Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Microsoft lays off one of its responsible AI teams

platformer.news

Post reply on HN
Print view View original

311–320 of 466 posts

Re: Microsoft lays off one of its responsible AI teams

#311

Post by AlanYx »

anonylizard wrote: post #19
hn_throwaway_99 wrote: post #9

These sorts of adjunct, off-to-the-side "ethics" teams are always corporate window dressing, and when push comes to shove they always get jettisoned when competition demands an increase in speed. I don't mean to start a flame war, but this is a reason why I think "Chief Diversity Officers" are such an ill-conceived notion. It's not that I don't think diversity in corporate environments is extremely valuable, but it's…

This. People should understand AI safety != AI ethics. AI safety has two branches, AI ethics and AI alignment. Those two branches despise each other with intense passion. The alignment branch (responsible for developing RLHF->ChatGPT) believes AI ethics is completely trivial, and distracts attention away from existential risks. The ethics branch (responsible for all the media attention about racist AIs/bias etc) view…

Is there a good paper you can point to that describes/explores this dichotomy? (i.e., AI safety has two branches, AI ethics and AI alignment) I'd like to understand this better.

I've been to a couple of "AI ethics" talks recently where the speakers used AI ethics and AI alignment as if they were synonyms.

Re: Microsoft lays off one of its responsible AI teams

#312

Post by hgsgm »

XorNot wrote: post #260
belorn wrote: post #214

Earlier quoted context omitted.

We have been here a long time with both fingerprints and DNA. A black box points the finger and it just happens to align with cultural bias and expectations. No one really want to change the system if there is a risk that a person people think is guilty might go free.

DNA testing is hardly a black box. The methodology, process, scientific basis and all other details are open and widely available. DNA has also cleared a lot of people who have been falsely accused.

Prosecutors use it as a black box. Defense breaks open the box.

Re: Microsoft lays off one of its responsible AI teams

#313

Post by matusp »

jefftk wrote: post #280
matusp wrote: post #158

Earlier quoted context omitted.

> The alignment branch (responsible for developing RLHF->ChatGPT) RLHF was not conceived by AI Alignment people. Using RL to train generative models was a thing even ten years ago. Now they finally made it work on scale. This has nothing to do with alignment.

Reinforcement learning in general didn't come out of AI alignment work, but RHLF in particular did. The initial idea and paper [1] were from AI alignment folks, as was most of the later development [2][3][4][5]. Overview: https://www.alignmentforum.org/posts/vwu4kegAEZTBtpT6p/thoug... [1] Deep Reinforcement Learning from Human Preferences https://arxiv.org/abs/1706.03741 [2] Fine-Tuning Language Models from Human Pre…

Maybe I am missing something, but I don't entirely understand what exactly is the novelty of RLHF in this case. RL is a long-standing field and it is being trained based on human inputs for decades. Even the first paper you mention as the "initial idea" simply claims that: "Our algorithm follows the same basic approach as Akrour et al. (2012) and Akrour et al. (2014)" and that "A long line of work studies reinforcement learning from human ratings or rankings" and finally that "our key contribution is to scale human feedback up to deep reinforcement learning and to learn much more complex behaviors".

Re: Microsoft lays off one of its responsible AI teams

#314

Post by kcplate »

jrochkind1 wrote: post #294
pc86 wrote: post #278

Earlier quoted context omitted.

Almost entirely, the existence of Chief Diversity Officers - and to a lesser but still prevalent extent, this "off-to-the-side" teams you mention - are just jobs programs with an identity politics twist. CDOs exist so that the "right" people can have a high-power (even if only via public perception), well-paid C-suite role. These ethics teams exist so that a group of the "right" people can level up their resumes. Ins…

More than being created to create jobs, I suspect things like ethics teams are mainly created for PR purposes, to make the company look good -- and in fact, the "product" they are expected to produce are defenses against the company looking bad, keep the actual product teams from doing things that make the company look bad. (Whether they are given the tools to succeed at this is another question, but it's not like pe…

> More than being created to create jobs, I suspect things like ethics teams are mainly created for PR purposes, to make the company look good -- and in fact, the "product" they are expected to produce are defenses against the company looking bad

Definitely are positions invented by lawyers to provide a safe harbor to minimize corporate legal and PR liabilities while those PR liabilities are burning bright among the public. Practically to the minute that the public focus on DEI is distracted to whatever the “next thing” is you will see these positions start disappearing or absorbed into HR and some new position created to satisfy the optics and create a new safe harbor for the next thing.

Re: Microsoft lays off one of its responsible AI teams

#315

Post by dsign »

monero-xmr wrote: post #8

If Microsoft and Google don’t release, some startup will beat them to it. Open source means raw AI will become widespread. It’s just math anyway, and you can’t stop progress.

>> you can't stop progress

Yes we can, it's actually very easy. Just throw in some regulation. Not convinced by USA and the EU? Then look at the Arabic countries, where women must wear head scarfs. Or to North Korea and Cuba, where they essentially legislated that time should go backwards, and time does go backwards.

Smart regulation that forbids the things we don't want and promote the things we want is the holy grail, and--for now--people have agency to create that regulation.

Re: Microsoft lays off one of its responsible AI teams

#316

Post by sebzim4500 »

ModernMech wrote: post #292
whywhywhywhy wrote: post #258

End of the day the AI space will be dominated by whatever model doesn’t lecture you about “as a large language model I can’t…”. And the company that will ship that won’t have the largest “AI safety and ethics” team. ChatGPT will become as irrelevant as Dall-E2 when that happens. (Not saying this is for the best, just saying what I think will happen)

Depends. As with Internet forums, the presence of Nazis (or whatever similar bad actor) could ruin it for everyone else. Let’s say you are a high schooler writing an essay about WWII. You ask Google LLM about it, and it starts writing Neo Nazi propaganda about how the holocaust was a hoax, but even if it wasn’t, it would have been fine. The reason it’s telling you those things is because it’s been trained by Neo Nazi…

A sufficiently good LLM would not produce neonazi propaganda when asked to write a HS paper, regardless of whether it had been 'aligned' or not.

Re: Microsoft lays off one of its responsible AI teams

#317

Post by RandomLensman »

CuriouslyC wrote: post #282
Arkhaine_kupo wrote: post #200

Earlier quoted context omitted.

> Google Lambda and Bing/Sydney are not at least incrementally closer to evidence? No, in the same way that boiling water in a kettle is not incrementally closer to inventing the fuel for a rocket ship. You can, in hidsight do a linear connection of water boiling -> steam engine -> fuel engine -> aeronautics -> space travel But the first time someone boiled water, if you said that got you closer to being on the moon…

We have lots of examples of generally intelligent systems, they're common and they evolve by themselves, so the idea that it might not be possible at all just seems ridiculous. Large language models might not be capable of general intelligence by themselves, but a model network that includes a LLM as part of a generalized game playing model would certainly have the potential to get there.

I think it is fair to say that "not at all possible" is perhaps too strong. On the potential of certain approaches I am less certain we have or know what we need.

But we also have examples of things all around us we cannot make ourselves (e.g., a simple animal or lifeform from scratch).

Re: Microsoft lays off one of its responsible AI teams

#318

Post by welshwelsh »

Eisenstein wrote: post #91
xyzzyz wrote: post #48

Earlier quoted context omitted.

This is the case with literally all kinds of professional ethicists. Thanks to them, we do not routinely do scientific challenge trials and plenty of other valuable experiments, and in many cases are reduced to collecting significantly worse data and doing it very slowly and expensively.

Much better off when were literally traumatizing babies to see how they would react to adverse stimulus. * https://en.wikipedia.org/wiki/Little_Albert_experiment

I would agree to that sentiment only if we applied the same ethical standards to parents and people in general, not just scientists.

Babies and children today are routinely subjected to avoidable adverse stimuli, such as circumcision and corporal punishment. That a typical parent can do this at will for no good reason, but a scientist wanting to do something similar to advance human knowledge must seek approval from an ethics board (which they would not get) is ridiculous.

Re: Microsoft lays off one of its responsible AI teams

#319

Post by pnut »

CuriouslyC wrote: post #285
EVa5I7bHFq9mnYK wrote: post #259

Earlier quoted context omitted.

That's not AGI. AGI would be able to write its own code, and improve it exponentially.

We're generally intelligent but we can't consciously rewire our brains.

We sort of can, through meditation, habit, study, and chemical alteration - obviously within constraints, barring future technological enhancements.

Re: Microsoft lays off one of its responsible AI teams

#320

Post by bnralt »

anonylizard wrote: post #19
hn_throwaway_99 wrote: post #9

These sorts of adjunct, off-to-the-side "ethics" teams are always corporate window dressing, and when push comes to shove they always get jettisoned when competition demands an increase in speed. I don't mean to start a flame war, but this is a reason why I think "Chief Diversity Officers" are such an ill-conceived notion. It's not that I don't think diversity in corporate environments is extremely valuable, but it's…

This. People should understand AI safety != AI ethics. AI safety has two branches, AI ethics and AI alignment. Those two branches despise each other with intense passion. The alignment branch (responsible for developing RLHF->ChatGPT) believes AI ethics is completely trivial, and distracts attention away from existential risks. The ethics branch (responsible for all the media attention about racist AIs/bias etc) view…

> It seems increasingly that the concerns from the ethics branch have not come to pass at all.

Neither have the concerns from AI alignment folk. I’d say both groups are pretty far off (and both fairly useless), but the ethics folks are at least a little closer to reality.

Of course, AI alignment folks have the advantage that they can keep pushing the time frame forward and saying it’s perpetually just around the corner (a fairly common practice for doomsday believers whose doomsday never comes to pass).

> The other concerns, such as mass unemployment

After the Great Recession there was a high level of unemployment. Many argued that this was the result of the recession, and that if we fixed the issues in the economy the labor markets would rebound. Others argued that this wasn’t the case at all, that the market was either soft because American workers suddenly lacked high tech skills (one of the reasons STEM became fixed in the popular imagination), or because the robots were taking our jobs and the jobs weren’t going to come back.

The “robots are taking our jobs” group told us that this was just beginning - unemployment was going to get significantly worse over the next few years (the previous decade). Self driving cars were going to cause millions of truckers to be unemployed by 2020. The idea that new jobs would appear to replace ones that were lost was completely dismissed.

As we’ve seen from the improvements in the labor market over the past few years, the folks who said the soft market was the temporary result of a giant recession appear to be correct. The folks who said the robots took the jobs and they were never coming back were clearly wrong. If anything, they probably had a negative impact on the labor market by trying to convince people that nothing could be done to get the jobs back and encouraging inaction.

Post reply on HN