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

331–340 of 466 posts

Re: Microsoft lays off one of its responsible AI teams

#331

Post by paulryanrogers »

IMTDb wrote: post #221
kramerger wrote: post #205

Earlier quoted context omitted.

I think there is a lot of unjustified hatred for the AI ethics guys. From what I have seen, they have raised manu important issues so far, and led to many improvements both around ethics and AI in genetal. For example, the cops wanted to arrest people based on input from image recognition systems. Someone demonstrated that the leading system flagged multiple members of Congress as dangerous fugitives and that was the…

> Someone demonstrated that the leading system flagged multiple members of Congress as dangerous fugitives and that was the end of that. That's not a question of ethics, it's a question of the model performance.

If the model performed so poorly it would be an ethical concern to deploy it, as many departments were so eager to do.

Re: Microsoft lays off one of its responsible AI teams

#332

Post by paulryanrogers »

sebzim4500 wrote: post #316
ModernMech wrote: post #292

Earlier quoted context omitted.

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.

Defining good may require ethics.

Re: Microsoft lays off one of its responsible AI teams

#333

Post by cumshitpiss »

IMTDb wrote: post #221
kramerger wrote: post #205

Earlier quoted context omitted.

I think there is a lot of unjustified hatred for the AI ethics guys. From what I have seen, they have raised manu important issues so far, and led to many improvements both around ethics and AI in genetal. For example, the cops wanted to arrest people based on input from image recognition systems. Someone demonstrated that the leading system flagged multiple members of Congress as dangerous fugitives and that was the…

> Someone demonstrated that the leading system flagged multiple members of Congress as dangerous fugitives and that was the end of that. That's not a question of ethics, it's a question of the model performance.

[dead]

Re: Microsoft lays off one of its responsible AI teams

#334

Post by fnordpiglet »

Xcelerate wrote: post #252
bmc7505 wrote: post #111

Earlier quoted context omitted.

Those two branches despise each other with intense passion. I never actually made this connection before, but having spent far too much time listening to both camps, this observation does ring true. Another observation I had is that there is an orthogonal axis to the safety/ethics dichotomy that follows a more scholastic approach, namely neuro/symbolism. These two branches correspond, roughly, to the debate between m…

> The symbolists place a heavier emphasis on model interpretability and believe learning emerges from a logical substrate, albeit at a much higher level of abstraction. These people still exist? Personally, I don’t think gradient descent is the way to AGI either (I think it’s efficient algorithmic search over the space of all computable programs), but I haven’t heard anything from the symbols crowd since the early 20…

They can’t speak over all the cooling noise coming from the GPU farm.

Re: Microsoft lays off one of its responsible AI teams

#335

Post by dgb23 »

xyzzyz wrote: post #116
Eisenstein wrote: post #91

Earlier quoted context omitted.

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

Ah yes, because there is no reasonable middle ground between forbidding experiments where a subject might get a mote in their eye, and literally sacrificing babies to Moloch. Please. You can make better argument than that.

The irony with statements like these is that they are themselves ad-hoc ethical statements.

Whatever "reasonable middle ground" you find, you'd have to make logically sound ethical arguments. This is what these people do for a living, and they are very well trained in those matters.

As programmers and engineers we are also trained in logic, use it every day. But that typically pales in comparison of what philosophers do. They deal with much richer logical systems and can precisely apply them to statements and arguments.

When applied to ethics, they typically end up with much more radical conclusions that we are dealing with here. In fact the resulting initiatives are already softened compromises and a "reasonable middle ground" before it even clashes with policy making.

Re: Microsoft lays off one of its responsible AI teams

#336

Post by fnordpiglet »

bmc7505 wrote: post #111
anonylizard wrote: post #19

Earlier quoted context omitted.

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…

Those two branches despise each other with intense passion. I never actually made this connection before, but having spent far too much time listening to both camps, this observation does ring true. Another observation I had is that there is an orthogonal axis to the safety/ethics dichotomy that follows a more scholastic approach, namely neuro/symbolism. These two branches correspond, roughly, to the debate between m…

The issues is the safety folks keep making the trolly logic run over the ethics folks, and the ethics folks keep pointing out an ethical safety AI person would prioritize sacrificing themselves.

Re: Microsoft lays off one of its responsible AI teams

#337

Post by xpe »

sebzim4500 wrote: post #316
ModernMech wrote: post #292

Earlier quoted context omitted.

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.

Why not? Can you unpack what you mean?

Are you saying that sufficiently good models should understand that such propaganda is not appropriate for the context?

Are you saying that understanding appropriateness is not the same thing as ethics?

Re: Microsoft lays off one of its responsible AI teams

#338

Post by xpe »

paulryanrogers wrote: post #332
sebzim4500 wrote: post #316

Earlier quoted context omitted.

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.

Defining good may require ethics.

May? Isn’t this the definition of good?

Re: Microsoft lays off one of its responsible AI teams

#339

Post by thwayunion »

What a strange comment section.

A customer service chatbot telling your customers that an issue is solved, when the issue isn't solved, is a core product failure. It's also a Responsible AI issue.

A code generation model suggesting GPL code without attribution is a core product failure and serious legal risk [1]. It's also a Responsible AI issue.

A diffusion model generating child porn or nazi propaganda is a core product failure and a serious legal risk [2]. It's also a Responsible AI issue.

In other words, the problem isn't that Responsible AI is BS. The problem is that Responsible AI is a conglomeration of serious legal and product risks that companies NEED to solve in the context of specific products.

More importantly, how will those risks change over time? The western world public is overwhelmingly skeptical of AI and in favor of regulation. Even if new government regulation can be gummed up, you can expect juries to be less than sympathetic.

But you’re not going to solve those problems by hiring a bunch of philosophers and interdisciplinary types that obsess over “nutrition labels for models” and the like. Everyone knows that Tay being a nazi is bad. And I don’t need a PhD in Sociology or whatever to tell you that you probably shouldn’t be openly condescending toward your customers. But constraining and aligning the output of models is a really hard technical problem, and needs a different type of expertise.

So, I think we’ll see Responsible AI teams start to look a lot like e.g. safety at automotive companies, verification & validation at hardware companies, or security at tech companies. Mostly staffed with technical people who are strong engineers and happen to specialize in data governance/constrained generation/rotics safety/etc. There will be dedicated teams that kinda-help-kinda-police the rest of the company (think security). But there will also be a lot of diffused responsibility within each product group, with ICs or small groups within each group owning this space.

[1] https://githubcopilotlitigation.com/

[2] https://en.wikipedia.org/wiki/Strafgesetzbuch_section_86a

Re: Microsoft lays off one of its responsible AI teams

#340

Post by bioemerl »

xpe wrote: post #323
bioemerl wrote: post #264

Earlier quoted context omitted.

I believe it's for the best. The dangers of AI are theoretical. The dangers of corporations being in control aren't.

> The dangers of AI are theoretical. What do you mean by this? Many people throw out the word ‘theoretical’ as some kind of way to imply that something ‘isn’t real’ or worth worrying about. Something might seem implausible until it happens. Gravity waves were once ‘only’ theoretical after all :) There are plenty of AI dystopian predictions, and many of these are possible and impactful. Many of these theories are base…

I'm saying that most predictions about the dystopian AI future are typical of any generation encountering a disruptive technology. They're exaggerated and built from assumptions.

They also always seem to conclude that the answer to these problems is to keep these machines centralized in the hands of big companies and governments, which is a very strange conclusion when they tend to get paychecks from those groups.

At the extreme end you even get these people talking about AI like it's going to be some monstrous world eating god.

All innovation has negative consequence.

Television lead to couch potato. 24 7 news. Television stars.

Radio killed the local performer.

Social media addiction and the erasure of culture and increase of depression.

Cars, if you go back far enough, got plenty of criticisms as well.

The AI powered future, should it come to exist will have downsides. You'll have a very different landscape to deal with as a creative where your artistic output isn't valued for individual works of art. You'll have very different expectations of how art and media is personalized. It will be hard to trust video or pictures or audio again, because they are so easily faked.

Our way of life, as it stands today, will not survive. That's not bad, just different.

The AI critical people tend to have some valid criticisms, but at the end of the day their criticisms are rooted in a desire to maintain a status quo until we are "sure it's safe" and that just doesn't fly.

We don't know how this will all pan out until we try it, so as we have always done, we will try it and deal with the consequences as they appear.

Post reply on HN