Tuesday, June 13, 2023

Lie to Me

     In a perfect world, lawyers do not defend the guilty. We know who the guilty are, and they are adequately punished. This may not be impossible for long. The more data that we collect about an individual, the more we know about them. If you had a camera trained on O. J. Simpson for his entire life, you would know that he was a murderer. If you were a superintelligent AI system, you would likely not have to try that hard to become the world's greatest detective (sorry Batman!) and convict O. J. of murder. Maybe there are actual lie detection techniques that certain AI systems will be good at, but even just by combing through massive amounts of data and using simple inference, I am sure that the policing systems available in the future will be extremely powerful. Powerful enough to trust, and powerful enough to do away with the current "jury of your peers" legal system. Now, there is a trade-off to this, the same trade-off we always face: safety vs human rights. 

    Authoritarian regimes focus on safety. Not safety of their citizens, but safety of their regime. They would want to know when a civilization was lying: "no, I wasn't at the protest last night." They will not want their citizens to have use of this technology: "hey, did you see that Robot 3000 proved that Xi Jinping was lying last night about the Uyghurs?" Use of advanced lie detection in the legal system will certainly change human interaction. Fooling modern-day lie detector machines is a bit of an ironic, because polygraphs are shown to actually not work in any sort of reliable fashion. What about in the future? If your dog knocks over a vase, and then tries to fool you into thinking it happened on its own, would you believe it? We can see right through the attempts of animals to deflect blame or straight up lie. They simply do not have the required mental capacity to string together a convincing argument. That may be us in the future, trying to convince our technocratic overlords of our innocence. It does matter if we turn a 10% wrongfully convicted rate into a 0%, and a 90% rightfully convinced rate into a 100%. What matters much more is what the sentencing requirements are for breaking the law. What matters most is who is writing the laws.

No comments:

Post a Comment

Doing Good, or Not Doing Bad?

      Effective Altruism, as a philosophy, is very simple. Basically, the argument is that if you shouldn't do bad in the world, that me...