– Reflections on Artificial Intelligence – A Guide for Thinking Humans

Recently I read Melanie Mitchell’s brilliant book Artificial Intelligence: A Guide for Thinking Humans. It is a comprehensive introduction to the world of AI for anyone who wants to know when, how and how far we have come. The book was published a few years ago, so it doesn’t have all the latest developments in the field, especially the impact of LLMs, but it has enough.

While reading the book, at various stages, some thoughts triggered which I noted down. This essay is first in a series of six.

Simulated Masters and the Myth of Neo

In one of the most iconic scenes in the movie The Matrix, Neo opens his eyes, exhales, and murmurs “I know Kung Fu.”

And then we are with him and Morpheus in the dojo, watching him do gravity defying stunts. I recall that his learning, specially after he became a helicopter pilot just like that, became a running joke in my group of friends. We joked about how to learn to ride a bicycle. And the answer was obvious. It involved no falls, no bruises, no cuts.

I recalled this scene when I was reading about reinforcement learning in Mitchell’s book.  

But then, Neo did not learn Kung Fu. He acquired it. Simply like a download replacing discipline. No aching muscles, no torn ligaments. No silent humiliation in front of a patient master. Just the performance of competence, stripped of history.

And yet, there he is, with Morpheus, dancing between punches, gravity-defying, beautiful. What we see is the afterimage of thousands of hours that led to this mastery. The cracked knuckles, the broken bones of forgotten monks and masters compressed into executable code.

But then we realize, Neo is still plugged in. The real Neo has never touched a wooden floor with bare feet. His spine has never known the repetition of a thousand front kicks. He carries the simulation of mastery, not its memory.

So who is real? The one who stumbles on the dirt outside the dream? Or the one who floats through fantasy with a master’s grace? And then the more dangerous question: which version of ourselves would we choose to become?

It is, after all, the seduction at the heart of artificial intelligence. Not just that machines can learn. But that we might learn like them, efficiently and without pain.

Reinforcement Learning, or the Romance of Repetition

Machines learn by trying, failing, adjusting and then trying again. Like children, but much faster.

This is reinforcement learning. A really simple method, classic carrot and stick. Reward what works and penalize what doesn’t. Let the system iterate until the pattern emerges—a strategy in chess, a movement in robotics, a path through the maze.

And while reading this, I recalled Neo’s “I know Kung Fu”. Can we build a loop like that for ourselves? Could we bypass the experience of learning often marked by red pen (at least in my case). Could we wrap ourselves in simulations, train against tireless AI opponents. Rehearse life until we get it right?

Could we, like Neo, wake up knowing Kung Fu?

The Neo Fallacy

Neo doesn’t understand Kung Fu. He performs it. He has the moves, perfected by a program. His body knows the sequence, but doesn’t have the memory of either effort or pain. He is a virtuoso with no memory. He has not earned the right to teach, because he hasn’t truly learned. There is something missing in that loop. Something dense and difficult.

And so we must ask: what is learning, really? Is it a transfer of data, or a transformation of being? When I say I know how to console a friend, is that the same as having watched a thousand simulations of it? What does it mean to know something not with the mind, but with the whole body?

We learn not just through correction, but through context. Through the shame of forgetting a friend’s birthday and through the beauty of failing at something you love. Through the awkward dinner where you say too much, and the walk home where you wish you hadn’t.

Anyone who has chatted with the GPTs of the world, and noticed their loss of context or sudden hallucinations, knows this: machines skip the sting. The LLMs are prediction machines. But the sting is the thing.

Reinforcement Learning for Humans—Is It Possible?

There are ways, of course, to simulate experience. The theatre of the mind. Athletes use visualization; therapists use roleplay. VR headsets promise immersion, and there are training modules now where you can rehearse your next difficult conversation with an AI actor who never gets tired. There are tons of YouTube videos and tutorials on how to prepare for a job interview with help from AI.

But these are shadows on the cave wall.

Because we are not code. We are consequence (another one of Keanu Reaves’ iconic ones, different movie though).

We remember, we regret. And we anticipate heartbreak and flinch before the blow. And that anticipation is what makes us who we are. A bundle of hope and despair.

Reinforcement learning may teach a robot to pour tea. But it cannot teach why the tea matters. Why the guest across the table must be heard. Why the silence between sips might be sacred.

If we could learn like Neo, we might gain skill. But we would lose story. And what is mastery without myth?

Maybe a day will come when skill along with context could be transferred in a data packet. Till then, iterate at your own pace.