22.08.2024

Super Human Intelligence

People have speculated that it is possible to create a god-like artificial intelligence. The way to do this is to build a smart computer that could modify itself to become even more intelligent, which would allow it to make even more improvements, and that recursive self-improvement process could continue until its ability far surpassed humans.

If we assume that computers will become far smarter than humans, then we must also assume that intelligence is a form of computation. If intelligence is a kind of computation then humans and computers are on some fundamental level doing the same thing. That would mean humans by themselves, humans using computers, computers using humans, and computers by themselves, would eventually be able to do the same tasks, given sufficient amount of time to finish the computations. Thus, humans should also be able to achieve god-like intelligence by recursively improving their own computational tools. I'm reminded of Douglas Engelbart's thoughts on the topic.[1][2]

Computers are a way of quickly doing lots of logical operations. I think more computation is required to solve some difficult problems, but I do not think that computation alone is sufficient to solve all the problems in the world, because logic isn't enough to make something valid.[3][4] The conclusions derived from the logical operations can only be valid if the initial premises are also valid. The way people test the validity of premises is ad hoc, but the main ways are through some combination of consensus, empiricism, intuition, and trusting authority.

Currently, AI can help solve problems of scale. These tools can generate more media, they can automate more manufacturing, and they can analyze larger sets of data. In these ways AI is superhuman.

Another version of superhuman intelligence is discovering new, fundamental truths about the world. Maybe analyzing larger and larger data sets is enough to do this. I do not know, but, the nice thing about computers is that a person would eventually be able to reach the same conclusions by following the same sequence of operations.