Live data from Hacker News

AI at Google: our principles

blog.google

411–418 of 418 posts

Re: AI at Google: our principles

#412

Earlier quoted context omitted.

All the arguments have already been used in the past threads. No use to repeat them here.

As a neutral observer, I've not been on past threads. Most people who don't have particular interest in this haven't. It would be nice to hear both sides of the argument

Go back to the threads on github acquisition. There are at least 4 of them during the past week. They are very long, very rich and very divided, so making a tl;dr would be too much work.

Re: AI at Google: our principles

#413
post #4

The best way to lead is by example. Thank you, Googlers. The choice not to accept business is a hard one. I've recently turned away from precision-metrology work where I couldn't be certain of its intent; in every other way, it was precisely the sort of work I'd like to do, and the compensation was likely to be good. These stated principles are very much in line with those that I've chosen; a technology's primary pur…

> ”non-surveillance purposes”

Not what it says. It says:

“...surveillance violating internationally accepted norms.”

Thanks to leaks, we have a glimpse of the new normal.

Re: AI at Google: our principles

#414

Earlier quoted context omitted.

Really? Looking at drone strike causualities you’d have no clue how much the US even tries. Its funny cause I haven’t heard of large scale Chinese or Russian drones flying over other countries targeting terrorists but ending up murdering children on more than one occasion. Perhaps these technologies can be just evil and the US is the only country powerful enough to get away with using them. I really have no stomach f…

If you ever end up in a war with China, you'll be seeing plenty of drones. Obviously China doesn't use drones for killing terrorists - China doesn't kill terrorists because terrorists have no way to get into China in the first place.

They have plenty of terrorists and frequent attack, but also no free press so you do not get to know much about it...

Re: AI at Google: our principles

#415
post #55

So, I'm all for giving someone the benefit of the doubt if they have a change of heart upon reconsidering an issue, but this coming after the fact rings a bit hollow to me. I think the only principle at play here is that it became a PR issue. That's fine, but let's be honest about it. Early emails between Google execs framed this project only in terms of revenue and potential PR backlash. As far as we're aware, there…

Google’s disregard for the ethics of the arrangement is further supported by this interview with one of the Googlers who organized the resistance to this project: https://jacobinmag.com/2018/06/google-project-maven-military...

Jeez, that's just brutal. I try to avoid content-free comments like this one but... damn. Just brutal.

If things actually went down like this person describes I would have been out the door too. (Due to the betrayal. I actually think Project JEDI is pretty cool.)

Re: AI at Google: our principles

#416

Earlier quoted context omitted.

It doesn't have anything to do with censoring your favorite conservative search result. But that's not correct. That's exactly what it's to do with. A big part of the philosophy of conservatism is to accept the world as you find it. Conservatives, at least in theory (not saying the Republicans are a great implementation of the philosophy), eschew large social engineering schemes, they eschew attempts to remould attit…

Obtaining the correct categorization of something is not social engineering, it's returning a FACTUAL answer. I didn't ask the AI "are men more likely to be doctors" I asked "what is a doctor" Do you understand the difference? I don't expect it to tell me "Doctors are male individuals who...." No where in my original post did I say the AI has to lie about gender ratios, I would expect any AI to tell me also, the soci…

I didn't ask the AI "are men more likely to be doctors" I asked "what is a doctor"

I'm afraid you're obfuscating. The most important paper in question is this one:

http://papers.nips.cc/paper/6228-man-is-to-computer-programm...

It documents the researcher's "discovery" that word embeddings trained on all available text learn that most doctors are male and most nurses are female, along with many other relationships, like volleyball / football being a female / male analogy.

Word embeddings can't answer a question like "what is a doctor", even people would struggle to give a good answer to such a vague question. So they asked it specific questions about gender, namely, "if you ask for a gender relationship starting from doctor what do you get" (answer: nurse). And then they decided this was bias, and wrong, and should be edited out of the model.

So yes - if you asked such a model "are men more likely to be doctors" it would answer "no" although the correct answer is "yes".

I wonder if that changes your views?

Right, progressives are the only ones with ideology, conservative positions are arrived at by cold, hard logic?

Both positions arrive at their conclusions via logic, but working forward from different assumptions and premises.

The difference is progressives are much more likely to try and impose change on the world top down, whereas conservatives are much more likely to leave the world be (note: attempts to remove top-down imposition of progressive views is often cast by progressives as equally equivalent "change", but it's not, likewise, not attempting to change old traditions is often cast as inherent support for them rather than a general aversion to imposing top-down change).

Your post is a great example of the dangers of this attitude:

Maybe just maybe, the status quo isn't good for women?

I happen to think the status quo, where women are routinely handed jobs and money simply for being women (e.g. the whole 'women in tech' movement), is excellent for women!

But our views on this are both irrelevant because it's not what the argument is about.

The point is, do we want to build AIs that understand the world as it is today or which have been given an understanding of the world as Googlers believe it should be.

You are very clearly arguing for the latter, here, to the extent that earlier in your post, you argued a good answer to the question "what is a doctor" would actually involve a lecture on the "socio-historical-cultural perspective". Who says users of AI give a crap about any of that? Maybe they just want to get a definition of the word doctor, without some AI trying to change their kids ideas about what job they'd like to do along the way?

A Google AI that is constantly engaging in proxy social advocacy with me would be annoying as hell. A Google AI that doesn't but has a warped and distorted view of reality because its creators feel they're towering moral actors with a mandate to change the world on behalf of billions of people they never met? Even worse! I'd rather it was at least open about it.

Look at it this way. If Google AI was constantly subtly suggesting that all progressives were naive, hated America, that government intervention always failed and markets were the best way to do things, , you wouldn't be very happy about it, would you? Especially not if the AI hadn't actually learned such things but such beliefs had been added later by libertarian programmers convinced that they'd make the world a better place by doing so.

We've had anti-sodomy laws, anti-booze laws, anti-miscegenation laws, all of them enacted by social conservatism

I suspect there's a slight terminological difference here between libertarianism and conservatism (I've been meaning primarily the former).

But regardless, are you sure you aren't assuming that?

Let's take anti-booze laws. Prohibition in the USA was a bipartisan issue at the time and both Democrats and Republicans voted in favour of it, in fact more Democrats voted in favour than Republicans did.

The constitutional amendment that gave women the right to vote was introduced by a Republican. It took a long time to get passed (decades), but this was partly because - just like the UK - the woman's suffrage movement hurt its popularity through aggressive tactics and association with unrelated social policies, in particular with pacifism and refusal to join World War 1.

And as for religion, I'd note that the USA is probably the most conservative (or libertarian) country in the world and it's also famous for separation of church and state.

Re: AI at Google: our principles

#417
post #381

Earlier quoted context omitted.

India (part of the British Empire) also sent millions who died fighting in North Africa. The reasons every country got into the war were complex. But it is also largely true that America spilled blood and treasure out of a sense of obligation to fight Nazism, despite having a significant immigrant German population (who fought against their brethren on the other side), going so far as to impose a draft in the later s…

This is so much BS. You should actually read the history of the reasons for the involvement of the US in the WW2 theaters. You sound like a young kid who has never actually read about history and geopolitics. Please dude, actually read up on the involvement of the US in the secret wars in Cambodia by Kissinger, the School of the Americas, instrumental in teaching South American militaries repressive strategies that k…

> Please dude, actually read up on the involvement of the US in the secret wars in Cambodia by Kissinger, the School of the Americas, instrumental in teaching South American militaries repressive strategies that killed tens of thousands of innocents with the full support of the American government, the Iran-Contra affair, the ridiculous involvement in Vietnam.

Kissinger was just pure evil. Apart from his involvement in the things above, he also actively supported dictators in Pakistan, and indirectly did nothing to stop the slaughter of civilians in (then) East Pakistan.

But, I still stand by my point: Americans have strived to right the wrongs (and there have been many and monumental ones). Eventually all these 'secret' activities have come out and the public has ensured that the people responsible were shamed or held accountable, to some extent (it never is a full reckoning, unfortunately). I can't think of that happening in China, for instance. Name any other great power that hasn't had stuff like this?

Re: AI at Google: our principles

#418
post #24

Earlier quoted context omitted.

I concur. This post seems to be specifically targeting the negative perception the resignations were generating.

Is there any room in your cynicism for any other view? For example, maybe Google leadership did hear the complaints, then realized what was happening and what could happen, and decided that something needed to be done to create and uphold values. I don't approve of defensive cynicism, I don't believe it makes a better world. In fact, I think that defensive cynicism is the reason why the world sucks. Let me guess, you…

Sorry for your pain. I understand this realistic approach bothers you a bit. As it happens, I expect decency from individuals, not corporations. My skepticism is justified in Google's post, since at the end of all the words, absolutely nothing was changed. If Google were truly listening, they would stop working on the project immediately. Therefore, rather than outright contrition, this post reads more like "Sorry, not sorry."

Politicians follow Hotelling's Law. It doesn't actually matter who you vote for as long as you vote. The only thing that changes is the flavour, and that's a matter of personal preference.

Post reply on HN