Finding meaning on the road to AGI

Supposedly the goal of all of this GenAI stuff is to find a path to artificial general intelligence. The GenAI labs have put all of their eggs in the "scale is all you need" basket and they are betting trillions of dollars on that assumption. The problem is that this assumption is fundamentally wrong. Large language models are just that. They are not intelligence models. The work well enough for a broad range of applications, but the only way that they can be said to achieve general intelligence is by redefining general intelligence.

In my book, Algorithms are not enough: Creating general artificial intelligence, I describe what would be needed to achieve general intelligence. More recently, I have begun to describe some initial steps toward achieving it. Among those steps is for a system to represent meaning. Systems can do a lot with symbols related to other symbols, but general intelligence requires that the system also represent that those symbols have meaning and that it exploit that meaning in creating intelligence.

Today's GenAI LLM models depend on the distributional hypothesis that the meaning of a word is how it is used in the language. But, as has been known for more than 40 years, symbols relating just to other symbols is insufficient to establish their meaning. Symbols need to be grounded in the world, not just in other symbols.

The linked article continues the discussion of what it would mean for a computer model to understand meaning. Meaning is not all you need to accomplish general intelligence, but it is essential. Meaning is not just connections to co-occurrence profiles, not just reference, not just a mapping from words to worlds, but is more subtle than these.

At Reliath AI we are building a foundational model of AI that is based on facts, not on word tokens. It does not hallucinate. It is available for licensing and for applications to be built on top of it. It is not sufficient to achieve general intelligence, but, like meaning, it represents an essential step in that direction.