Thread by @AlejandroPiad

Today is #PhilosophyFriday 🤔!

The year is 2035. You're sitting comfortably in your L5 self-driven car, zooming across the highway.

Suddenly, the truck in front drops a big boulder. In a split second the car AI has to make a choice: brake or dodge... 🧵👇

⚠️ Even with a full brake you're not guarantee to survive the crash.
🚙 To your left there's a van driven by a human.
🚴‍♂️ To your right there's an unprotected biker.

❓ What should your AI do?

>

⚠️ If the AI chooses to brake, the boulder will still hit you, with potentially disastrous consequences.

And it's very hard to sell you an AI that will make the choice with the least chance to save your life.

🚙 If it chooses to hit the van, you'll most likely survive, and the other driver has a pretty decent chance too.

🚴‍♂️ If it chooses to hit the biker, he/she will almost certainly die, but you'll maximize your chances of not getting hurt.

If the AI is trying to save you at all costs, it will kill the biker. That's the optimal solution for you.

But we probably don't want that, so let's program it to hit the van, the choice that minimizes the odds of someone getting killed, right?

👉 Now, let's suppose it's two bikers, one each side, but here's the catch: one of them is wearing a helmet.

What's the option that minimizes the odds of someone dying?

Even if by a very small chance, now hitting the biker with a helmet seems the best option.

🤔 Suddenly, doing the right thing as a biker (wearing a helmet) makes you *more* likely to be hit by a car.

We probably don't want that either, or everyone will be biking with the least possible protection.

But wait, it gets worse... 👇

👉 What if there's two identical bikers, but one is carrying a baby, the other is carrying a pregnant women, and you're carrying two small children?

What do you value more, a potential life lost or a certain life lost? Does it matter what is the expected lifetime left?

If you're the one driving, you'll do whatever your instincts tell you.

And whatever that decision is, any court or judge will have the consider the fact that you didn't have enough time to think.

But an AI had enough time to think. It made a conscious decision.

👉 That decision was either preprogrammed, or computed from a preprogrammed formula, or learnt from data, or reinforced, ...

In any case, there are humans behind it who had plenty of time to think carefully about it.

And they made a conscious choice, or didn't they?

A possible solution is to refuse to make a choice at all.

🎲 In an impossible situation let the AI flip a coin, so whatever happens it was due to luck.

But is it morally correct to refuse to solve a literal life-or-death problem if we have even a slight chance to solve it?

❓ Why is this question so hard? Well, moral is tricky. It feels immoral to even consider there might be a predefined answer for who's life is worth more.

But an AI needs a formula in this case, even if it's "choose random", and we better come up with one we can live with.

As usual, if you like this topic, reply in this thread or @ me at any time. Feel free to ❤️ like and 🔁 retweet if you think someone else could benefit from knowing this stuff.

🧵 Read this thread online at <apiad.net/tweetstorms/ph…>