A THOUGHT EXPERIMENT · ANDRÁS KOLOSZÁR
I often come across the idea that decision-making is the limitation of artificial intelligence.
“It can do all sorts of things, sure, but making decisions? It couldn’t do that well.”
01 / THE DECISION
There are several reasons why this makes me wonder why people see it that way (although I think I understand what they mean and why they say it).
First, because the most basic thing those large language models — the ones most people have in mind when talking about AI — do is decide: what is the most likely next word?
It makes a decision:
every.
single.
word.
Just as we do.
We decide whether to say or write the next word. Quite a responsibility, actually...
Second, because I haven’t experienced this inability to make decisions. What I have experienced is that it makes decisions constantly. If I ask it to decide, it decides.
“Yes, but,” the sceptics say, “it often makes bad decisions!”
Hmm… let’s pause for a moment.
Don’t people?
And for another moment. What does ‘bad’ mean? Not the way I would have decided?
How do I know what the right decision is? Could I mean that AI’s decision is ‘wrong’ because I don’t understand it, or don’t agree with it?
The same applies to human decisions. So how is an AI decision different from a human one?
02 / THE THOUGHT EXPERIMENT
Let’s go back to the beginning: given how AI works, the decision it makes is the most likely one. That’s how I understand it. What does it base its decision on? The information available to it and the context. Could it be that there is no ‘right’ or ‘wrong’ decision? That the information available to us is simply different?
Don’t we do exactly the same thing — make the decision most likely to be right, based on the information available to us?
Well... If I look at myself, I have to say:
no.
I don’t make a significant proportion of my decisions based on the information available to me. People smarter than me say I don’t make any of them that way. Information matters, but it isn’t what ultimately decides.
I make the decision my emotions dictate — and in many cases, I rationalise it afterwards.
So if we continue this thought experiment and say it isn’t decision-making, what is the limitation of artificial intelligence?
It could be, for example,
emotions.
How esoterical, much self-help, such wow!
“So what?”
— comes the thought in the mind of an educated person living in a capitalist world...
From what I see and hear, quite a few of us are trying to identify what only we can do, where our value lies, and what we’ll be needed for in a future with AI.
Feelings. Emotions.
Here, for example, there is a clear distinction between us and AI. As I currently understand it, these are ours. Of course, emotions can be imitated. Based on the information available to it, ChatGPT can write or speak the messages most likely to carry the right emotional tone. But it doesn’t feel.
And imitating a feeling and actually feeling it are two very different things.
Who do our emotions matter to?
Other people.
03 / WHAT BELONGS TO US
We joke about AI arguing for the opposite of what it claimed just one sentence earlier. Many people take this to mean that AI is malfunctioning. That it’s wrong. Yet all that happened was that it learned something since its previous statement that made a different answer the most likely one. For example, it learned what I think.
I won’t persuade a person to change their decision that way — except, perhaps, on matters of very little importance. Not because I can’t give them enough information (although that is a limitation too), but because they have an emotional attachment to their decision.
So if our emotions are one of the fundamental things that distinguish us from artificial intelligence, could it be that we should strengthen this ability?
WHAT?!
Emotions?
At work?!
What kind of value is that? Who’s going to pay for it? Is that how I’m supposed to become a more valuable resource than AI?!
Possibly. Maybe.
Because workplaces are full of people. With feelings. With emotions. The goods those workplaces produce are consumed by people. They choose products and services based on emotions. They make decisions on an emotional basis. All those reports count (for something), but leadership decisions are based on feelings and emotions too. Many people buy and sell shares based on emotions.
I don’t think this is the answer to the AI puzzle taking shape around us. But I do think it’s an important piece.
And no, I don’t think decision-making is where we’re better than AI.
I wouldn’t stake keeping my job — or its future — on ‘I can make better decisions than AI’.
04 / WHAT I WOULD STRENGTHEN
I would be much more willing to stake it on this: if empathy matters in my work, if creating value through my work involves another person’s emotions (and I think that wherever there’s a person, it does), then I want to be good at expressing and understanding human emotions.
I’m reluctant to strengthen a skill at which AI is probably a thousand times better than me, or will be.
I’d strengthen what
it can’t even compete in.