Questioning the AGI Dream

I don't really know what to think about these AGI (Artificial General Intelligence) stories. Many players like OpenAI, Anthropic or Deepmind are pursuing this ancient dream of creating an artificial intelligence surpassing our own.

To achieve AGI, we would first have to define it precisely. But its definition is not really consensus, and a fortiori neither is that of intelligence. On Wikipedia, AGI is defined as follows: "artificial intelligence capable of performing or learning practically any cognitive task characteristic of humans". Does this include the ability to contribute to science? To lead people? To develop the ability to lie? To disobey? Does general intelligence imply some form of autonomy or free will?

One could argue that even without being able to formally define intelligence, this does not prevent recognizing an intelligent person when you come across one. Similarly, it will not prevent researchers from getting closer to AGI without being able to define it. But would we be able to recognize an AGI if we had one in front of us? Would we be humble enough to put a machine on the same pedestal as the human being? That's not sure.

Without completely ruling it out, I am in truth rather skeptical that we will ever achieve it. I am not sure either that the transformer architecture alone allows us to get closer to it. Even by training an LLM on all available knowledge, nothing suggests that we would achieve AGI. We could also increase the LLM size and training time, nevertheless, we are already approaching the limits of what is financially feasible.

Faced with these architectural and economic limits, a new trend is emerging. That of designing systems based on a set of specialized models: MoEs (Mixture of Experts). This approach has gained popularity following rumors that GPT-4 itself is built on this principle.

Even if in the future we realize that AGI is a utopia, it's not that bad. Utopias have been responsible for many advances in the history of science. Moreover, why this desire to pursue AGI? It is not necessary to reach such a level of intelligence to disrupt our societies.