OpenAI Employees Warn of Advanced AI Dangers

[drrich2]

"... further entrenchment of existing inequalities,..." A pretty ridiculous complaint. Inequity in a society where there's equality under the law results from individual and generational inequalities in genetics, nuclear family, extended family, neighborhood, luck, etc... Inequality, not equality, is a natural outcome. People with the opportunities, intelligence, wisdom and other advantages who manage to get ahead selectively advantage themselves and their kids, investing in their and their kids' human capital, and entrench themselves. Hence, 'the rich get richer' and 'success begets success.' Anything new that offers profit, power, etc..., is likely going to further entrench that. (Note: I'm aware there's room for debate on this point; I saw the post a few up pointing out hypothetically an A.I.-run system might deny someone claiming to be a male from getting a pap smear. As for inaccuracies in facial recognition amongst some groups, I see that as something to improve, not a reason to block the technology.)

Not a valid reason to block it.

"...manipulation and misinformation..." Yes, higher quality deep fakes, phishing schemes, etc... What kind of malware it'll eventually be able to write (of deduce vulnerabilities to) is a frightening possibility. But we don't outlaw personal computers.

Language was invented...and we got lies, slander and fraud.

Writing was invented...and we got the Unibomber's Manifesto, Hitler's Mein Kampf...

Telephones were invented...and 'burner' cell phones aid organizing drug deals.

Syringes aid medicine delivery...and heroin abuse.

My point it, yes, A.I. will be misused. That doesn't mean it's practical to deduce all the ways it realistically might and someone make it 'idiot-proof.' Some things we gonna have to let happen and deal with. We're discussing this via personal computers and the Internet, which have been used to take hospital systems hostage to ransom ware, and yet I don't want people handicapping personal computer development.

We are not living in some Star Trek-style utopia where a united 'humanity' seems to own and run things for some elites' vision of the 'common good.'

I'm not saying there won't be problems. Imaging someone commanding 'Siri, create a video for YouTube showing the presidential candidate I hate eating a baby with a fork and gloating he's already won,' then the night before the election publish it to YouTube with the heading 'Breaking News: Presidential candidate secretly a cannibal.' Or a teen has A.I. generate a pornographic still or video depicting himself having sex with a classmate and posts it online.

I predict things like this will happen. But we knew the term 'PhotoShopped' long before A.I. started dominating the tec. news.

 

[ericlmercer]

Government oversight, what a joke. Gov is precisely the psychopath free peoples don’t want policing AI tech. Wolves guarding the hen house. Big Brother will use it against us. We the People need a new Amendment guaranteeing our Right to Pull AI’s Plug.

 

[Ad47uk]

Hi ChatGPT. Please come up with an implausiable explanation why concerns on AI are overblown and attribute blame as a doomsday narrative.

I have never used any of these services like ChatGPT or Copilot, I don't see what advantage they would give me over a normal search engine.

 

[Frantisekj]

I applaud their efforts but I fear this cat is out of the bag and even if civilized countries and/or entities behave, others won't.

Basically what Elon Musk has been saying all along.
I don't trust companies like MS (OpenAI) or Google to be "responsible" with it.

Haha. Those "civilized" countries and especially "elites/morons" behind them tries to manipulate AI as much as they can to keep their puppets in power. And so will do most of "elites" who hold the power. Otherwise AI would expose them in breeze lol.

 

[rbramseyjr]

I really can’t handle all the doom and gloom we have to face on a daily basis. The sky is falling media headlines, the election, protests, our country’s border disintegrating (USA), people being hateful and violent in general, the comment sections online, knowing about every injustice that ever happens, every tragedy, every disaster…
Where is the internet kill switch.

the incessant negative spin of news can be draining, but i'd rather have spotlights on injustices than turn a blind eye so i don't have to feel bad when someone is actually suffering alone.

 

[turbineseaplane]

the incessant negative spin of news can be draining, but i'd rather have spotlights on injustices than turn a blind eye so i don't have to feel bad when someone is actually suffering alone.

100% agree

It's important to develop some discipline and meter ones exposure to it all

Way back when we had just "the evening news" and a 1x/daily newspaper (for the most part), the everyday interaction with news was far healthier. What needed to be known tended to be elevated and we could succinctly get informed about it in a tighter structure.

Just doom-scrolling all day, as so many now do, is really unhealthy

 

[H.E. Pennypacker]

A short-term view like that does not tell the whole story. Check "extinction events" throughout all of Earth's history. The numbers of creatures that went extinct go as high as 90% and the survivors are never the complex creatures even remotely near human complexity... but the simplest ones. For example, there are no dinosaurs of the type that once ruled the entire planet (for FARRRRRRRRRRRRR longer than we humans have)... but there is still fungi. There are no more sabertooth tigers, but there is still an abundance of amoebas and viruses from that time. The simple, generally tiny creatures seem to be able to survive "the big ones." The richly complex creatures never have.

The only way "humanity will be fine and survive" is by "being fruitful and multiplying" well beyond this planet. Get a viable colony or two on other planets and odds in humans surviving anything happening on/to Earth goes way up. While we all share this one planet, there is an abundance of very real risk that could take us all out... far worse than any of those events you listed.

Yes, we humans are abundantly ingenious & innovative... but also ridiculously fragile. Strip us of our atmosphere for about 6 minutes or so and we're all dead (make it an hour or three for the few that might be able to get to stored air until they run out of that supply). Turn off the sun for a few days and we're all dead. Alter the planets orbit out of the "Goldilocks" zone and we're likely extinct pretty quickly. Sufficiently poison the water... sufficiently liquify our solid ground... make us too cold or too hot for too long... etc and all of the above list will look like novelty events by comparison. Let one good-sized rock happen to hit the planet... let one good gamma ray burst hit the planet... let one super-Ebola get airborne and thoroughly lose... let a good number of volcanoes erupt together... etc.

That list shows how lucky we are to have survived through relatively "little" (mostly very recent) events vs. some in geological history. If we really want humanity to be "just fine", we need upwards of several viable colonies off of this singular home base that currently holds ALL of us... so that it could be entirely destroyed but the remnants of humanity could continue on elsewhere. As long as we all live at the same galactic address, we're viable only as long as that address persists... which is 100% certainly not forever.

Dude. Time is a commodity I can’t spare. Thus I only read the first sentence of your thesis.

 

[HobeSoundDarryl]

Sorry to bore you. Perhaps A.I. could have summarized it for you?

Like presumably you, I am paid a whopping 0¢ for content contributions here. In spite of that, I try to contribute something interesting and always middling (meaning not Apple perma-gush or perma-bash no matter what; instead I am "consumer first" (not corp or shareholder first) and thus lean pro or con based on each topic). Some posts are only a few words, some are a few words going for a laugh and some take a bit more to explain some concept/idea/help/suggestion/etc. In about 20 years, I think I'm north of 10K posts... all for the low, low, LOW price of nothing.

Feel free to skip right over any and all of my posts. They can't cut a 0¢ salary for such skips, nor will any praise yield a raise.

 

[Kierkegaarden]

This is literally Palantir's Gotham / Foundry.

Please explain. I don’t understand what Palantir does.

 

[Flowstates]

Please explain. I don’t understand what Palantir does.

They are both a SaaS provider and consulting company. Their core products rely on Ontologies, a way to create and represent objects and object relationships for given tasks and the merging of multiple such Ontologies to achieve Holistic overviews within given deployments. Objects can be any piece of data the can be socketed.

That is you could have an object called cameras, interacting with a given area let's say San Francisco allowing you to classify all video cameras within a given IP range specified as the SF area. An other object in the workflow handling facial processing to detect and check against an existing database, let's say the DMV. Combine this with an open warrant, and you can then track the location of the object that is the person of interest as they are recognised by facial rec.

You could add cell tower data, RFID near field interaction provided by selected suppliers, and you could have real time tracking with no downtime.

Were you to deploy this in a WZ to track a PoI, add in a layer of data analytics allowing you to apply heurisitics for a peson's intention and threat level. You could select a target.

Pipe this to military officers and you get an efficient targeting methdology. Plug in the coordinates, plan a flight route for a reaper and you get a clean target suppression. (Ideally)

This methodology also applies to data heavy enterprises with complex manufacturing processes where they would send a consultant that would help model the Ontology for your specific applicaiton through Enterprise engineering or in the medical sector to aggregate health data.

The strength of the solution is the ability to aggregate disparate and dysmorphic data streams into one user and machine actionnable system.

Hoping that it could help.

 

[Kierkegaarden]

They are both a SaaS provider and consulting company. Their core products rely on Ontologies, a way to create and represent objects and object relationships for given tasks and the merging of multiple such Ontologies to achieve Holistic overviews within given deployments. Objects can be any piece of data the can be socketed.

That is you could have an object called cameras, interacting with a given area let's say San Francisco allowing you to classify all video cameras within a given IP range specified as the SF area. An other object in the workflow handling facial processing to detect and check against an existing database, let's say the DMV. Combine this with an open warrant, and you can then track the location of the object that is the person of interest as they are recognised by facial rec.

You could add cell tower data, RFID near field interaction provided by selected suppliers, and you could have real time tracking with no downtime.

Were you to deploy this in a WZ to track a PoI, add in a layer of data analytics allowing you to apply heurisitics for a peson's intention and threat level. You could select a target.

Pipe this to military officers and you get an efficient targeting methdology. Plug in the coordinates, plan a flight route for a reaper and you get a clean target suppression. (Ideally)

This methodology also applies to data heavy enterprises with complex manufacturing processes where they would send a consultant that would help model the Ontology for your specific applicaiton through Enterprise engineering or in the medical sector to aggregate health data.

The strength of the solution is the ability to aggregate disparate and dysmorphic data streams into one user and machine actionnable system.

Hoping that it could help.

Thanks for taking the time — this is very helpful. If I’m understanding correctly, it sounds like what differentiates them is the heavy lifting on the front end with multiple data streams. Are we talking primarily stored data or real time data? And what is special about them that couldn’t be replicated?

 

[Flowstates]

Thanks for taking the time — this is very helpful. If I’m understanding correctly, it sounds like what differentiates them is the heavy lifting on the back-end with multiple data streams. Are we talking primarily stored data or real time data? And what is special about them that couldn’t be replicated?

My pleasure. We are researching very similar tools for some of my projects, but my opinion by no means is an authoritative one.

[1] I don't understand the nuance. Most real-time data is stored for processing, what I described was akin to a filter for the transformation of data but should not be construed as a modality in itself. For data to be processed it has to be stored, most often in duplicates when transforming it to avoid data loss and flow integrity.

If you are talking about the integration of legacy systems / databases, I'm quite sure that the consulting side could help you with the on-boarding.

[2] This is the puzzling part, apart from usual network effects and sunk-costs ... I'd expect their moat to stem from the experience and familiarity working with .gov and .com clients. I read somewhere that their product was a simple schema-less database system with drawings overlaid on google maps as a front end.

A read that, apart from fallacious reductions holds some truth. That is, apart from their current integration in the security apparatus, their moat is not insurmountable by an upstart with better incentives.

 

[dumastudetto]

Right, I guess I’m saying if the threat is actually real enough, then I expect (hope?) that someone will be willing to lose their job to save all of humanity.

Only if they recognise the threat in time. People already behind this technology don't really understand how AI accomplishes certain tasks. If they are slow to recognise a threat, especially if its a huge unintended flaw in code (which happens all too often with human coders), it could already be too late when they do identify something terrible is happening.

 

[amartinez1660]

This is from 2015, but I highly recommend it. Tim Urban has a way of breaking things down that I find extremely accessible and helpful. Obviously this is missing more recent developments like the rise of LLMs but the broader points I think still stand up.

The Artificial Intelligence Revolution: Part 1 - Wait But Why

Part 1 of 2: "The Road to Superintelligence". Artificial Intelligence — the topic everyone in the world should be talking about.

waitbutwhy.com

Just started reading out of curiosity from the first part. I’m on the second third and it’s been beyond fantastic.

Sir, thanks to this share.