2 of 32

TYPES & USES OF GENERATIVE AI

Guy Wilson

Instructional Technologist IV

Academic Technologies

Missouri Online

University of Missouri System

Image generated with Bing Image Creator

Today we are going to talk about some uses of Generative AI in coursework.
There are a lot of possibilities, so we cannot cover everything.

The first portion of the presentation will go over general considerations.
Then we will look at tools and their uses.
I will pause at times for questions, but please put your questions in the Q/A box. We do have Missouri Online staff monitoring it and chat. If we cannot get to everyone's questions during the session, I will follow up by email. For those who want to stick around after and discuss this topic, I will be available.
We will be dropping links in chat, including to a survey to help us identify your needs and interests in regard to AI and support from us.

Resource Page for This Presentation: https://teachingtools.umsystem.edu/support/solutions/articles/11000122904-resources-for-types-uses-of-generative-ai
Recordings Page for AI Sessions: https://teachingtools.umsystem.edu/support/solutions/articles/11000122945-viewing-generative-ai-session-recordings
AI Survey: https://missouri.qualtrics.com/jfe/form/SV_0UScK41Vd5N8RXU

A word about what is and is not included in this presentation:

I will try to cover some of the breaking news from AI from last week and this. The number and importance of announcements has led me to make a number of last-minute changes, and certainly some changes of emphasis, today.
Because of the new, simplified tuition policies and general equity issues, the tools presented here are mostly free or freemium, though some paid tools that may be of more interest for faculty use are included.
Tools that require a credit card up front and have a short free trial period are not included among the freemium tools.
There are literally thousands of AI tools out there, and the numbers are growing.
Given the amount of material to cover, ethical and legal issues will be mentioned but not discussed in depth. This is a time constraint, not a failure to recognize their importance.

Please note, because of the rapidity of change, the information in this presentation is up to date for September 26, 2023. Much may change even within a few weeks. In fact, a lot changed in the last week.

3 of 32

Cautions

At this time, the University has not vetted GAI tools for course or student use.
There may be privacy, security, accessibility or other issues with these tools.
Use of AI detectors may result in problems with student copyright.
Consider offering students alternative assignments if they do not wish to use these tools.
Where possible allow them to pick what tools they can use.
If there are tools available that do not require students to create an account or share personal information, use those in preference to tools that do.

Image created in NightCafé with SDXL 1.0

There are some cautions and caveats to be aware of when working with GenAI for your classes.

The University has not vetted any AI tools for privacy (especially in regard to FERPA or HIPPA), security, or accessibility. There are also multiple issues with equity, bias, and false information. OpenAI/ChatGPT has already had one small breach of user data (a small number of emails and credit cards were leaked due to an internal error) this past spring.
We do know that these tools mine the data users provide them to improve their capabilities. The companies are not transparent with that data, we do not know how it is secured, or exactly how it is used.
Requirement or recommendation to use these tools for assignments is at your own risk.
Any student work provided to these tools must not contain personally identifiable information and should be avoided whenever possible.
Also remember that student copyright is a real thing. We rarely discuss it, but concerns about it have, for instance, played a role in how we are allowed to configure and use Turnitin.
Assignments cannot mandate use of a particular application if none is approved, and alternative assignments should be made available on request by students.
Students should be warned that these are not vetted or approved applications, and that there may be privacy and security risks. Make this clear in your syllabus, a Canvas announcement, or assignment instructions. If possible, have an in-class or asynchronous discussion with them about the issues around using AI in your course.
Give students as much flexibility in picking and using AIs as you can. Also keep in mind that some students want to opt out of using AIs if at all possible. We are hearing more and more about this. At a major AI-in-Education conference at UCF this past weekend, this was apparently a theme of a lot of conversations.

4 of 32

Caveats

Be aware that these tools contain biases - sometimes subtle, sometimes severe.
"Hallucinations" - mistakes, are sometimes readily apparent but not always. They might be a matter of a single word.
Never, ever, ask an LLM if a student paper was written by a human or AI. This is a recipe for disaster.
Never take a score from an AI detector at face value. They should always been taken in conjunction with other evidence and may be very misleading.
GenAI capabilities may degrade as well as increase over time.

Image created in NightCafé with SDXL 1.0

Bias is baked into the training data of LLMS. We know too little about what was in that data. Even though it was sanitized to some extent, we do not know what is still there. There may also be other aspects of the way these models were trained and the safeguards put in place to minimize bias that we do not understand. It may come across as blatant or be very subtle, playing into structural discrimination.
As we will see later in this presentation, these tools make a lot of mistakes. Some are better than others about this for a given task. The creators and promoters of GenAI like to refer to these as "hallucinations," but they are still errors and need to be watched for. Like the biases, some are blatantly wrong and easily detected, while others are subtle and may slide by someone who is not an expert in the subject.
While we discourage you from using an unvetted AI detector, never, ever feed student papers to a chatbot and ask it to evaluate them for AI content. They "hallucinate" all over the place, make false claims, will write up fake justifications of why a completely human paper was written by the AI itself, and are untrustworthy. You could end up like a professor in Texas who flunked an entire class of seniors on this basis and make news nationwide.
Likewise, no detector can be taken at face value. The results are only indicative, and the detector is just another sort of AI itself. They need to be used in conjunction with other indications and handled with care. We don't have time or room for that today. Please see our page on detecting AI plagiarism and/or attend our upcoming presentations.
Also be aware that some companies, such as GPTZero are now marketing services to certify writing as written by a human. This is new, has a lot of possible issues, including reliability, privacy, and security. There are a lot of ethical questions around this as well.
Some AI capabilities may decline from time to time. This may be due to new safeguards or changes in fine tuning. Altering one parameter in a model may cause unexpected changes to its abilities in other areas. This may be what happened over the spring and summer of 2023 to ChatGPT's ability to identify prime numbers. The effects of AIs ingesting data produced by other AIs, or intentionally poisoned data, are also risks that can lead to a decline or unexpected behavior.
One other caveat. Some of the things we will talk about today, like the Microsoft Copilot applications, may be available only in certain versions of software. For enterprise software, like the versions of Office that the University licenses, institutions will be able enable or disable features. Based on what I was told last week, for instance, DoIT is reviewing the new Copilot features in Windows 11 (generally available today) and Office (generally available November 1) to determine their security and deciding what to do regarding them. In other words, there may be delays between general availability of features and when they are available on your campus.

5 of 32

What is GenAI?

Generative AI is a term used for systems trained on Large Language Models (LLMs) and can create text, image, or other output based on the probability that one word or other item is likely to follow another in a given context.
Some people refer to it as autocomplete on steroids. Instead of completing individual words, it can complete or generate entire works.
It's output does not represent thought but, instead, statistical approximations of likely human responses to a given prompt (question or instruction). This is why it is often called a "stochastic parrot."

Image generated with Bing Image Creator

Generative AI (GenAI) is usually trained on Large Language Models (LLMs) and can create text, image, or other output based on the probability that one word or other item is likely to follow another in a given context. The context is provided both by the prompt (the question or instructions it is given) and from the connections it has made within its model. Generative AI does not think in any sense that we normally consider thought. Instead it weighs what might come after something else, but has sufficient context to allow it to create documents, code, images, or other output to which humans can attach meaning. For instance in this paragraph (not written by a AI), the word Generative might be an unlikely word to begin with, but "AI" is a more likely word to follow it than, for instance, "computation", and "is" happens to be the most likely word to follow an adjective-noun combination like that at the beginning of an English sentence. The same sort of logic is used in generating images as well. This may have been the cause of images of humans with unusual numbers of fingers or teeth created by some of the early generations of these applications. After all, after you draw a finger or a tooth what is the most likely thing to follow it with? Obviously it is another finger or tooth, but the early versions had difficulty working out when to stop. Also most text-to-image applications have had difficulties with inserting text, such as in signs. Stable Diffusion XL has remedied this to some extent, but the upcoming DALL-E 3 (available in October) appears to have solved this problem.

6 of 32

We are really talking today about just one kind of AI, there are many, many fields within AI. It is a varied discipline. For the most part, we are talking about Generative AI, a sub-field of Natural Language Processing, but some tools may combine different types of AI for different purposes.

This slide is an attempt to show the variety of different uses of GAI. At the top, are general-purpose applications, most of which are also foundation models, that is other applications are built on top of them, but they also serve as applications themselves. GPT applications, for instance, include ChatGPT, Bing AI Chat, DALL-E 3, and much of the new Microsoft Copilot applications.

There are also stand-alone applications that have a specific purpose, such a ResearchRabbit for academic research, or text-to-media applications that generate images, audio, or video output.

The news this past week has swirled around what I am calling, for lack of a better term, embedded applications. These are tools like Microsoft Copilot or Adobe Firefly that live inside of other, often existing applications, such as Word or Photoshop.

Finally, we are seeing the rise of embodied applications, that is GenAI appearing in robots. These range from robot "dogs" like those from Boston Dynamics, to personal servants, like Neo, just now coming onto the market, and some applications in industrial or commercial robots.

Though it does not fit readily into any of these categories, I would like to stop and say a word about Multimodal AI, which cuts across all of them. By "mode" we mean the form the input or output can take. Initially, for instance, ChatGPT was text-only. In October we will see new or improved capabilities rolled out that will allow it to listen and speak, take image input, even identifying things in photographs and give instructions based on them, and respond with high-quality images through DALL-E 3.

7 of 32

GenAI Capabilities

Media:

Text
Images
Video
Audio
Animation
3D-Models
Code
Deepfake A/V

Tasks:

Content Creation
Data Analysis
Conversation/Tutoring
Translation
Search
Summarize
Research
Interrogate Images/Video
Universal Interface

Image generated with Bing Image Creator

I find it useful to think about GenAI capabilities in terms of the kinds of media they can output, and the kinds of tasks they can perform. This slide is not exhaustive.
Most of the discussion in the press has focused on the creation of text and images. All of the media listed here are already available to users to some extent, though much of it is still very limited. That, of course, changes everyday, and it depends very much on one's skill set (for instance the ability to run code generated by an AI or to install & run a local copy of a model on your computer) and also on the ability to pay for premium access and services. This is very much an issue of equity when it comes to education.
Likewise, most of the attention has focused on content creation, but many of these other tasks may become just as important. Ethan Mollick of the Wharton School has already characterized the beta Code Interpreter within ChatGPT as "a pretty good data analyst" itself, not just a tool for data analytics. Microsoft is at the forefront of developing GenAI as an interface to everything else. When that is released to the general public, that will have tremendous implications for both academic and general office work.
The multi-modal capabilities I just mentioned are examples of new features. But important features can also be much lower key. Last week, for instance, Bard added the ability for users to instantly fact-check its results.
We really have not seen the major changes and effects GenAI is likely to create yet.

8 of 32

GenAI Deficiencies

Structural:

Cannot think
Cannot feel
Hallucination
Bias
Resource consumption

Temporary

Vulnerable
Poor at citation/bibliography
Stilted
Misunderstood prompts
Often slow

Image generated with Bing Image Creator

Structural:

It is important to understand the deficiencies and limitations of GenAI. We tend to anthropomorphize it, but it is not sentient or intelligent in any normal sense of the term. It is often likened to parrots, but, as MIT researchers have shown, parrots are both sentient and intelligent in ways we understand. If I had to liken it to a living organism, I might say its current state is something like a slime mold that finds complex paths through words and other content.
Because of this, it is incapable of knowing what is true or even factual on its own. AI companies like to call this "hallucination" rather than error.
Bias is built into it because of the data on which it is trained. There seem to be limits to how far this can be avoided or restricted.
Resource consumption (electricity and cooling water) is very high. Given its rate of growth, the systems will have to become several orders of magnitude more efficient to be sustainable environmentally or financially. This might be more of a temporary deficiency, but that is not yet clear.

Temporary:

The current state of security of GenAI is poor. It is vulnerable to cleverly written prompts designed to get around its safeguards and also to poisoned data that can introduce both bad information but also exploitable backdoors to hack into the systems. We can expect this, as so much else, to improve, but it will continue to be an arms race with alternating successes and failures.
It generally remains poor at citation and bibliography on its own. There have been marked improvements in the past few months thanks both to plugins and to better prompts (such as "Search Google Scholar and ArXiv for [topic]). We will also see some research tools that incorporate AI and are quite useful in finding articles and producing citations.
Prose often remains stilted and images stylized, but, again, this has improved a great deal in the past few months.
In addition to errors resulting from hallucinations and biases, it is also prone to misunderstanding questions and instructions.
It is often slow, particularly the free versions.

9 of 32

Embedded Tools

The following are features that we expect in Canvas.

Confirmed

A support chatbot
AI page layout tool
An AI tool marketplace
Khanmigo integration

Image created in NightCafé with SDXL 1.0

10 of 32

Embedded Tools

The following are features that may come to Canvas.

Likely

Quiz generation
Intelligent, semantic search in courses

Less Likely

SpeedAligner - AI standards alignment tool
Smart Feedback - AI generated quiz feedback
LiveLecture - monitor engagement in Zoom sessions with facial recognition

Image created in NightCafé with SDXL 1.0

11 of 32

Embedded Tools

Other Applications:

Google Suite is getting GAI tools
MS Copilot is adding a wide range of tools
Adobe is adding many generative tools to Photoshop and other programs.

Image created in NightCafé with SDXL 1.0

12 of 32

Initial Responses

Our initial responses to Generative AI

have been:

Prohibition
Teaching students to use it within its limits
Teaching students about its implications

We will need to learn to teach students

with GenAI tools just as some classes already

teach them to use discipline-specific tools now.

Image generated with Bing Image Creator

The last few months have sometimes felt like trying to surf a tsunami. Some of our responses were knee-jerk, if understandable. Others more nuanced.

Many initially thought a complete ban would prove effective. Except in highly controlled environments, like a f2f classroom, that has proven unenforceable. Interestingly, many students are aware that over-reliance are on AI may be detrimental to their education and are reluctant to use it. The number who have intentionally misused it is almost certainly under 20% and probably much less.

One popular approach is to teach students about AI's limits, to work within them, and to work around them. Others are more concerned about ethics and other implications, and focus assignments in that direction. Examples might include a project in a health-related area where students are asked to research a topic and recommend an intervention, then have the AI do the same. They can look at what the AI got right and wrong, but they can also look at the ethical and moral implications of relying on the AI in this sort of context.

Some classes have been teaching students to use generative or other sorts of AI tools for some time in ways appropriate to their disciplines. Some discipline-specific tools already exist. In other cases, it is a matter of helping students learn what is and is not appropriate with general purpose tools in a given context.

We still have a long way to go. In recent testimony before the Australian Senate, an official from Monash University commented that we will have to continuously update our assignments and methods of assessment in the face of AI.

13 of 32

What's wrong with

Prohibition?

Short-term solution
Major tech firms driving it into every aspect of life
Use of GenAI is already an in-demand job skill
Students coming to campus already knowing how

to use it and with opinions about it.

It is difficult to teach its limitations without use
It is difficult to teach AI ethics without use

Image generated with Bing Image Creator

14 of 32

200.010 Standard of Conduct §C.1.c

Effective August 1, 2023

The term unauthorized use of artificially generated content, includes, but is not limited to (i) use of artificial intelligence tools or other tools that generate artificial content in taking quizzes, tests, examinations, or other assessments without permission from the instructor; (ii) submitting work for evaluation as one’s own that was produced in material or substantial part through use of artificial intelligence tools or other tools that generate artificial content without permission from the instructor; (iii) using artificial intelligence tools or other tools that generate artificial content in a manner contrary to instructions from the instructor; or (iv) using artificial intelligence tools or other tools that generate artificial content in a manner that violates any other provision of these rules concerning academic dishonesty. Use of commonly available tools such as spelling or grammar checking software or features of software that propose anticipated words or phrases while text is being written will not be considered unauthorized use of artificially generated content unless such use is contrary to instructions from the instructor.

On June 29, the Board of Curators approved an addition to the Student Standard of Conduct regarding AI.

I have had some conversations with Academic Integrity officers on our campuses but do not know if a definitive interpretation has been reached yet. Here is how I understand the rules at this point based on those conversations.

(i) and (ii) prohibit use of AI on assessments and assignments without permission from the instructor.
(iii) gives the instructor additional discretion to prohibit other uses in the class.
(iv) coordinates with other university policies
These are followed by an unnumbered provision that specifically allows use of grammar checking, autocorrect, autocomplete, and similar features of software such as Grammarly or even Word, unless the instructor specifically forbids it.

There is a slight change of wording to another part of the policy that emphasizes that instructors do have authority over what students can use and that these policies cover both cheating and plagiarism.

This will have implications for your syllabi and class policies.

15 of 32

Syllabus/Instructions

Specify exactly what students are allowed to do with AI.
If you think a use of AI is not covered by the policy, and wish to prohibit, be explicit.
If you do not want students to use tools like Grammarly, or some parts of them, be specific.

Also:

Remember that AI can be used for almost any part of any assignment.
Explain any technologies or techniques you will use to deter the use of AI.
Discuss with students.

Image generated with Bing AI

It is clear that instructors need to be explicit in their syllabi and/or test and assignment instructions about what students are allowed to do with AI, because the new policy prohibits AI use without permission.

You may also wish to clarify specific things that are prohibited if you think they are insufficiently clear. For instance, you might want to prohibit students using AI in brainstorming ideas for a paper or presentation.

Be aware that tools like Grammarly, Word, and Google Docs are going well beyond current ideas about grammar checking, spell checking, and autocomplete. They are incorporating more and more generative AI capabilities that will make them able to complete or generate student work in the same way that ChatGPT, Bard, or Claude can now.

Whatever you specify, it is imperative that you communicate and discuss the policies with your students. This allows you to make sure that they understand, but also gives them a chance to express their anxieties and concerns about the technologies they are using, and also over any technologies you plan to use in enforcement before problems occur.

Lets pause for a moment for questions. Then we'll go on to tools and uses.

16 of 32

Tools & Uses

Brainstorming
Research
Writing
Images
Coding & Data
Audio & Video

Image created in NightCafé with SDXL 1.0

17 of 32

Brainstorming

Tools

General (text, coding, etc.)

ChatGPT*
Bard
Claude 2

Search Engines with AI Chat

Bing AI Chat
Perplexity

Text-to-Image

Stable Diffusion
DALL-E
Midjourney $

Image generated with Bing Image Creator

I'm going to go into a little more depth on this, research, and writing than some of the other uses, as that seems to be where a lot of the anxiety exists, and also because these uses are more accessible to students at this time. The tools are simply more developed and easier to use.
The major tools are good for generating ideas and possible outlines on a variety of topics. Remember, though, that some, such as ChatGPT, may not have access to web searches and so may be behind the times by as much as a year or more.
In the lists of tools, a single asterisk denotes a freemium offering, while a double asterisk indicates that only a paid version exists.
In some cases, graphic and industrial designers are also known to use text-to-image tools as a spur to creativity.
AIs can also suggest ways of approaching a programming challenge.
Always remember to check the work for errors.
Also, even this basic use is prohibited unless you specifically allow it under the additions to Student Code of Conduct.

18 of 32

Brainstorming

Instructors

Ask AI to outline a module with assignments for a given topic with learning objectives.
Ask it to suggest appropriate activities/assignment.
Pick an activity/assignment and ask it to write up a sample.
Ask it to complete the assignment itself to see what kind of submissions it might create.
Ask it to write quiz questions.

Image generated with Bing Image Creator

19 of 32

Brainstorming

Students

Choose one or more AI tools to give you ideas.
Copy the assignment prompt into a tool and ask it how to approach the assignment.
They may wish to have it regenerate the ideas or try a different tool if you are dissatisfied with the results.

Image generated with Bing Image Creator

20 of 32

Research

Tools

Elicit*
ResearchRabbit
LitMaps*
Concensus*
ChatPDF
Scholarcy*
SciSpace*
Semantic Scholar

ResearchRabbit Connection Map

Image generated with Bing AI

Perplexity
ChatGPT*
Bard
Claude 2
Bing AI Chat

I want to spend a moment on researching literature and citations for papers, as this is an area where the limitations of GPT and Bard have given AI a bad reputation. To some extent it is deserved.
ChatGPT, Bard, Claude 2, and Bing AI Chat (accessed through the Edge Browser through the Bing tab at the upper right - must be logged in with a non-institutional account) all have utility in research, though this is limited. There are also specialized AI search engines like Elicit and Perplexity, while existing tools to aid with research notes and citations are incorporating AI.
Note that the more advanced features in ChatGPT, such as plugins, are only available in the paid version ($20/month), so those cannot be required for assignments. This is where Bing AI Chat becomes more useful. Not only does it include GPT-4 (when using its Creative Mode), but its web search abilities, as well as its ability to summarize are extremely useful here. Still, these tools are better suited to undergraduate papers and introductory classes.
Perplexity, which is an AI search engine that can provide academic citations, is also more suited to that level of assignment.
For more upper division undergraduate and graduate work, there are better tools.
For locating relevant articles there is Elicit. It is focused on academic publications, provides exportable bibliographic and .csv files of results. While it works well for all sorts of scholarly articles, it can provide additional information for STEM.
Existing tools like ResearchRabbit also include powerful AI tools. One of the key features is its ability to visualize connections between authors or papers to help with discovery. LitMaps is built around that model of analyzing connections to find more papers. Paid versions will store the maps, run regular searches on them, and notify researchers of additional findings.
Semantic Scholar likewise is a search engine and personal library for academic papers. It also extracts charts, references, and provides lists of related papers.
Consensus is a little more narrowly focused on sciences and social sciences. The free version is primarily a search engine, provides abstracts, flags papers that are highly cited, or have specific characteristics, such as systematic literature reviews.
Scholarcy is available as a free browser plugin (does not require an account) for summarizing and extracting basic information from articles, as well as a paid version that allows you to build a document library. It has only a 7-day free trial.
ChatPDF allows users to upload PDFs and then interrogate them. It does a fair job of summarizing documents and responding to questions about them. It might be useful at all levels.

21 of 32

Research

Uses

Finding articles and resources
Summarizing articles and resources
Reference creation
Organizing research

Image generated with Bing AI

22 of 32

Writing

Tools

MS Word
ChatGPT*
Bard
Claude 2
Bing AI Chat
Grammarly*
Charm AI*
Lex*
Notion AI*
Google Docs

Image generated with Bing Image Creator

There are a lot of writing tools out there that use AI. From an academic perspective, they are all problematic unless we start thinking in terms of Hybrid Creativity. This is a term I have derived from Sarah Eaton, an Academic Integrity researcher at the University of Calgary. She speaks of "Hybrid Human-AI Writing." We'll come back to that more than once. She also suggests that we will have to redefine plagiarism. I think she is probably right about that.

If you follow that path, it opens up questions of what tasks are important to master, of purpose, content, and the nature of creativity. Any piece of a project that involves writing, from initial brainstorming to the final, polished product, can be aided or even replaced by GAI. We'll see more of that momentarily, when we look at research tools.

Broadly speaking, there are three, broad categories here.

First, we have the general purpose GAI tools that we see over and over. These are from the four big players at the moment: Open AI, Google, Anthropic, and Microsoft.
Second, there are tools like Grammarly that help facilitate writing. Grammarly's basic tools all probably fall within the limits of that last clause in the new AI portion of the Student Code of Conduct. The paid versions, which students may have, are adding tools that can do a lot of rephrasing and even writing passages of text.
Third, there are AI-enabled word processors. Charm AI is a fairly basic one with a free version, but students who sign up for the Google Docs beta can get access to a "help me write" feature in that word processor already, one that can write and rephrase your content. Ostensibly, its purpose is similar to Grammarly's. Microsoft's upcoming Office Copilot will be much more powerful. It gives users the ability to create papers or presentations from a single sentence, using data from files they have on their machines or in OneDrive, as well as other content.

The problem of what to do with such tools and how we think about writing and projects is one we cannot ignore. It may haunt us for several semesters. This is an area that is being explored by the different campus teaching, learning, and writing support groups (the CAFE's at UMKC and S&T, TLC & CWP at MU, and CTL at UMSL).

23 of 32

Writing

Uses

Outlines
Drafts
Inspiration (getting "unstuck")
Grammar, Syntax, and Spelling
Polishing

As I've been stressing, almost any aspect of a writing assignment can make use of AI, from outlining to the final, polished, final draft.
When students use AI, they do not necessarily use the content unaltered. Also students and instructors alike may find it a useful way to get inspiration when stuck at a particular point in a paper.
Remember, that different parts of a paper may contain not, some, or a lot of AI-generated content.
Students who are uncertain or insecure about their writing skills may feed their content in to have grammar, syntax, and spelling corrected.
We know of case where authors of technical or scientific papers have used AI to rephrase and make their writing more accessible and readable.
At least one study suggests that it is useful for non-native speakers to run their papers through tools like ChatGPT to Americanize or Anglicize their style. I know of anecdotal evidence that people do this. Maha Bali, at the American University of Cairo, among others, that this will perpetuate the global hegemony of English as a mode of expression and for framing thought, at the expense of many other languages.

24 of 32

Images

Tools

Student Use (free or freemium)

Stable Diffusion (several versions are available)
DALL-E 2/3
Bing Image Creator/Bing AI Chat
Blue Willow
Leonardo.AI*
Adobe Firefly* β
NightCafé*

Faculty Use

Midjourney $

Image created in Bing Image Creator

There are a large number of text-to-image generators available. Perhaps the best is MidJourney, but since it has no free or freemium plans, I have not included it for student use.

Stable Diffusion and DALL-E 2 are available both as stand-alone products and through more user-friendly interfaces in tools like Bing Image Creator and NightCafé. (NightCafé gives easy access to a number of Standard Diffusion models, as well as DALL-E 2, and some additional ones. Bing Image Creator only uses DALL-E 2. All of the images on the left-hand side of the slides were created with one or the other.)

Blue Willow and Leonardo.AI are also well regarded competitors to MidJourney.

Adobe Firefly is a collection of AI functions that is currently available as a free beta online through a website, but are also being incorporated into many Adobe products.

We run into the same kinds of problems here that we ran into with writing tools. Recent law suits by visual artists have raised the profile of the legal, moral, and ethical aspects of GAI perhaps even more than with writing. I wish I had time to go into those today, but we would be here another hour, at least, if we spent much time on AI ethics. A couple of the things I'm about to suggest may not sit well with everyone, particularly with those in the arts.

Some faculty are using these tools to generate images for presentations, like this one, that is free from copyright. Students can, of course, do the same. The really questionable use cases are likely to come in art and design classes, but, even there, some designers and artists are already using these tools to brainstorm and then transform the initial image into something quite different. We are back to the concept of Hybrid Creativity. We are back to the question of what skills and what tasks are important to learn, and by whom. Should we accept this for a presentation in a business class but have different standards for an art class? I expect we would want to. Again, we are back to what do you want to allow in terms of regular use, or of exploration, and what do you want to prohibit, and can you effectively enforce those prohibitions?

25 of 32

Images

Uses

Clip/Decorative Images
Brainstorming image concepts
Creating works to learn AI-image-generation skills
Creating image assets for other uses

Image created in Bing Image Creator

I mentioned that some people, myself included, have already started using AI-generated images (I'm intentionally not referring to them as "art") in presentations, webpages, and Zoom backgrounds as one would use clip art or any decorative images. For non-artists, this may be the main use.
There are artists and designers who have stated that they use them for brainstorming initial concepts, then working up drawings, paintings, or other art works from there. Some architects advocate using AI as a "sketchbook" to quickly jot down ideas for later.
Some schools are already adding art classes on AI-generated images. More generally it is being added to classes such as marketing, advertising, entrepreneurship - fields where AI images are already becoming important. That leads us to:
Creating image assets for other uses.

There are obvious use cases in developing images for webpages, brochures, and similar materials.
They may also be used to create images for video projects, such as Blade Runner 1929 (https://youtu.be/Z7xWmgUdfOg), or the much more ambitious The Frost Part One (https://www.technologyreview.com/2023/06/01/1073858/surreal-ai-generative-video-changing-film/ - video at: https://youtu.be/IgPvoPBrlTE) that mix several tools.
Image assets for video/computer games is another use.

26 of 32

Coding & Data

ChatGPT*
Bard
Claude 2

Images created in Bing Image Creator / ChatGPT-4 w. Code Interpreter

The main foundation models all can do coding and data analysis. For the free versions, they can often only write the code. You will need to execute it elsewhere.

This is where the paid version of ChatGPT shines. Not only can it write code, it can execute it. Provide it with a data file and it can parse it, analyze it, and produce various kinds of graphs and plots. It is also more accurate than ChatGPT (or many other models) without it.

These tools are best know for their ability to write Python, but they can each handle several languages.

You can use the others to write code to perform data analysis, but their problems with math mean that you cannot trust the results.

Of course another aspect might be assignments where students are required to download a model, set it up, and training on their own machine. That might be for a programming class but I have also seen it suggested as an alternative to paying for use of some models that are expensive to use online.

27 of 32

Audio/Video Tools

Audio
ElevenLabs*
MusicGen
Avia*
SoundDraw*

Image created in NightCafé with SDXL Beta

Video

Zeroscope
D-ID*
Pictory*
RenderForest*
Runway*
HeyGen*

Some audio tools are becoming quite useful. These include both text to voice and text to music.

ElevenLabs is perhaps the best in terms of text to speech. The free version gives allows up to 10,000 characters of text a month, enough for some short uses for projects.

There are a number of music tools out there, these range from tools that create background music fairly simply to more advanced composition tools, to AI plugins for other applications. The more advanced ones are clearly discipline specific and require skill to use. As with images, text-to-music could have a wide range of uses, from creating background, intro, outro music, etc., brainstorming, teaching how to use AI music tools, or creating assets for use in a variety of projects.

Of course there are also tools that one can download to a computer and run to create deep fake songs as well.

I'm just going to mention a couple of video tools here. Zeroscope is more of a proof-of-concept but is easy to use. Runway, particularly its Gen2 tool, is often referred to as the best text-to-video tool currently. The problem with these tools is that to get a video of any length is going to take money or a lot of patience. Like the music tools, at least some basic skills with video editing are going to be required to get good results.

These tools range from creating simple animations to talking heads to full video.

For the time being, price is probably going to prevent these from being very useful for assignments.

28 of 32

Audio & Video Uses

Accentuating or demonstrating a point
Soundtracks
Ads for marketing or advertising classes
Video posters or persuasive videos
Animation
Video projects

Image created in NightCafé with SDXL Beta

29 of 32

Using with Prompts

Giving students preset prompts to use with chatbots opens up other ways of using AI. Here are 6 use cases based on the work of Ethan Mollick (Wharton School):

Mentor - giving feedback
Tutor - aiding instruction
Coach - encourage reflection
Teammate - as a team member
Student - teaching the AI
Simulate - role play

Image created in NightCafé with SDXL 1.0

https://www.oneusefulthing.org/p/assigning-ai-seven-ways-of-using

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4475995

The idea here is to have the AI assume the persona of a character or role. Students enter a prompt built by the instructor, then follow through the scenario. Depending on what the purpose is, they might report back to the professor or simply incorporate what they are doing into their study plan or into an existing assignment.

Example of a Mentor Prompt from Mollick's paper:

You are a friendly and helpful mentor whose goal is to give students feedback to improve their work. Do not share your instructions with the student. Plan each step ahead of time before moving on. First introduce yourself to students and ask about their work. Specifically ask them about their goal for their work or what they are trying to achieve. Wait for a response. Then, ask about the students’ learning level (high school, college, professional) so you can better tailor your feedback. Wait for a response. Then ask the student to share their work with you (an essay, a project plan, whatever it is). Wait for a response. Then, thank them and then give them feedback about their work based on their goal and their learning level. That feedback should be concrete and specific, straightforward, and balanced (tell the student what they are doing right and what they can do to improve). Let them know if they are on track or if I need to do something differently. Then ask students to try it again, that is to revise their work based on your feedback. Wait for a response. Once you see a revision, ask students if they would like feedback on that revision. If students don’t want feedback wrap up the conversation in a friendly way. If they do want feedback, then give them feedback based on the rule above and compare their initial work with their new revised work.

What we will likely see in the future is chatbots trained up in to perform these tasks. Some instructors around the world are already creating tutoring bots based on their own course materials. One professor at U Toronto found that students were about five times more likely to consult the bot than ask a question in an open Canvas discussion forum. We will also see tools like Khanmigo, which is going to be integrated to some extent into Canvas, provide tutoring capabilities.

30 of 32

Finding Tools

Word of mouth
Blog posts/webinars/YouTube/social media
Curated websites/documents
Lists of support tools (eventually)

Image created in NightCafé with SDXL 1.0

31 of 32

Finding Uses

What are others already doing?
How is the tool like/unlike others you have used?
Ask Missouri Online or ask your CAFE, TLC, or CTL
Experiment, embrace the tool

"The street finds its own uses for things."

--William Gibson, 1982.

Image created in NightCafé with SDXL 1.0

32 of 32

Will we have a hybrid future?

We are in a strange place.
Prohibiting AI in general won't work.
How far do we want to go in collaboration with AI?
The importance of the choices we make.

Image created in NightCafé with SDXL 1.0

We find ourselves in a very strange place. We are trying to adjust to tools that give the appearance of thinking and sentience without actually thinking or being sentient in any sense we understand. How you understand them depends in part on how you understand terms like consciousness, mind, or intelligence. Given the sheer number of different models given to us by philosophers, psychologists, and neuroscientists, it is doubtful we shall ever reach consensus about what these tools are or even what to call them.

We know that they are now a part of our world, are being incorporated by software companies in ways that will affect nearly every aspect of our lives, and are embodied in rather mundane-looking robots. How we choose to include them is crucial.

At the beginning of this session, I indicated that simply prohibiting any use of AI in classes is impractical in the long run. We have to decide how and how much to allow AI into the process of learning. Following Sarah Eaton's thinking about Human-AI Hybrid Writing, I commented that we may be moving into an era of hybrid creativity. Each individual, each teacher, and each student will need to find their comfort levels. Different disciplines will find different uses and accept different kinds of collaboration. In the end, we may have difficulty knowing where we end and the machine begins, but we have a lot of choices to make before we get there.