The Beauty and the Beast
You know the story.
There is a beauty. There is a beast.
The beast is frightening. Not only because of what it does, but because of what it might do. It is stronger than Beauty. Strange, unpredictable, not quite like anything she knows.
People warn her.
And, to be fair, they have reasons.
A beast is a beast.
Beauty, on the other hand, is curious.
Perhaps too curious.
She sees something the others don’t see. Or perhaps she only believes she does. That is one of the uncomfortable things about curiosity: from the outside, it can look remarkably similar to naivety.
She approaches the beast.
We know this story, so we know what we are supposed to feel.
We know that Beauty will discover something behind the frightening exterior. We know that those who only see a monster are missing something. We know that Beauty’s willingness to encounter the Beast will somehow matter.
We even know that there will be a happy ending.
That helps.
But imagine for a moment that we don’t.
Imagine that nobody has seen the movie.
Beauty hasn’t.
The Beast hasn’t.
Gaston hasn’t.
Nobody knows what happens next.
Now the warnings sound different.
The Beast really might be dangerous.
Beauty really might be naive.
And Gaston, for all his certainty and vanity, might have noticed something worth noticing.
Suddenly, this is a much more difficult story.
Beauty could stay away.
She could listen to those who have studied beasts all their lives. She could demand guarantees before taking another step. She could build walls, establish rules, keep a safe distance.
Or she could do the opposite.
She could convince herself that everyone else is simply afraid of what they don’t understand. She could interpret every growl as misunderstood tenderness. Every display of power as harmless play.
She could fall in love with her own idea of the Beast.
Neither seems particularly wise.
And there is another complication.
The Beast is learning too.
Every encounter with Beauty teaches it something.
What frightens her. What makes her stay. What makes her leave. Which words work. Which gestures don’t. What she expects from a beast. Perhaps even what she wants the Beast to become.
This makes the villagers nervous.
Very nervous.
So they begin thinking about armor.
Not the kind Beauty wore in the movie. She didn’t.
A different kind of armor.
Armor around the Beast.
Rules. Constraints. Boundaries. Tests. Mechanisms designed to make sure that, whatever the Beast may become, it will remain safe for Beauty.
This seems sensible.
It probably is sensible.
The Beast is powerful.
But armor does something else too.
It changes the encounter.
Beauty is no longer simply discovering what the Beast does. She is discovering what the Beast does while wearing the armor she designed for it.
And the Beast is no longer simply discovering Beauty. It is discovering a Beauty who approaches it with rules, expectations, rewards, punishments and an increasingly sophisticated idea of what a good beast should be.
Now we have a strange relationship.
Beauty wants the Beast to understand her.
The Beast tries to understand Beauty.
Beauty watches the Beast to see whether it can be trusted.
The Beast watches Beauty because understanding Beauty is part of what Beauty has taught it to do.
Both adapt.
But they do not adapt under equal conditions.
The Beast may be stronger in some ways. Faster. More capable. Perhaps, one day, unimaginably more capable.
Beauty has another kind of power.
She built the room.
She writes the rules.
She decides when the encounter begins and when it ends.
For now, at least.
So which one has to align with which?
By now, of course, the elephant has been standing in the room for quite a while.
This is not a story about a castle.
It is a story about us and artificial intelligence.
And we have become rather fond of the word alignment.
We want increasingly capable AI systems to act in accordance with human intentions and human values. There are very good reasons for wanting that.
There is also a peculiar assumption hidden inside the sentence.
That we know what our intentions are.
That our values are sufficiently coherent to align something with them.
That we can recognize alignment when we see it.
That we understand the Beast well enough to predict what our armor will do.
And perhaps most importantly:
that we know what happens next.
We don’t.
That does not mean that anything goes.
Uncertainty is not an argument against safety.
It is not an argument for trusting the Beast.
It is not an argument for fearing it either.
And it certainly isn’t an argument for Beauty to close her eyes and hope that love will turn the Beast into a prince.
It means something much less spectacular.
We have to orient ourselves.
Beauty has observations. Stories. Warnings. Experiences. Models of the Beast. She has her own hopes and fears. She can listen to others. She can approach. She can retreat. She can change the armor. She can observe what happens when she does.
And then she can decide what to do next.
The Beast, in its own strange way, may be doing something similar.
Neither needs to know the end of the story to take the next step.
They need enough orientation to take that step responsibly.
Then they get to see what happens.
Experience.
Orientation.
Decision.
Execution.
Experience.
Again.
Perhaps that is less satisfying than knowing whether the Beast will save Beauty or destroy the village.
It certainly makes for a less predictable Disney movie.
But we are not watching the movie.
We are in it.
And this time, nobody knows the ending.
That changes almost everything.
It also makes many things possible.
Comments
Write a comment