Cake
  • Log In
  • Sign Up
    • Researchers at UC Berkeley and Stanford have done research into how accurate algorithms are at predicting the likelihood of a defendant being arrested for a new crime. Their research concludes that algorithms are much more reliable than people, leading to a debate about how involved artificial intelligence should be when it comes to criminal justice reform. The publication from UC Berkeley is here. Super interesting read. Curious to get some thoughts from everyone on here.   

    • In the article they compared predictions by an AI, compared to predictions made by volunteers, and it turned out AI made better predictions. What conclusion can you make? That an AI is better in recidivism prediction than a random dude on the street. Why is that in any way relevant to the court cases that are argued and judged by career professionals? AI can make a better Chinese translation than a random person, sure. But, can it translate a piece of text better an a professional Chinese translator? That's the relevant comparison. This is something else entirely.

      I would be very cautious with applying machine learning to areas like this, or anything related to it.

      Machine learning if fraught with examples of systems learning the wrong thing and applying it blindly. Polar bear detection that recognises snow instead of a bear. Skin cancer detection that detects hospital lighting instead of cancer. And so on. Putting systems like that in charge of human destinies - I would double, and the triple check. And then still defer to opinion of a professional.

    • Interesting article - but reread the second sentence in the paper .....

      "When assessing just a handful of variables in a controlled environment, even untrained humans can match the predictive skill of sophisticated risk-assessment instruments, says the new study by scholars at Stanford University and the University of California, Berkeley." The article then goes on to discuss judgements involving not a handful of variables, but a very complex series of variables - where it states AI logic can outperform humans.

      I submit that Machine Performed Algorithms can help humans - judges, parole officers - make better decisions - but I have major reservations about machines making legal decisions, without oversight by human professionals - not random volunteers, but human professionals.

    • Great thoughts, @jpop! Yeah, I think AI should definitely be used as a tool to help professionals make more informed decisions than the other way around where they replace them. That’s for sure. Your analogy about the Chinese text (中文) made a lot of sense. 👍