Segment | Materials |
Introduction (10 minutes) - Machine learning is a form of artificial intelligence (AI) which allows software to learn from data where we instruct the software how to learn.
- It can be used to identify patterns and solve problems.
- Many leading companies (Apple, Google, Microsoft, etc.) are utilizing it.
- Machine learning is slowly becoming more integrated in our daily lives.
- Search engines, personalization, recognition, autocorrection, etc.
- Use of labeled datasets to train algorithms that to classify data or predict outputs accurately.
- Training time (giving the model inputs and correct outputs) vs Test time (model generates outputs given inputs)
| Presenters |
Activity – Rapid Resume Reports (10 minutes) - Each student will get a set of resumes and their corresponding resume screening report scores.
- They will spend 5 minutes reading through the resume reports and filling out the observation sheet.
- Then, they will spend a few minutes discussing with a peer on their observations.
- Have volunteers share what they wrote and any interesting observations they had.
| Participants |
Reflection (5 minutes) - What are some things you noticed about the activity we just did?
- These inaccuracies were the result of biases in machine learning.
- Humans have developed many biases in their time; simple biases include colors or foods, but complex biases can form towards personal identities.
- Computers are not impartial; Our own biases translate into bias expressed by technology
- There are many main types of machine learning biases, here are a few:
- Algorithmic: Bias within the algorithm processing data itself.
- Implicit: Biases within the individuals creating the algorithm.
- Sample: Bias within the sample of data which is being processed.
- Reporting: Inaccurate sample size in comparison to real-world.
- Social: Individual norms leading to biased perceptions of data.
| Presenters
|
Connector (5 minutes) - Resume screening bias against women in Amazon algorithm
- Overconfidence in these algorithms can cause barriers in the workforce and in future job prospects
- Gender bias in previous versions of Google Translate between languages with no gendered pronouns to a language with gendered pronouns (eg. Turkish to English)
- Facial recognition bias against darker skin tones
- A high profile incident involved a chatbot named “Tay”, released on Twitter by Microsoft in 2016. Designed to mimic a “19-year-old American girl” by learning from interactions with Twitter users, it was suspended after 16 hours since it started spewing hate speech (Twitter users taught Tay such speech).
- Machines learn what we teach them, even if it is vulgar.
- How can we mitigate biases in machine learning?
- Have equitable representation in supervised learning.
- Assess for biases in unsupervised learning.
- Have more representation/diversity within the AI field and in datasets.
| Presenters |
Closing Activity – Generative Image AI (10 minutes) - Students open generative image AI tool:
- They will choose one or two of the given prompts and fill out the observation sheet.
- They can also come up with their own prompts to input into the image generator.
- Have volunteers share what they saw and any interesting observations they had.
| Participants |