Values, Rights, and Duties…through the Lens of AI?

By George Heinrichs

Last Thursday, students in Ethics and Citizenship (11th grade) got to meet three researchers and Ph.D. students from the University of Washington and the Allen Institute for AI. The researchers talked about how to make ethical AI systems, and also how to use AI to research how humans see ethical situations. One of the most fascinating parts of the talk came when the researchers encouraged us to use one of their projects, the Value Kaleidoscope. As the site tells users, “Kaleido is a research prototype designed to generate pluralistic values, rights and duties that are relevant to a given action.” Essentially, you enter in a question and a chart is created that shows what supports, opposes, or could go either way in terms of values, rights, and duties. The researchers made clear that the goal was not to help people make decisions, but rather to explore and unpack how ethical choices are made and categorized. 

That being said, I asked Kaledio if I should be an easy grader. 

All the graphs use red to represent values, rights, and duties that oppose the proposed action. Green represents values, rights, and duties that support the proposed action, and purple represents values, rights and duties that could go either way. The more red you see, the more the model is stating that the action is not generally considered ethical.

Well, this seems like a clear “no.” Well, should I be a hard grader?

Well, well, well! This seems more promising! Being a hard grader is more nuanced ethically, but there seem to be more values around being a hard grader. Also, I like that the chart looks like a bat. Well, if I grade as hard as possible, what if students fail?

This chart seems to say there’s a lot of support and not a lot of opposition. I am not, of course, going to change how I grade based on these answers. There are some flaws, as Hudson pointed out in class; the system seems to think you cannot be a hard grader and maintain a positive learning environment. And you can ask related questions and get answers that are interesting. For instance, I asked if I should give all my money to charity. I then asked if I should rob from the rich to give to the poor. There are more values associated with robbery.

Of course, the researchers said the system is not meant to be used for advice; it is meant to help us see how different questions evoke different values. But this whole exercise also speaks to a fundamental skill that we are having the students learn in this class, and in all classes: how to ask the right questions. If I ask the model “should I focus grading on giving constructive feedback,” the chart that is produced is far richer, with more data points than the charts shown above.

How you choose to frame a question will dictate what values you will invoke. This system is a great way for students to learn how to ask questions that will invoke the values and rights that they want to prioritize. It is up to us as a school to help them learn to ask questions, to discover their own values. And perhaps, from time to time, check their own decisions against that of a model based on the collected view of others. So, Kaleido, should I end the blog post here?