STEP 5:
ANSWER Multiple Choice and Short Response Questions
QUESTIONS:
Download the following document to answer the questions (submission instructions will be given by your teacher):
Document: Discriminating AI: Can Machines be Unfair? Assessment Questions
Complete the assignment before class and follow your teacher's submission instructions.
All of the questions are listed below for reference. Download the document when you are ready to answer the questions.
1. [Google video] Fill in the blank:
a. In ___________________, people hand-code solutions to a problem step-by-step; meanwhile, in ___________________, computers learn the solutions by finding patterns in data.
b. Say a dataset contains many examples of athletic shoes, but few examples of high heels, thus preventing the machine from learning to recognize high heels. This is an example of _______________ bias.
2. The Google video provides examples of how machine learning (AI) is used in everyday life. For each example, can you think of a relevant app on your phone which uses AI?
a. Navigation __________________
b. Suggestions __________________
c. Translation __________________
d. Voice recognition __________________
5. [TED Talk] How could the programmers of Microsoft's Chatbot "Tay" done a better job?
7. [NPR interview] What are three characteristics of algorithms (machines) that worry Cathy? Briefly describe each one. If you could choose one characteristic to remove from unfair AI, which would it be? Why?
BONUS. Below are two figures, each with an example of unfair AI. For each infer (a) what the AI is trying to predict (b) how the AI is being unfair.
3. [Google video] Multiple Choice: When training a machine to recognize human faces, how can you mitigate selection bias?
a. Add more photos to your dataset
b. Ensure photos in your dataset represent everyone
c. Limit dataset to photos of a certain type, e.g. faces of celebrities
4. [TED Talk] True/False:
_________: Human programmers may have bias which leads AI to make unfair decisions
_________: Say you're turned down for a mortgage or
6. [NPR interview] In the example of teachers' "growth scores," what is the negative consequence of trusting numbers too much? What else should be taken into consideration?
