, which are also pioneers in the use of AI, were founded in Europe.

Social Studies

🚨 🚨 Internet access required!

AI Snapshots

325 of 382

This is…

AI Fact!

Here are the top 10 internet companies as of early 2022:

Amazon – USA
Alphabet – USA
JD.com – China
Meta – USA
Tencent – China
Alibaba – China
Suning.com – China
ByteDance – China
Netflix – USA
Paypal – USA

Social Studies

AI Snapshots

326 of 382

Hey, that’s my neighbor…

How would you design an AI that can learn to solve this problem using images from the country’s war zones?

Early in Russia’s 2022 invasion of Ukraine, officials scrambled to identify dead soldiers, prisoners of war, and enemy combatants. In the midst of so much chaos, dedicated Ukrainian leaders have been overwhelmed by the task.

Design

challenge!

Social Studies

AI Snapshots

327 of 382

You may have considered…

Who would benefit from this technology
How images would be collected
What happens if the technology is inaccurate
Who has access to the acquired information

Social Studies

AI Snapshots

328 of 382

Out with the old, in with the new

How could this technology be most effectively used for good?

A new AI tech called CoupCast predicts whether countries are likely to have a sudden leadership change soon. The company gathered data about factors that contributed to coups in the past to train their AI to make predictions, successfully predicting 2021 coups in Mali and Chad.

Social Studies

AI Snapshots

329 of 382

You may have considered…

using the tech to increase aid to countries that might experience a coup
helping explain how multiple factors combine to influence sudden leadership change
keeping the tool openly accessible and well-explained so it’s transparent and easy to use

Social Studies

AI Snapshots

330 of 382

AI field trip

How would you design an AI that can learn to solve this problem using images of museum exhibits?

Museum visits are a great way to get excited about history. However, people in major metropolitan areas tend to have access to more museum exhibits than their counterparts in smaller cities and rural areas.

Design

challenge!

Social Studies

AI Snapshots

331 of 382

You may have considered…

How people will access the technology
Representing 3D objects based on 2D images
Quality of the museum experience
Impacts on attendance at local museums

Social Studies

AI Snapshots

332 of 382

AI 101

What do you think an average person needs to know about AI in order to make good everyday decisions?

While the US is one of the strongest AI superpowers, the average person doesn't know much about AI. In a recent study, only 12% of Americans could pass a basic test about AI.

Social Studies

AI Snapshots

333 of 382

You may have considered…

how AI uses data
what the definition of AI is
how AI can be biased
the future of AI
careers in AI

Social Studies

AI Snapshots

334 of 382

Flying solo

Who could benefit from banning this technology and who could be harmed by it?

A grim wartime question: should a drone, a gun, or a bomb be able to decide on its own whether to attack with lethal force? A UN conference recently moved to restrict the development of these autonomous weapons.

Social Studies

AI Snapshots

335 of 382

You may have considered…

autonomous robots make military operations easier and cheaper, increasing their frequency
people are reluctant to trust major moral decisions to machines
robots don’t have human limitations like needing oxygen

Social Studies

AI Snapshots

336 of 382

Voting via AI

Use your Google search skills to find out: who am I?

→ I believe that "diverse minds create diverse solutions."

→ I invented PRIME III, an open-source voting technology that allows users to cast their vote using touch or voice.

→ My research is on human-centered computing and AI.

Social Studies

🚨 🚨 Internet access required!

AI Snapshots

337 of 382

I am…

Dr. Juan Gilbert!

Creating a voting technology designed to accommodate all voters regardless of mobility, vision, hearing or other impairments
Runs the Human Experience Research Lab at the University of Florida

Social Studies

AI Snapshots

You may have considered…

How complicated dishwashing tasks are
Teaching AI to manage dynamic physical work
Impacts on the job market

Social Studies

AI Snapshots

342 of 382

Now hold still for 3 seconds

How would you decide whether this technology was overall beneficial or harmful?

The Baltimore Orioles' stadium now uses AI-powered tech to scan attendees as they enter. The tech can scan 1000 people per hour and doesn't require pat-downs, cutting wait times. However, the stadium has not released the accuracy of the technology.

Social Studies

AI Snapshots

343 of 382

You may have considered…

how often the technology lets weapons through
how often the technology wrongly flags a weapon that isn’t there
privacy may be increased if pat downs are phased out
how the AI determines if a weapon is present

Social Studies

AI Snapshots

344 of 382

For your review

How should governments decide whether an AI tech should be approved for wider use?

The European Union recently approved the use of ChestLink, an AI-powered chest x-ray reader. ChestLink either generates a report if it finds the x-ray healthy, or forwards it to a radiologist for review if it detects an anomaly.

Social Studies

AI Snapshots

345 of 382

You may have considered…

how accurate the technology is
how easy it is for doctors to use
if it’s an improvement over a doctor in accuracy or speed
whether patients are willing to trust an AI with their scan

Social Studies

AI Snapshots

346 of 382

Sound alike

Use your Google search skills to find out: is this AI fact or fiction?

Most AI assistants are male.

Social Studies

🚨 🚨 Internet access required!

AI Snapshots

347 of 382

This is…

AI Fiction!

Siri, Alexa, Cortana, Google Assistant, and more all have female-sounding voices. This phenomenon is an example of gender bias in AI.

Social Studies

AI Snapshots

348 of 382

Not so whole foods

How would you design an AI that can learn to solve this problem using USDA reports?

A food desert is an area that has little to no access to affordable nutritious food. Areas with high rates of poverty are more likely to be food deserts, meaning their residents have to travel far distances to find healthy choices.

Design

challenge!

Social Studies

AI Snapshots

349 of 382

You may have considered…

Why food deserts exist
Motivating private companies to open stores in new areas
Making cross-city travel easier
Programs to help residents pay for healthy groceries

Social Studies

AI Snapshots

350 of 382

AI for a better workplace

Use your Google search skills to find out: who am I?

I think robots are stupid.
I originally studied psychology and wanted to go into medicine.
I believe that while technology may displace workers, their output will get redirected to things that are much more productive.
I founded Catalytic, a company that makes workplace automation software.

Social Studies

🚨 🚨 Internet access required!

AI Snapshots

351 of 382

I am…

Sean Chou!

Interested in “crypto, startups, data, automation, RVs, nerd stuff”

Social Studies

AI Snapshots

352 of 382

AI for zoomers not boomers

You're in charge. How would you address this issue of age bias in healthcare AI?

Are you looking out for grandma and grandpa? New research shows that healthcare AI may not help older people as much as it should. A lack of data on older individuals plus a resistance to technology among this population means that healthcare AI generally works better for the young.

Social Studies

AI Snapshots

353 of 382

You may have considered…

gathering more data from older people
designing AIs that specifically help older people
build trust with older people by explaining how the AI works and its limitations

Social Studies

AI Snapshots

354 of 382

Mine! Mine! Mine!

Use your Google search skills to find out: is this AI fact or fiction?

The Federalist Papers, a collection of essays written by John Jay, Alexander Hamilton, and James Madison, include 12 “disputed” essays: papers claimed by multiple authors. The authors of the disputed essays were first discovered using AI.

Social Studies

🚨 🚨 Internet access required!

AI Snapshots

355 of 382

This is…

AI Fiction!

The authors of the disputed papers were discovered through careful text analysis, but not AI. You can see in the chart below the most common words in Paper 10, written by James Madison:

Social Studies

AI Snapshots

356 of 382

Up and coming

What questions would you ask about a country to figure out if it could become an AI powerhouse?

While China and the US are clear leaders in developing new AI technology, they aren't the only countries trying.

Social Studies

AI Snapshots

357 of 382

You may have considered…

whether it has a strong education system
whether its education system includes computer science and engineering
how much funding it has for AI projects
how rapidly it usually adapts to technology trends

Social Studies

AI Snapshots

358 of 382

More diversity in AI

Use your Google search skills to find out: who am I?

I study intelligent systems for healthcare delivery.
I believe that technology "represents the values of the humans that are behind the design."
I am the co-founder of the nonprofit AI4ALL, which aims to increase diversity in AI education, research, and policy.

Social Studies

🚨 🚨 Internet access required!

AI Snapshots

359 of 382

I am…

Fei-Fei Li!

Born in 1976
Professor at Stanford University
Previously worked at Google Cloud, focused on “democratizing AI technology and lowering the barrier to entrance for businesses and developers”

Social Studies

AI Snapshots

360 of 382

Founding great-great-great-grandpas

How would you design an AI that can learn to solve this problem using images of founders?

Black and mutilracial descendants of the Founding Fathers have faced challenges in getting recognized by the founders’ families and estates. While genetic testing can determine relation, there are many descendants who have no idea they might be eligible for testing.

Design

challenge!

Social Studies

AI Snapshots

361 of 382

You may have considered…

How to source images of potential descendants
Traditional methods for tracing family history
Accuracy in AI image recognition
What happens after descendants are identified

Social Studies

AI Snapshots

362 of 382

Smoke and mirrors

How would you design an AI that can learn to solve this problem using photos of Earth?

Traditional rectangular maps are scaled by latitude to make navigation at sea more accurate. Land masses farther from the equator are depicted larger than they really are, making continents like Africa seem smaller. Making an accurate 2D map to visualize a 3D planet presents a problem to cartographers.

Design

challenge!

Social Studies

AI Snapshots

363 of 382

You may have considered…

New ways to visualize maps
How people will react to unfamiliar maps
Collecting accurate images of the planet
What 2D maps are used for

Social Studies

AI Snapshots

364 of 382

Taxes, am I right?

You're in charge at your local government. Would you adopt this tax-calculating AI and why or why not?

They set off the Revolutionary War. They strike fear into the hearts of everyday citizens. How do we make taxes fair and beneficial for all? Financial firm Salesforce is building an AI tech that finds the optimal tax rate for both workers and politicians.

Social Studies

AI Snapshots

365 of 382

You may have considered…

if the AI would lead to political backlash for you
if the AI would build more trust in the tax rate
whether it’s easy to explain why the AI chose the tax rate it did

Social Studies

AI Snapshots

366 of 382

Recognizing the face recognition debate

What are the potential benefits and risks of facial recognition technology?

In early 2020, the city of London began rolling out live facial recognition in some public locations to “scan crowds and identify individuals in real time”. At the same time, the city of Berlin decided to cancel its effort to install facial recognition in train stations due to claims that it was unconstitutional.

Social Studies

AI Snapshots

367 of 382

You may have considered…

being constantly monitored feels like an invasion of privacy for many
the technology isn’t 100% accurate and can lead to mistaken identity
the technology has been used by authoritarian governments
people have low trust in companies to use the tech for good

Social Studies

AI Snapshots

368 of 382

AI for GTA

How would you design an AI that can learn to solve this problem using using court records?

In LA, grand theft auto carries a sentence of up to 3 years in prison. But judges have the power to decide between, no jail time, 3 years, and anything in between. Left in the hands of individual judges, sentencing can end up being racially biased.

Design

challenge!

Social Studies

AI Snapshots

369 of 382

You may have considered…

Bias in previous court records
How sentencing guidelines are determined
Who should build this technology
What happens to people who have already been unfairly sentenced
Potential pushback in the legal community

Social Studies

AI Snapshots

370 of 382

The masses love karaoke

Should politicians be allowed to use deepfakes when running for office and why or why not?

South Korea’s recently elected president used deepfakes in his campaign: AI-generated videos of him speaking that appear 100% real. His avatar was created using 20 hours of video and made him appear friendlier than usual, answering questions about his personality type and favorite karaoke songs.

Social Studies

AI Snapshots

371 of 382

You may have considered…

whether a deepfake is a reasonable reflection of a person, if they approved it
how much time and money deepfakes might save for a campaign
if it’s realistic to be able to prevent deepfakes

Social Studies

AI Snapshots

372 of 382

Catching Snorlax but it’s for my education

Use your Google search skills to find out: who am I?

I am curious about financial machine learning, ethics, democracy and philosophy.
I studied Pokemon Go, asking questions like "do games teach better than schools?"
I once wrote a paper called "Are Algorithms Affecting the Democracy in Brazil?"

Social Studies

🚨 🚨 Internet access required!

AI Snapshots

373 of 382

I am…

Victor Silva!

Ph.D. Student at the University of Alberta, Canada
Publishes in both English and Portuguese
Shared that working organizations like LatinX in AI makes him “feel embraced by a multicultural community”

Social Studies

AI Snapshots

374 of 382

Can’t trust this

Use your Google search skills to find out: is this AI fact or fiction?

Humans tend to favor suggestions from automated decision-making systems while ignoring contradictory information made without automation. This tendency is called "automation bias".

Social Studies

🚨 🚨 Internet access required!

AI Snapshots

375 of 382

This is…

AI Fact!

Humans tend to think of AI as more objective or impartial compared with our own judgments. It’s natural to follow the decision made by an AI without questioning it.

However, AI systems are just as flawed as humans are. It’s important to question why and how an AI comes to its conclusions.

Social Studies

AI Snapshots

376 of 382

Automating wall street

How would you design an AI that can learn to solve this problem using using stock prices?

Human stock traders make decisions based on knowledge, experience, facts, and intuition. A successful trader will have a 50 - 60% accuracy rate which can translate to millions or even billions of dollars of mistakes over time.

Design

challenge!

Social Studies

AI Snapshots

377 of 382

You may have considered…

How humans use facts to make predictions
How to account for the loss of human intuition
What makes an acceptable level of accuracy
Who has access to this technology

Social Studies

AI Snapshots

378 of 382

Dating texts (but it's not what you think)

Who could benefit from this technology and who could be harmed by it?

An AI called Ithaca can “restore missing text of damaged inscriptions, identify their original location, and help establish the date they were created … Ithaca achieves 62% accuracy in restoring damaged texts … and can date texts to within 30 years of their ground-truth date ranges.”

Social Studies

AI Snapshots

379 of 382

You may have considered…

could help researchers understand the history of old texts
could cause political conflicts if artifacts need to be returned to their home country
if the AI is very accurate and easy to use, archaeologists may be out of a job

Social Studies

AI Snapshots

380 of 382

Can you hear me now?

You're in charge. How would you respond to this report?

Have some companies resorted to collecting audio data for AI from video calls even while you're muted? Researchers at the University of Wisconsin discovered evidence that some video chat platforms listened in while users were muted some or all of the time.

Social Studies

AI Snapshots

381 of 382

You may have considered…

directly requesting clarification from companies about whether they do this
asking about how that data is used
passing a law preventing this practice or fining companies that have used it

Social Studies

AI Snapshots

382 of 382

Social Studies

Want to help us improve

AI Snapshots?

Take our 5 minute survey to share your experience:

tinyurl.com/aieducatorsurvey

AI Snapshots