The black box of thought

One of the biggest problems with artificial intelligence (AI), perhaps the main one of them all, is its black box nature.

You know the information you put into it, you look at the text it returns, you check its correctness or its usefulness, but you have no idea how it did it.

This is very frustrating.

ChatGPT and its colleagues, soon to be hundreds and then thousands, generate results that no one, not even its own creators, had imagined.

Some are pure and simple lies, or fantasies if you prefer to call them that, but others find angles that the user had not anticipated.

Large language models

(

LLM), the class of algorithms to which ChatGPT belongs, exhibit certain emergent properties, properties of the system that cannot be predicted by its components.

The emerging complexity seems like magic, but there is nothing exotic about it.

Neither nitrogen (N) nor hydrogen (H) smell like ammonia (NH₃), so the smell of ammonia is an emergent property.

A gene is a string of letters a, g, t, c, but the information it contains is not in the letters, but in their sequence.

Genetic information is a phenomenon that emerges from those humble chemical compounds that mean nothing in themselves.

If you only study neurons, you will never understand the neural circuit that this article allows you to understand.

Your understanding of language is made entirely of neurons, but there is only one level higher, at the level of circuits.

Emergent complexity is a central theme of science, from particle physics to neurology and beyond.

What is so special about AI generating unpredictable ideas?

Chance doesn't serve as an explanation for me, because I'm not talking about a brother-in-law who utters the first stupid thing that pops into his head, but rather about relevant and meaningful, sometimes fruitful, associations, the kind of idea that makes you exclaim: "What?" Incredibly stupid not to have thought of it!”

I have put this phrase in quotes because it is exactly what Thomas Huxley said when Charles Darwin explained his theory of evolution by natural selection.

A little history-fiction exercise: if ChatGPT had existed in the 19th century, could it have come up with Darwin's idea?

Are you sure not?

Another way of asking the same thing is: what does human creativity consist of?

Darwin had spent five years on the

HMS Beagle

studying fossils, turtles and finches, and it was only after returning to British soil that, by chance reading a book by the Reverend Thomas Malthus, he came across one of the best ideas in history, natural selection: nothing directs the adaptation of individuals;

It is simply that the palm does not adapt, and the population as a whole, made up of the survivors, becomes more adapted by definition.

How incredibly stupid not to have thought of that!

Let's now do a second history-fiction exercise: what would have happened if Darwin had not come across Malthus's book?

ChatGPT's great competitive advantage is that he has always read Malthus's book.

He has read everything, and therefore he can find links—links, analogies, metaphors—between very distant and disparate sectors of knowledge, like the finches of the Galapagos Islands and the economic theories of a reverend.

Human creativity consists precisely in finding those unexpected links, but it requires prior and detailed knowledge of each of the two sectors that are going to be linked.

Deep down, our creativity is a box as black as artificial intelligence.

Both find links between disparate things, but we have no idea how they do it.

That's the thing.

Subscribe to continue reading

Read without limits

Keep reading

I am already a subscriber

_