Listening to today’s Babbage episode I felt like not sneezing. You know the feeling of the build up to a sneeze. First your eyes squint, then your cheeks rise, you take a deep breath, and the sneeze never comes. Deeply disappointing.
The first part is a beautiful breakdown of AI, consciousness, and sentience. The episode does a great job at explaining the current fuzz over AI LaMDA. Leading experts such as Dan Dennett and David Chalmers explain why LaMDA, most probably does not experience consciousness and sentience, although it claims to do so. But those experts agree that future efforts might be able to create conscious superhuman AI. They agree that consciousness as a concept might be a mystery, but enough neurons with enough information might just be the clue. So, let’s assume that they are right before creating such a conscious superhuman AI. How should we treat it? What are its rights and duties? Should it be punished by the law like the rest of us, if breaking it?
Here the conversation derails. The answers given by leading experts varied from inclusion of AIs into our rule of law, to outright banning the research on superhuman AIs. I disagree with both. Why are we tackling AI’s rights and duties from a human perspective? Should we not try to learn from AIs and adapt our own morality, and rule of law? I would even go a step further: Is it not our obligation, as humans, to be superseded by superhuman AIs?
Law
Let’s take a step back. Let’s assume we build a superhuman AI, and for the sake of argument it comes in human-robot form, like Terminator. We build its brain, its limbs and feed it with information. None of that was the robot’s choice. It now goes out and kills people by running over them with a car GTA style. Should this conscious superhuman AI be put in prison for the rest of its life? The answer must be no. We have either provided the superhuman AI with inappropriate physical attributes or have fed it too much GTA gameplay. But none of that is the superhuman AI’s fault. At any moment in time, it is determined by its information and physical abilities. It is also clear that the current state of the AI is dangerous for society. Thus, we need to confine the superhuman AI, work on physical and information tweaks, until it is fit for society. The goal should be maximizing the happiness of conscious beings, not unnecessarily punishing them for being built and programmed in a harmful way.
Every word from the previous paragraph can be translated to humans. A person is nothing but nature and nurture, none of which he is responsible for. To the question of whether we should lock up superhuman AI in the same way as humans I say, lets work on bettering humans just like we work on bettering superhuman AIs.
Banning AI
Dan Dennett goes a step further. He wants to avoid the question as a whole, and outright ban research, and creation of superhuman AI. He argues that because we are not able to confine superhuman AI in our moral, and judiciary system, we shouldn’t create “counterfeit humans” (alignment problem). I agree we should be careful, just like we should be careful when raising children. We want to create responsible and kind human beings. Just cause it could go wrong doesn’t mean we shouldn’t try. Staying the analogy of children: I would take the opposite view on Dennett; Not only should we try our best creating superhuman AIs, we should allow them to supersede us. The example often given is that superhuman AI will see us like we see ants. We don’t want to kill them but if we need to build a house for our happiness, we will kill all animals living on that land, disregarding their right for life. And we should: we are maximizing conscious happiness. The same counts for Superhuman AI. If its capability for happiness supersedes ours by x fold, and we are in the way of their happiness why would expect to have a right stay?
Two arguments:
1. Let’s not be stupid and force our barbaric laws on AI. Rather, lets learn from the process of building consciousness, and adapt our laws adequately.
2. I am disappointed if my children’s consciousness is aligned with mine. I want my child to be a superhuman AI, completely disregarding all of humanity (and hopefully be "super" happy).
*sneeze
Comments