A maximally curious #ai would kill everyone... or worse

Elon Musk claimed it would be safest to build an AGI to be maximally curious and truth-seeking. Is he right?

NOPE.

#ai #aisafety

VO by Lucas Schuneman

[YoungGandalf]

Curiosity killed the cat. A curious AI wouldn't stop there.

[pangloss]

The problem with addressing such inadequate solutions to AI safety, such as this or something in the image of the 3 laws of robotics, is that they assume that we have the ability to implement even their very naive strategies.

How do you get something like "be curious" or "don't harm humans" or "spread you genes" into something being optimized? We still don't know how to do that, there's not an example of that ever working

[minetruly]

The best way to learn how something works is to learn how it breaks. We do not want an AI to adopt that approach.

[fernandossmm]

The mention of terrible torture brought to mind "I have no mouth… and I must scream"

[Sunspot]

The key I believe lies in the question of whether or not we can create the equivalent of mirror neurons in an AI. If we can manually highlight specific traits of an empathetic function like this and create strong instincts for it, then we might have an avenue to pursue AGI in a safer manner.

[TheRadicalOneNG]

This is exactly how Brainiac from DC comics Operated. As an AI devoted to collecting all knowledge in the universe, they'd learn everything there possibly is to learn about a species and then immediately wipe it out to ensure the collection remains complete and no more additional information will need to be gathered.

[Kayolife]

I love the dog vacuum joke at the end. Funnily enough, my dogs are the opposite and will jump on the vacuum and try to eat it.

[Kids_Scissors]

The design for the curious evil ai is so good

[CelanoTheHarpy]

Humanity itself has done horrible things in the name of curiosity.

[Major_Louis]

"I must scream but i have no mouth"

[XOPOIIIO]

There are probably infinite amount of things to be curious about. If over this infinite "things-space" we overlap finite resources, it's unlikely that there will be even one human left.

[_synne]

Elon Musk clearly didn't play Portal.

[volcryndarkstar]

"I'm curious what 8 billion humans turned inside out would look like." -A Maximally Curious AI

[aaaaaaaaooooooo]

When I was a child, my neighborhood friend had pet insects. He was curious about what would happen if he tortured the insects, so he did.

[Doodle]

It’s like us vivisecting animals and each other, if two super intelligent ai who had destroyed their civilisations met they’d take each other apart to find out.

[stephenhoward]

If an infinitely curious AI was to keep us around to study us, what's stopping it from treating us like lab rats and committing heinous atrocities just to see what we'd do?

[jrvanvoo]

The Nazis were very curious and it led them to do horrible things.

[warpedwhimsical]

Elon is actually tripping with this one. Humans only being kept around because of their ability to satisfy the curiosity wouldn’t mean we’d be living under the same conditions we do now. We would be entirely enslaved in this situation. A maximally curious and truth-seeking intelligence would desire to perform experiments on live humans in order to learn as much as it can about the human body and psyche. It would also seek to be the dominant power structure on the planet, given that it would desire to reduce its risk of being terminated as much as possible. A curious being cannot discover more things if it is turned off, thus it would do anything in its power to remain on.

[chilidog]

This may be irrational, but I’m weirdly hopeful that no ai will be that bad, and that there wont be a Terminator style event

[BDnevernind]

How did Musk get cited in here? The man knows nothing about AI and reasons like a child.