AI Research

Overview

Starting point:

  • To be intelligent is to understand the world,
  • The machine at issue is electronic,
  • The process of understanding is a semantic process,
  • The semantics of electronics are a mystery.

Turing knew a lot about electronics but didn't address the most important question for AI: How can a computer have semantic content (human  minds do). John Searle's Chinese Room Argument exposed the extent of Turing's omission. To this day, AI research hasn't addressed the semantics of electronics and as a result doesn't understand how a computer could have human-like general intelligence.

LLMs are amazing statistical engines, but its clear from edge cases, "hallucinations" and other fundamental issues that they lack human-like general understanding. Also, they are built from instances of binary differences created by human minds on the basis of already existing human knowledge.

It's the same old problem of Symbolic AI. Humans use human intelligence to define the behavior of the machine. The system itself isn't intelligent. It's as dumb as a brick. 

The same goes for LLMs. Machine behavior is determined by a vast statistical analysis of binary differences created by already existing human intelligence. The LLM itself is as dumb as a brick.

We will know when machines are on the way to human-like general intelligence when they survive in the wild - in a hostile, novel, complex and changing environemnt. 

Some classic papers and books

Some classic papers and books.