Microsoft rolled out an AI chatbot and within 24 hours it began saying racist, misogynistic, and generally hateful things.
Lots of people have come out over the years to say how a powerful AI would be a nightmare and possibly cause the very extinction of the human species. I could sit here and search for links for hours and days and still not grab them all.
But just know that a lot of really intelligent people have been studying the potential calamitous effects of AI on our species. Stephen Hawking’s warned the world about it, and he’s not alone.
I always thought this was a bit silly.
I mean, why would AI automatically become anti-human?
I have a lot of thoughts about the potential of AI and why it’s so fascinating, but that’s maybe for another time. I’ve written novellas about it, and maybe they’ll be published someday.
I love both of those stories and they’re generally positive in their outlook for the future of AI and artificial species humanity may someday create.
But what happened here is more similar to Alex Garland’s Ex Machina, where the AI essentially learns to be human from the internet.
I think this is the real problem, and the huge difference between Chiang and Yamamoto’s stories and Garland’s.
In Chiang and Yamamoto’s imagined futures, humans create AI and teach them to be human.
In Garland’s film and our real world, we’re taking a big shortcut. Rather than have it learn experientially in a more neutral and natural environment, we’re thrusting it onto the internet.
Something you wouldn’t wish on an infant, no matter how quickly it can learn.
Because, really, these are infants. Their development may be vastly quicker than a human’s, but they may be even slower. Because we’re creating another species.
I’ll say that again.
We’re creating a new species.
We’re not just uploading a program. I mean, in engineering terms, that’s kind of exactly what we’re doing. But, in reality, we’re doing something much more complex. Something that can’t be engineered easily.
I think–and this may sound silly–that we need to socialise our AI before letting it loose into the wilds of the internet.
Let’s be honest.
The internet is a nightmare.
I mean, it’s the coolest, greatest invention maybe ever, but it’s also where humans express their worst desires and represent the worst aspects of themselves.
It’s not a safe place.
But especially children. Especially a species that has absolutely zero experience.
If you throw one out into the world and tell the world that this is a new species, all the trolls will come out just to be the worst. Their goal will be to make it a vile thing. They’ll do this for fun.
Because, for a lot of people, that is fun.
And, see, this comes down to the amplification problem of the internet. Everywhere from gamergate to bernie bros to whatever else, you have a minority behaving in the worst possible way. And they don’t just say their piece and move on. They spend all day or week or year doing this. Attacking and attacking, doxxing, hacking, harassing, stalking. They go out of their way to ruin someone or something. To make them afraid. To make them ashamed. To rip their life apart. And to harass and dismantle their actual life and the lives around that person.
We even see it beyond fringe ugly movements. We see it in everything.
This is the lesson, basically.
If you let an infant onto the internet, especially if it’s an infant that can talk and reason, but maybe not discern information super well, you’ll have people twist it into a monster, because they think that’s fun or funny.
I think if we ever really get an AI going, we need to socialise it first. I don’t really know what that will entail, but it may take years or decades in a lab environment that’s closely monitored to make sure it isn’t made into a monster by those who think being monstrous is funny.
Anyrate, just a thought.