I found the Detain/Release simulation to be a fascinating experiment. The basic premise is that one must make choices pertaining to detaining or releasing people who had been charged with various crimes. The likelihood of showing up to the court date, committing another crime, and how violent risk the person posed was all ranked from low to high. There was also a statement on the crimes committed, a recommendation for detention or release by the prosecutor, and a statement from the defendant on how jail time would impact their life. In all of this, the user was expected to balance public fear while ensuring that the jail did not exceed capacity. This meant you had to release some of the people charged while ensuring that the public felt safe about your decisions. What was sometimes frustrating as you worked through the 25 cases you were to rule on is that occasionally what would have been a safe bet (someone with low rankings across the board) would suddenly commit another crime and your public fear and jail capacity would go up. I had this happen a couple times and I was worried that very suddenly I would fail the simulation. It felt relatively realistic as it is quite hard to predict human behaviour on such limited information.

Think about the implications and consequences that AI-informed decision making brings to certain aspects of life.

I think what this simulation does a good job of is showcasing what limitations there are with AI in decision making; particularly when human behavior is the key factor in determining the outcome. I felt I was making the right decision based on the information that was available to me but I could not predict the human behavior element. While one may argue that an AI may be able to make purely unbiased decisions (without getting into the biases of those who design the algorithms of AI), AI cannot really assess the ‘human’ element of situations like the cases in the Detain/Release simulation much like the user could not. My takeaway from this simulation is that AI should not be involved in the decision-making process for these types of cases.