Tag Archives: GenAI

Humans Are Not as Different from AI as We Think

Humans Are Not as Different from AI as We Think

GUEST POST from Geoffrey A. Moore

By now you have heard that GenAI’s natural language conversational abilities are anchored in what one wag has termed “auto-correct on steroids.” That is, by ingesting as much text as it can possibly hoover up, and by calculating the probability that any given sequence of words will be followed by a specific next word, it mimics human speech in a truly remarkable way. But, do you know why that is so?

The answer is, because that is exactly what we humans do as well.

Think about how you converse. Where do your words come from? Oh, when you are being deliberate, you can indeed choose your words, but most of the time that is not what you are doing. Instead, you are riding a conversational impulse and just going with the flow. If you had to inspect every word before you said it, you could not possibly converse. Indeed, you spout entire paragraphs that are largely pre-constructed, something like the shticks that comedians perform.

Of course, sometimes you really are being more deliberate, especially when you are working out an idea and choosing your words carefully. But have you ever wondered where those candidate words you are choosing come from? They come from your very own LLM (Large Language Model) even though, compared to ChatGPT’s, it probably should be called a TWLM (Teeny Weeny Language Model).

The point is, for most of our conversational time, we are in the realm of rhetoric, not logic. We are using words to express our feelings and to influence our listeners. We’re not arguing before the Supreme Court (although even there we would be drawing on many of the same skills). Rhetoric is more like an athletic performance than a logical analysis would be. You stay in the moment, read and react, and rely heavily on instinct—there just isn’t time for anything else.

So, if all this is the case, then how are we not like GenAI? The answer here is pretty straightforward as well. We use concepts. It doesn’t.

Concepts are a, well, a pretty abstract concept, so what are we really talking about here? Concepts start with nouns. Every noun we use represents a body of forces that in some way is relevant to life in this world. Water makes us wet. It helps us clean things. It relieves thirst. It will drown a mammal but keep a fish alive. We know a lot about water. Same thing with rock, paper, and scissors. Same thing with cars, clothes, and cash. Same thing with love, languor, and loneliness.

All of our knowledge of the world aggregates around nouns and noun-like phrases. To these, we attach verbs and verb-like phrases that show how these forces act out in the world and what changes they create. And we add modifiers to tease out the nuances and differences among similar forces acting in similar ways. Altogether, we are creating ideas—concepts—which we can link up in increasingly complex structures through the fourth and final word type, conjunctions.

Now, from the time you were an infant, your brain has been working out all the permutations you could imagine that arise from combining two or more forces. It might have begun with you discovering what happens when you put your finger in your eye, or when you burp, or when your mother smiles at you. Anyway, over the years you have developed a remarkable inventory of what is usually called common sense, as in be careful not to touch a hot stove, or chew with your mouth closed, or don’t accept rides from strangers.

The point is you have the ability to take any two nouns at random and imagine how they might interact with one another, and from that effort, you can draw practical conclusions about experiences you have never actually undergone. You can imagine exception conditions—you can touch a hot stove if you are wearing an oven mitt, you can chew bubble gum at a baseball game with your mouth open, and you can use Uber.

You may not think this is amazing, but I assure you that every AI scientist does. That’s because none of them have come close (as yet) to duplicating what you do automatically. GenAI doesn’t even try. Indeed, its crowning success is due directly to the fact that it doesn’t even try. By contrast, all the work that has gone into GOFAI (Good Old-Fashioned AI) has been devoted precisely to the task of conceptualizing, typically as a prelude to planning and then acting, and to date, it has come up painfully short.

So, yes GenAI is amazing. But so are you.

That’s what I think. What do you think?

Image Credit: Pixabay

Subscribe to Human-Centered Change & Innovation WeeklySign up here to join 17,000+ leaders getting Human-Centered Change & Innovation Weekly delivered to their inbox every week.