Deciding about, by, and together with algorithmic decision-making systems

According to Alan Turing, Artificial intelligence (“AI”) is “the science and engineering of making intelligent machines, especially intelligent computer programs.” The most advanced AI systems employ complex “machine learning” techniques which use algorithms to deduce decision rules from input data and store them in, e.g., decision trees or neutral networks (algorithmic decision making, or “ADM”). Over time, the AI tool improves itself by learning from its past decisions, correct or incorrect. 

The overarching ambition of this project is to examine whether there are limitations to this kind of ADM, within the range of AI systems used today. ADM systems are becoming increasingly popular among cash-strapped criminal justice systems. 

In the US, civil liberties unions including the ACLU have even advocated their use at all stages of the criminal process to avoid possible human biases. The uptake of ADM within the criminal justice system, coupled with the grave potential consequences of erroneous decisions, makes this an important research area to compare, on the one hand, how humans make decisions about other humans compared with how ADM systems alone make the same decisions, with the ways in which humans in conjunction with ADM systems take decisions about other humans and the limits to the use of ADM systems. A closely related question is how a given polity decides whether and how to use an ADM system within its criminal justice system.