Hacker News Reader

Comment by 908B64B197 | Hacker News Reader

Again with the "Ethics in AI".

You can't say it publicly but ethics is a trend. It's trendy right now to apply it to AI because there's a lot of academic funding going into AI so fields where there's no value production (like ethics) will slap "AI" on their grant proposal.

The point of ethics is to shame people or practices you don't like via some kind of cancel culture of peer pressure [0]. Of course, it's impossible to do it in court as you must actually prove things. So you get a large enough mob instead.

There's mostly no point in debating with the ethics crowd (you really can't do that since their papers typically won't be reproducible). Or acknowledging their existence really.

[0] https://syncedreview.com/2020/06/30/yann-lecun-quits-twitter...

View on HN

It just isn't true that "The point of ethics is to shame people or practices" or that ethics has "no value production".

In general, a primary factor reviewers in computing conferences are asked to consider is the degree to which a submission make a "substantial contribution" to the community. What is or isn't a substantial contribution is subjective and entirely dependent on the prevailing ethical perspectives in various communities. Papers -- a key unit of academic progress (for better or worse) are entirely subject to concerns of ethics. Certainly, there are interesting argument around how much time and paper space should be spent on speculating about negative impacts, and people are having those conversations.

The fact that ethics is "trending" is because more researchers would like tackle ethical challenges explicitly, instead of falling back on the default of a given community. For instance, here is a paper that quantitatively (with reproducibility!) analyzes the values in ML papers. https://arxiv.org/abs/2106.15590 This is one way to have a very empirically-grounded discussion of the topic.

IMO, many researchers can and do debate topics of ethics in AI, and in doing so move the field forward (and increase likelihood that computing will have more positive impacts than negative ones).

> The fact that ethics is "trending" is because more researchers would like tackle ethical challenges explicitly, instead of falling back on the default of a given community

The trouble isn't that ethics is trendings, it is that it is happening in 2021.

Rant: The world has gotten poisoned by social media, and that has enabled screwball US academic philosophy to leaked out. In the current climate, philosophers are either cracked or cancelled. Most people who wax lyrical on ethics in AI (and basically anybody who actually calls themselves "an ethicists") are insufferable.

To think about ethics in technology you need to step outside the narrow confines of our current cultural assumptions of right and wrong, because they're are prone to fashion and hysteria. Right now, if you try to do that you run the risk of ending your career.

The “point of ethics” isn’t cancel culture and if you believe that you’re either being willfully dishonest about what ethics is and it’s entire history to make your own political statement or are actually ignorant of what ethics is, both in reality and historically.

By all means if you want to make a political statement about how ethics is being applied (which would actually imply ethics being used as a smokescreen in this case) go ahead but we’ve been grappling with the concept and idea of ethics for thousands of years as a species.

Let's consider your comments applied to another ethical area: eugenics. That was a previous area where scientists did something which is now considered highly unethical (remove the sexual autonomy of people due to incorrect scientific judgements). Would you say the people who opposed eugenics were "shaming" the scientists, or helping "guide them to making better decisions?"

I think there are many concerns around ethics in AI, but the majority of the players are not contributing (folks like Timnit, with the stochastic parrots paper, the sections on power consumption are completely technically wrong). Sara Hooker is a much better spokesperson for this, but I'm really struggling to see the point in her paper, beyond "we should identify existing data biases and make algorithmic changes to reduce them" which just sounds like "we need regularization that is the opposite of L1 and L2".

Your first paragraph is called the Motte and Bailey fallacy [1]. You can't win your original argument, so you pick a different one that's easier, and then pretend that you won the original.

[1] https://en.wikipedia.org/wiki/Motte-and-bailey_fallacy

That's a very uncharitable (and in my judgement inaccurate) characterization of GP's argument - the usage of the MaB fallacy as you've done neglects the principle of steelmanning your opponent's position, and shifts the discussion from a fact-finding mission to "winning an argument", both things that are anathema to good discourse. The article you linked expands on this criticism of the MaB fallacy in the 'Criticism' section.

GP took advantage of an analogy to counter the claim that ethics is only about shaming and calling people out. Analogical reasoning is a powerful philosophical tool that can lead to important insights and discovering new questions. In this case, GP used the analogy where we apply ethics (relatively uncontroversial) in a scientific context. If the comparison is successful, it counters the claim that ethics is only a fashionable way of shaming people.

Lastly, GP never claimed to 'win' the argument with the analogy (even if it were a motte-and-bailey). It's entirely possible that GP has more work to do after getting a response to the analogy.

I didn't even make argument; I'm just asked the author to reconsider their statements given that there is historical evidence that ethics is more than just shaming and cancel culture. It's more https://en.wikipedia.org/wiki/Argument_from_analogy and https://en.wikipedia.org/wiki/Appeal_to_emotion

Your comment is the "fallacy fallacy" where every statement is countered by the claim that it is in fact a fallacy.

The person you responded to has a cogent reply and you haven't done anything to address the meat of the matter.

Ironically, it's just what Timnit Gebru did.

"I can't prove algorithms are unethical, therefore I'll make a point that the (clean) energy used to train them contributes to global warming (by assuming all the energy required came from jet fuel)". [0]

And then ethics tries to claim it's a real science!

[0] https://www.technologyreview.com/2020/12/04/1013294/google-a...

What is truly extraordinary about that paper is that it completely ignored the fact that ML is only a tiny contributor to overall power usage in computing, and also attempted to compare the training to intercontentintal flights, which use the same amount of fuel empty of full (and the planes do fly nearly empty). All of it makes it look like she just wanted to attack ML, rather than make a good faith argument.

The other part of the paper (the danger of trusting things that stochastically parrot well enough to exit the right side of the uncanny valley) is interesting.

I think she's all to willing to throw the baby (the transformer model) with the bath water. She doesn't care about AI advancement at all, and that makes me distrust her paper.

Since "I can't prove algorithms are unethical" implies a demand or requirement to prove a negative, I'm inclined to conclude that you're misrepresenting your source.

It's interesting you had to pick illegal practices to attempt to make a point, while my point was that ethics was used to shame practices when the legal system couldn't be used because of a lack of rigor.

eugenics wasn't illegal at the time. https://www.nytimes.com/2014/10/14/science/haunted-files-the...

When the Eugenics Record Office opened its doors in 1910, the founding scientists were considered progressives, intent on applying classic genetics to breeding better citizens. Funding poured in from the Rockefeller family and the Carnegie Institution. Charles Davenport, a prolific Harvard biologist, and his colleague, Harry H. Laughlin, led the charge.

...

By the 1920s, the office had begun to influence the United States government. Laughlin testified before Congress, advocating forced sterilization and anti-immigration laws. Congress complied. The Immigration Act of 1924 effectively barred Eastern Europeans, Jews, Arabs and East Asians from entering the country. And, at the state level, thousands of people who were deemed unfit were sterilized.

>while my point was that ethics was used to shame practices when the legal system couldn't be used because of a lack of rigor.

I'm confused as to why that would be a bad thing. For instance, liars are shamed, even though it would be ridiculous (and harmful) to create laws against lying in all cases. At the same time, the law sometimes takes account of ethics (rather than harm) to make laws, for example, the illegality of corpse desecration is not about harm, but about ethics - what it says about our society if it were legal.

"Liars are shamed" had a different meaning in 1990's than today. We've gotten to the point of pervasive surveillance and infinite storage. That's why public shaming, a social tool based on mob justice, is becoming an even worse act of violence. It destroys lives with little due process, riding on powerful preexisting social and political emotions.

> For instance, liars are shamed, even though it would be ridiculous (and harmful) to create laws against lying in all cases.

But lying under oath or false advertising are illegal.

> for example, the illegality of corpse desecration is not about harm, but about ethics - what it says about our society if it were legal.

That's morals, not ethics. And it's rooted in religions.

>But lying under oath or false advertising are illegal.

That's why I said "in all cases". Clearly, lying isn't only immoral in the cases of lying under oath or false advertising. It's still immoral for me to say that a ball is "out" in tennis when I clearly saw that it's "in", even if doing so is not illegal in most cases.

>That's morals, not ethics. And it's rooted in religions.

The difference is immaterial. Even most ethical codes are rooted in religious thinking at some poitn in their formation. That doesn't make them invalid ethical codes. Ethical codes are generally constructed on the basis of morality, or in some cases on the basis of professional conduct - but even that isn't only practical.

There was nothing incorrect about their scientific judgments. Modern behavioral genetics have confirmed both the concerns and the potential effectiveness of the eugenicists. What changed was that, as a society, we no longer think the trade-off is worth it: we are now willing to sacrifice future generations to quiet down our conscience. How much of this is actually a considered choice, and how much of it is simply down to the fact that future people aren't around to complain...I'll leave it to you to decide.

I think you're probably limiting yourself to the idea that "selection can be applied to human breeding" which I do agree seems to be scientifically possible. They were roughly correct about that (but far from being able to exploit that knowledge).

The ethical concerns in eugenics weren't about that at all, though, they are about removing autonomy from people based on morally and scientifically questionable ideas (the eugenicists had many more scientific ideas which were just not supported by data, then ones that were).

I am not as cynical as you but I see your point. I think many in the "Ethics in AI" community are terrible in communicating their ideas. They also have the tendency to claim moral superiority (supremacy?) whenever mere mortals like Yann Lecun make sensible observations.

That was a disgusting moment to watch. They refused to debate the issue in a proper format (meaning not Twitter). They called on him for being part of a privileged class (white man) as if that explains his accomplishments.

Later I got interested to know more about their views so I found out they have a policy of not giving a "platform" to their opponents - refusing to debate them in other words, preferring cancellations, shaming and appeals to emotion instead.

> I found out they have a policy of not giving a "platform" to their opponents

wow, that explains a lot! It sure seems like they take this policy very seriously.

Reply on news.ycombinator.com