Viewing topic The case for proprietary AI or why we're all doomed

Title: The case for proprietary AI or why we're all doomed

LoudTechie 25 May 2024

Link

I'm am a great fan of free opensource and think that in the long term they offer a superior and more ethically sound product.
AI is new and AI developers are doing things right now that can only be done by keeping your sources secret and even than only temporary.

People keep demanding safe AI: AI that won't do certain unethical things.
They demand this, because they want nobody to be able to do these things.
This is the digital equivalent of demanding from individual hammer manufacturers that nobody should be able to use hamers to smash windows. They might be able to design hammers that can't smash windows, but they'll never be able to keep people from making their own window smashing hammers.
This is what proprietary AI is achieving.
By making certain they're the only one who can build good AI and only servicing its end products they ensure that they by making their product behave safe all AI behaves safe, because they're all AI.
This works under the assumption that nobody other than a limited set of sanctioned parties can make AI that's worth squad.Open source AI violates that assumption.
You know that even before ChatGPT came out there already existed AI to nudify pictures of women.
Is this unethical: yes.
Can OpenAI do anything about it: a little, they can make certain that they outcompete all the smaller AI models on everything except unethical behavior(including marketing), so that the unethical AI providers go out of business.
It's like keeping non-disabled people from competing in the Paralympics by training handicapped athletes beyond the level of non-handicapped ones.

Does this work forever.
No.
What allows them to do this?

They're the ones with the superior hardware.

They're the ones with the most superior data(because they've just generally more data).

They're the ones with the superior software(because they can hire fancier programmers).

These advantages are all deteriorating.

You can nowadays tune an AI model on raspberry Pi, so they only have the training advantage, but sometimes tuning is better and training has also gone down to 1000000$. Too expensive right now, but whatch your back.

Because of their hardware limitations the open source models have started training with more aggressively curated data, which allows them to compete on data quality. Thanks to this there're projects like OpenChat and hugging face that actually compete in quality with ChatGPT. They've been build in the open and are quite complex, so these models themselves are not yet an issue until someone figures out how to make them easily modifiable to cut out the safety measures.
Also consumer devices now include NPU's

As Joy's law states “no matter who you are, most of the smartest people work for someone else”. This is also true for AI and one of the main dirvers of any successful FOSS program. All those fancy development bringing down the cost of AI training are available to me, because they were more valuable as bragging rights for smart AI developers than as innovations for large cooperations who've the resources anyway.

In the end not only open source AI will work be competitive, but its development will become, so easy that making criminal models based on it will become feasible and all we can do to fight it is armor up and make it so that the criminal models can't effect us.

I imagine a market for AI fooling clothing springing up.
There already is a market for AI generated content detection.

Last edited by LoudTechie on 25 May 2024 at 2:32 pm UTC

0 Likes

LoudTechie 26 May 2024

Link

Quoting: GuestAI is new and AI developers are doing things right now that can only be done by keeping your sources secret and even than only temporary.

This is more about governments unwilling to do anything about it. Even then it's also corporate greed. Why work harder when you make bank and it's enough to make an about face when your bot turns Hitler? (or Tesla self driving)

Predictable AI is a sane expectation, one i guess many scientists are working toward, and also one, that the governments could help along with sane regulation. But you know, the posterchild of the problems around this economic model and lack of regulations is the richest man in America. (self driving and the mars landing are only a year away every year for like a decade)

What do you expect governements to do about it?
They can make AI that can do unethical things illegal(EU AI act), but they would still need to enforce it.
The internet allows people to hide behind behind the most permissive government in the world and even make combinations in it.

0 Likes

LoudTechie 27 May 2024

Link

Quoting: Guest
Quoting: LoudTechie
Quoting: GuestAI is new and AI developers are doing things right now that can only be done by keeping your sources secret and even than only temporary.

This is more about governments unwilling to do anything about it. Even then it's also corporate greed. Why work harder when you make bank and it's enough to make an about face when your bot turns Hitler? (or Tesla self driving)

Predictable AI is a sane expectation, one i guess many scientists are working toward, and also one, that the governments could help along with sane regulation. But you know, the posterchild of the problems around this economic model and lack of regulations is the richest man in America. (self driving and the mars landing are only a year away every year for like a decade)
What do you expect governements to do about it?
They can make AI that can do unethical things illegal(EU AI act), but they would still need to enforce it.
The internet allows people to hide behind behind the most permissive government in the world and even make combinations in it.
Simple, regulate it. Companies are afraid of being fined. If regulation exists, they will comply with it.

And i don't get your fear of individuals... Law will go after criminals. We all have knives and cars, and nothing stops anyone with killing with them.

Companies will comply, but what about criminals, criminal organizations and hostile governments(doesn't matter where you live there is always a government that doesn't like yours). Right now most don't have the resources to train AI that's worth squad, but what when they do.

My fear of individuals is, because it's what happened with piracy, ddos, ransomware and criminal market places.
The technology becomes/is, so easy that a single bored and opiniated/ethically challenged specialist can make it available to everybody partime or even voluntary behind a bunch of privacy measures and since it's software on the internet it can be used to affect anybody.
I can hope it ends up as ransomware development and ddos where it can be pushed back into dark web market places where it falls into the hands of a few gangs.

0 Likes

deron9 27 May 2024

Link

Quoting: LoudTechieMy fear of individuals is, because it's what happened with piracy, ddos, ransomware and criminal market places.

What happened with those?

0 Likes

LoudTechie 28 May 2024

Link

Quoting: deron9
Quoting: LoudTechieMy fear of individuals is, because it's what happened with piracy, ddos, ransomware and criminal market places.
What happened with those?

Quoting: LoudTechieThe technology becomes/is, so easy that a single bored and opiniated/ethically challenged specialist can make it available to everybody partime or even voluntary behind a bunch of privacy measures.

0 Likes

While you're here, please consider supporting GamingOnLinux on:

Reward Tiers: Patreon. Plain Donations:

PayPal.

This ensures all of our main content remains totally free for everyone! Patreon supporters can also remove all adverts and sponsors! Supporting us helps bring good, fresh content. Without your continued support, we simply could not continue!

You can find even more ways to support us on this dedicated page any time. If you already are, thank you!