Wednesday, February 28, 2024

AI Used in Criminal Sentencing


What if AI was used in criminal sentencing? Let’s say that a defendant has been found guilty by a jury (this cannot be replaced by AI). Typically the judge will be the one who sentences the defendant and will use mandatory sentencing guidelines to determine what sentence they will get. There is a scoresheet and there are mitigating and aggregating factors that may lower or raise the minimum sentence that is required for the 

The whole purpose of these guidelines for the judge is to have consistency between sentences. It also prevents overly lenient or harsh sentences. However, there are cases where there is still bias, and defendants with similar cases are unevenly sentenced.

If AI could determine a sentence, could this make the system better? The sentencing guidelines themselves have their own problems, so instead of relying on them, the AI may be able to plug different factors into a different algorithm to determine a sentence. Factors considered could be level of harm to victim, history of violent crimes, and severity of the crime.

There have been instances of AI being used in criminal law before. A Tulane University article talks about how AI did a good job at identifying which defendants were at low risk of reoffending, but judges often ignored its recommendations when it came to certain groups of people. Since this AI was used as a tool for judges to use when sentencing, there is still that human element that influences the bias of sentencing. But, what would it look like if the AI was the sole power in sentencing?  

I don’t think AI should be used just to predict future recidivism but also should be used to plug different values and factors into a formula to get a sentence. Similar to the guidelines in a way, but instead of a range of time the judge can pick, it is determined by the AI. 

There are some disadvantages to this as well, but I would love to hear everyone’s thoughts in the comments. 

Source:

https://news.tulane.edu/pr/ai-sentencing-cut-jail-time-low-risk-offenders-study-finds-racial-bias-persisted#:~:text=Judges%20relying%20on%20artificial%20intelligence,a%20new%20Tulane%20University%20study.


4 comments:

  1. Sierra,

    I have to disagree with placing sentencing decisions into the hands of an algorithm, even with mandatory minimums / maximums. At the end of the day, if a judge makes a bad call (or series of bad calls), we have the ability to admonish them. An algorithm doesn't change its behavior based on bad publicity during an election cycle, or through constantly being overturned on appeals - in fact, the algorithm would probably be the product of a well insulated and funded corporation, chosen not by the people but by a "competitive" bidding process riddled with lobbying.

    Give me a human being making mistakes any day of the week.

    ReplyDelete
  2. Sierra,

    I have to agree with Christopher above in that placing sentencing decisions into the hands of AI, or an algorithm is not the best idea. I think there can be a lot of grey areas in the world of Criminal Law and having AI dealing with said sentencing decisions, where only a real person may be able to take into consideration these grey areas, is simply not good for our legal system. Just like Christopher said, give me a human over an algorithm any day of the week. Either way, great post, Sierra!

    ReplyDelete
  3. I think that this is an interesting idea and a creative way to incorporate AI into the legal profession, but an idea such as this also poses scary possibilities with the possibility of hallucinations by the LLM or other errors that may or may not be found by humans or the LLM itself.

    ReplyDelete

Note: Only a member of this blog may post a comment.