The Logical Perspective

## Is artificial intelligence close to overtaking biological intelligence?

This question is at the intersection of evolutionary biology and artificial intelligence. Current generation artificial neural network models are quite different from human brains, both in terms of architecture and in the way they are trained. Large Language Models such as ChatGPT (which predicts the next word given a starting set of words) are touted as some of the most advanced artificial neural networks, possessing human-like intelligence. Many say (and fear) that in a few years, such neural networks would surpass human level intelligence in every field. However, I feel that there are some fundamental differences between how a human thinks and the way artificial neural networks make us believe that they are thinking. These fundamental differences arise from deeper roots, which in order to explore requires the understanding of how the natural process of evolution has acted upon us over millions of years, and has given rise to the current state of the human brain, starting from nothing but a single cell containing a single strand of a DNA polymer. Once we compare natural evolution to how the current state-of-the-art large language models are trained, it becomes obvious on what's lacking in the best artificial neural network models of today. Of which I can think of two fundamental differences:

(i) The structure of the brain evolves over time. The architecture itself changes. The genetic algorithm (which is a combination of mutation and natural selection/elimination) leads to various different possible candidate brain architectures which gets tested against the real world and only those that are "more fit" remains (natural selection) while the rest gets eliminated (natural elimination). The mutation (errors in copying of the genetic code) leads to slightly different brain structures across different generations of a species. Natural selection and mutation combined, leads to an evolving brain structure over time. In contrast, the current generation artificial neural models are fixed in architecture, and only the strength of the connections change during their training (i.e. when the parameter values are tuned via optimizing the loss function), but there isn't much flexibility in terms of what architectures a neural network can attain.

The animal brain probably started with a very simple architecture, just barely capable of processing some key survival tactics (like should I eat this food or not), and then evolved continuously over millions of years to morph into a more complicated structure, capable of performing more complicated processing (like estimating where one stands in the social hierarchy which was key to attracting mates). A few species have arrived at the ability to count (likely a measure to keep in check the total number of offsprings), and then humans are the only ones who with their increasing level of intelligence were able to craft special weapons for survival (with stone and iron during the stone age and in the iron age respectively) the designs of which required significant imagination skills. Once this level of intelligence was reached, it was easy for the human brain to ponder on philosophical questions (like who created us) and so on and so forth, which in turn led to hypothesizing, evidence gathering, theorizing, analyzing, discovering, inventing, etc. which in a way all fall in the category of "thinking".

(ii) The other major difference between animal brains and artificial neural networks is in the way they are trained. The animal brain is trained with the objective function of survivability, which I think leads to a quite different overall learning than training a next word predictor model. Even if the architecture of an artificial neural network is held constant, just the change in the objective function should lead to significant behavioral differences.

Therefore, the non-evolving nature of artificial neural networks and the way they are trained are key differences that set them apart from becoming "human-like". If you don't have these two ingredients (the flexibility of an evolving brain structure, and the survival of the fittest objective), then you will not be able to arrive at the quality of intelligence that biological beings posses.

***


The Logical Perspective

Biological Intelligence vs. Artificial Intelligence