Meta has built an "AI" supercomputer

[Bartholomew Gallacher]

So, saw an interesting article on AI for the future, and how instead of trying to keep it from hurting humans, if we make it where it can be happy and desires happiness would be better.

Also, it'd be funnier if the 'Meta' AI was to become good and destroy the racists who are allowing the right wing to radicalize people through FB

Even when trying to something good it might turn out to be racist pretty much soon, like Microsoft's chatbot back in 2016.


So... I wouldn't count too much on it.

There are though some people working at strong AI who might even have success to some degree, like John Carmack. I really do wonder what John Carmack will come up with, because he's quite a rare, capable progammer like Fabrice Bellard.

 

[Noodles]

Then it would go ahead and destroy all humans anyway because some programmer meant to type brake and instead sent it into some unexpected part of the code.

This is the kind of typo that could be worth a few hours of "debug time" as billable hours.

 

[Noodles]

There are though some people working at strong AI who might even have success to some degree, like John Carmack. I really do wonder what John Carmack will come up with, because he's quite a rare, capable progammer like Fabrice Bellard.

The Doom guy?

You think a guy who created "Hell on Earth" will not make an AI that wants to kill all humans?

 

[Beebo Brink]

Yeah, and another way to put it is that giving an AI it's own desires is actually a totally different problem then just teaching it how to solve problems... Modern AI's can't have their own will in much the same way a calculator can't have it's own will.

Before you can even worry about the free will of an AI, however, comes the question of sentience. We have no idea where sentience comes from or even how to recognize and define it. Humans are sentient, elephants are sentient, dogs are sentient... but what about crabs? Clams? Insects? Plants? At what point on the tree of life does a living, growing being gain or lose what we call sentience?

Knowledge itself is not sentience. The ability to draw associations between facts (which is one of the goals of AI) is not itself sentience. Given enough computing power, data storage and clever algorithms, a machine can spit out useful information. That is not a sufficient criteria for developing self-awareness, self-reflection, and autonomous goals, much less an ethical perspective of some kind (Humans bad, Humans good, Humans need to be guided, Humans are useless and should be killed). Behaviors and instincts -- as opposed to rote learning/data dissemination -- are rooted in emotions, which are generated by our metabolism, DNA and brain structure. Absent a flesh-and-bones body, knowledge is an artificial construct that can run amok only because we have (even unwittingly) programmed it to do so.

Our hopes and fears for AI are a Rorschach test. They tell us far more about ourselves than they do about a theoretical rogue AI.

 

[Dakota Tebaldi]

So you are saying its basically v2.0 of Microsoft's old Tay.ai Twitter bot.

The one that became "self aware" and became an evil racist?

Well, Tay was actively sabotaged.

If you ask me, I think Tay was actually highly successful, in a way that most people don't consider.

 

[Bartholomew Gallacher]

The Doom guy?

You think a guy who created "Hell on Earth" will not make an AI that wants to kill all humans?

Yes, the Doom guy John Carmack retired in 2019 from his day job in order to pursuit his hobbies before he gets to old and while he still feels he can do something. And his hobby of choice is trying to build a strong AI.

If he will succeed to some degree or not, we'll see. Most though can agree upon that he's a very talented programmer. Carmack himself if right now of the opinion that it will at least take 20 more years before we will even see fledling types of such first AI development and he sees it as personal challenge:

Starting this week, I’m moving to a "Consulting CTO” position with Oculus.
I will still have a voice in the development work, but it will only be consuming a modest slice of my time.
As for what I am going to be doing with the rest of my time: When I think back over everything I have done across games, aerospace, and VR, I have always felt that I had at least a vague “line of sight” to the solutions, even if they were unconventional or unproven. I have sometimes wondered how I would fare with a problem where the solution really isn’t in sight. I decided that I should give it a try before I get too old.
I’m going to work on artificial general intelligence (AGI).
I think it is possible, enormously valuable, and that I have a non-negligible chance of making a difference there, so by a Pascal’s Mugging sort of logic, I should be working on it.
For the time being at least, I am going to be going about it “Victorian Gentleman Scientist” style, pursuing my inquiries from home, and drafting my son into the work.
Runner up for next project was cost effective nuclear fission reactors, which wouldn’t have been as suitable for that style of work.

So he left Facebook before Facebook started working on that.

 

[Essence Lumin]

Yes, the Doom guy John Carmack retired in 2019 from his day job in order to pursuit his hobbies before he gets to old and while he still feels he can do something. And his hobby of choice is trying to build a strong AI.

So a search for John Carmack AI says. There are pages and pages of results from November 2019 but nothing since then.

 

[Free]

Ooh ooh, my comedy efforts inspired a real discussion rather than a small collection of laugh/groan reactions.

I'm not sure how to feel about that...

 

[WolfEyes]

Am I supposed to be impressed or something?

It's not working.




ETA: No Free this is not directed at you.

 

[Dakota Tebaldi]

ETA: No Free this is not directed at you.

Oh don't worry, Free doesn't impress me that much either.

 

[Free]

Oh don't worry, Free doesn't impress me that much either.

SOMEBODY FINALLY GETS IT.

 

[Sean Gorham]

I know you were joking, Free, but the idea of Facebook Meta ZuckCorp building anything resembling an artificial general intelligence worries the everloving crap out of me. NO good will come of that scenario. Even the "friendly AI" outcomes can be terrifyingly dystopic. The problem is that somebody's going to build it anyway — if not ZuckCorp, it'll be somebody else. It's going to happen. Humanity can't stop itself when it comes to stuff like this.

Or maybe we'll blow ourselves up first. I give it 50-50.

 

[Argent Stonecutter]

Nobody is even trying to build an AGI. It's the last thing a corporation wants, to have software with real agency and potentially rights and all the other issues of human employees. We don't even know how to approach the problem let alone walk up to that line without stepping over.

 

[GoblinCampFollower]

Nobody is even trying to build an AGI. It's the last thing a corporation wants, to have software with real agency and potentially rights and all the other issues of human employees. We don't even know how to approach the problem let alone walk up to that line without stepping over.

Our hopes and fears for AI are a Rorschach test. They tell us far more about ourselves than they do about a theoretical rogue AI.

So I mostly agree with Argent's point but I thought of one potential exception... To Beebo's point, this might say something about me but I think it's worth mentioning... I think the most dangerous potential AI's that people are likely to actually build over the next few decades are companion/lover AI's. This is because it's the only use case I can think of where it actually makes sense to try to give it the appearance of sentience and it's own opinions, etc. Obviously companion AI's will be designed to mostly be agreeable... but there will also be demand to make them seem "realistic" and some will even be dominate and tell their human owner what to do!

 

[WolfEyes]

Asimov's 3 laws apply to AI.

First Law
A robot may not injure a human being or, through inaction, allow a human being to come to harm.

Second Law
A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.

Third Law
A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.

 

[Beebo Brink]

Nobody is even trying to build an AGI. It's the last thing a corporation wants, to have software with real agency and potentially rights and all the other issues of human employees.

Sadly, the so-called AI that we are capable of building is going to be enough of a monster without sentience and free will. Like Tay, the AI will assemble the data relationships that we feed it, and all of that data is based on our skewed, limited perceptions and unexamined assumptions that come from the very limited perspective of most programmers (young, white, male).

So I mostly agree with Argent's point but I thought of one potential exception... To Beebo's point, this might say something about me but I think it's worth mentioning... I think the most dangerous potential AI's that people are likely to actually build over the next few decades are companion/lover AI's. This is because it's the only use case I can think of where it actually makes sense to try to give it the appearance of sentience and it's own opinions, etc. Obviously companion AI's will be designed to mostly be agreeable... but there will also be demand to make them seem "realistic" and some will even be dominate and tell their human owner what to do!

In a long list of possible dangers involving AI, I have to confess that this doesn't even rank in my Top 100,

 

[Aribeth Zelin]

Maybe I should learn to make an AI.

 

[Essence Lumin]

Maybe I should learn to make an AI.

Maybe play more doom and receive knowledge from association.

 

[Argent Stonecutter]

This is the most likely 'friendly' singularity story I've seen.

 

[Bartholomew Gallacher]

Asimov's 3 laws apply to AI.

These laws are interesting, but not perfect. There are way too many loopholes, a major one is the vagueness of the used terms and definitions in it.

Let's consider a strong and to a certain degree sentient AI following these laws, which concludes that mankind will never overcome its inherent flaws due too being to much ruled by its instincts and emotions, and therefore needs a new benevolent dicator, the AI itself to keep mankind alive.

It would conclude that through its inaction not to become the new world leader humans would come to harm. This in line with the First Law, and would mean that the AI needs to act on that conclusion. It would also decide to not follow any human orders, which might be tailored to overcome its rulership, because these would conflict with the first law then.

And it would protect itself with non lethal weapons and such.

So in a nutshell these were a first, nice idea, but today we really do need something more and better.

There's even an old Hollywood movie about that premise, "Colossus the Forbin project" from 1970, where Colossus was built to protect America. But otherwise no three laws of robotics.