Who Should Stop Unethical A.I.?
newyorker.com
newyorker.com
Less tersely: this article is one in a long procession of journalists trying to exert control over tech. The opening example (Speech2Face, which they aver is transphobic) is inflammatory and utterly unrepresentative of the usual topics of AI conferences. The other references are far better, but the choice is revealing-- it's not so much an abstract concern about an unaccountable few exerting control from the shadows, but alarm that someone else might be muscling in on their territory.
I think the article paints a good picture of the machine learning research community figuring out how to grapple with the growing number of people who want to probe its ethics, without portraying anybody as a villain.
On the other hand, I think it's legitimate to criticize research that has many obvious bad applications and few obvious good applications.
(Perhaps an important factor is why this research was being done in the first place? Did they build a "reconstruct what a person probably looks like from their voice" just to see if they could, or did they have some real application in mind? For instance, suppose they were working with a group of oceanographers who wanted a system to infer the characteristics of orcas from their whalesong as part of a project to track their migration habits, and the researchers decided to test their system on people first because it's easier and less expensive. If a project was undertaken for a good reason, but it opens the door to a lot of bad applications as an unintentional side-effect, should the researchers get a free pass on the latter?)
Just like the tool can not know if I have long or short hair based on my voice, it also won't be able to know anything else that has deviated from the average.
This here is the problem, and public discourse has a solution to the word you're seeking: transinclusive.
"-inclusive" succinctly provides a hint on how to be considerate, while being optimistic.
Hate & phobia issues aside.
Edit: Most research begins describing exactly why the research was undertaken.
We're all abnormal, which is why it's important to be careful how we use statistical models, but the acknowledgment and exploitation of normality is not an attack against people who deviate from it, it is the foundation of statistics and all of science. Calling the authors bigots does nothing to improve these issues, rather the opposite.
What kind of simpler statistical algorithms do you believe are used in news feeds (which do not belong in the "AI" category)?
What matters is that the authors of the article mean machine learning when they say AI. So they’re talking about unethical machine learning.
Which is basically unethical maths, whatever that means.
Are you suggesting that it would be strange to believe that ethics can be applied to mathematics?
So the real trick is, how can we turn down the rhetoric so that we don’t get instantly worked up by every fake video that crosses our feed?
At the end of the day the world is shaped more by material conditions than ideas. That doesn't excuse the cases you're worried about, but I think the majority of violently polarising media will only end up being so because there was already some resource dispute in the first place. It would make more sense, then, to tackle those issues, rather than fruitlessly trying to engineer the human brain to be less susceptible to rhetoric.
I suppose you have heard about religions and faiths. I'd hazard to say that about 1000 years of European history was shaped primarily by them. Major wars fought, new countries founded, major discoveries and inventions made or unmade and prohibited, people's daily lives shaped and defined.
A similar thing has been occurring to Arabic world and neighboring peoples for about 1500 years, and is actively continuing now.
Looking at the current situation in the US is also illustrative.
I would say that ideas greatly prevail over material conditions once immediate survival is secured.
Were it not for the special religious status of Jerusalem, they won't care. There were more important trade route chokepoints, and more convenient sea ports nearby, if trade control were the actual agenda.
People getting their news from random videos posted on Facebook are already lost to democracy for all practical purposes, so little would really change.
I'm curious about this. As I understand it, an ethicist is objecting to the publication of software that could be used for surveillance. If that is correct, does it follow that this ethicist would object to all software that could be used for surveillance? If so:
1. How does one distinguish between software that can be used for surveillance and software that cannot?
2. Presumably the ethicist would object to the dissemination of software that can be used for surveillance whether published openly or sold. If so, would they not object to the sale of iPhones, as iPhones contain software that can record video, and therefore surveil people?
It seems like whenever morality or ethics come into play, the slippery slopes start popping up like weeds.
These technologies can incarcerate people. It’s really important to dive onto the slippery slope and find out where it’s stable.
Therefore, I ask: how does one tell the difference between software that can and cannot surveil?
So most useful record-keeping software can be used for surveillance. Most software capable of inferring data about the real world from sensors can also be used for surveillance.
Most likely a better ethical framework for talking about surveillance is "how can surveillance (its components, owners, operators, and consumers) be identified in practice and what are its effects?" Surveillance is a subset of knowledge, and most knowledge is both useful and amoral. The use of knowledge for unethical things is the problem, not the knowledge itself.
Names, for example, are probably the oldest technology for surveillance; used since pre-history to refer to specific real people and data about them such as the names of locations they reside in or visit. No one blames names for the effects of surveillance.
Which I think most would agree is the standard definition of surveillance, and I also think the concept of being surveilled comes naturally to most, which is why parties that surveil people make clear efforts to obfuscate the fact that they do so.
The sticky wicket seems to be doing it in real time.
Rather than throw up hypotheticals which this ethicist is probably not around to refute, I think it's be better to look at the person's body of work in this area and if you think they have some sort of intellectual blindspot or bias, point out how you think that might detract from their analysis.
But the most likely explanation IMO is that Hanna isn't using the term "surveillance" as a pointer to any concrete idea about what will go wrong. It seems to be just an all-purpose explanation for why AI models need to respect her sensibilities.
I actually mean to ask the exact question: How does one distinguish between software that can be used for surveillance and software that cannot?
To be more specific, if an ethicist objects to the publication of software that can be used for surveillance, they are assuming that the software can be used for surveillance. How did they arrive at that conclusion?
Of the 16 linked papers/book-chapters a few were not freely accessible.
Of the ones that were accessible, 3 contain the string "surveil" in the body and 1 contains it in the bibliography. None of the references were helpful unfortunately.
Do you have any other links that may answer the question?
The result that most of technological society would have a small amount of ambient guilt is not entirely absurd; the idea that industry is not sharply, but diffusely and slightly evil is common in fiction and many people's intuition.
1. Miner mines iron
2. Manufacturer makes knife
3. Retailer sells knife to person A
4. Person A kills person B with the knife
I think most people living in the US would be ok with the general idea that person A should be held responsible for killing person B, and not miner/manufacturer/retailer. Perhaps I'm wrong.
Would you consider the knife retailer equivalent to the torture rack manufacture in your example?
Not GP, but doesn't
>> I think it's pretty straightforward to assign more or less blame depending on the specialization of the technology for the purpose of doing something evil
answer your question?
I think that "depending on the specialization of the technology for the purpose of doing something evil" merely shifts the question to multiple questions:
1. What does "specialization of the technology" mean?
2. Can technologies have multiple specializations?
3. What does evil mean?
4. Even if we had a perfect translation between "specialization" and "evil," if a technology has multiple specializations, which controls? The most evil? Does this require us to predict which specializations are most likely to be used?
However, to my knowledge, neither of "accessory to murder" and "criminal organization" have been applied to the manufacturer of a legal-to-own object.
This comment is meant to make an analogy between the manufacturer of a legal-to-own object and the publishers of an ML algorithm.
2. I instead presume ethicists object to the unethical surveillance of users by any means, and warn of risks they find or foresee.
There is no slippery slope; the unethical effects of surveillance should be countered by effective and ethical means. Banning technology is not ethical or particularly effective. Banning and punishing unethical behaviors is about the best we can do.
In other words, don't ban iPhones. Ban configuring iphones to surveil on users in unethical ways. 911/999 geolocation is ethical surveillance. Stalking apps are unethical surveillance. There's ethical grey in between.
What is the killer app for the technology under discussion? If the killer app is improving camera filters to make more vibrant pictures but it can also be used for surveillance then you do not worry about it being used for surveillance. If however the killer app is surveillance then you worry about it.
As a general rule surveillance however is not the killer app of these technologies, surveillance is just an instance of the killer app, and the killer app is the more general one of privacy violation of which surveillance is an instance.
>In this paper, we study the task of reconstructing a facial image of a person from a short audio recording of that person speaking.
I can imagine a potentially interesting application of this that has nothing to do with privacy violation or surveillance (that I can see):
What if we could take an audio recording of an ancestor or someone else of interest, for whom we have no pictures, and output a probable face for that person? Sounds interesting and potentially "killer" to me. Maybe I'm wrong.
I can see every police department and government agency paying plenty of money for being able to identify the speaker of a particular sentence in a meeting.
I can see the company with this technology providing services for academic research of the interesting kind you mention for the tax write-offs and also a sort of moral cover, but there is really only one thing I see as their main revenue stream.
I guess the question is, who decides what the killer app for a technology is?
I wonder if there is an example of someone objecting to a paper that did not release the training data or software that can predict (aka just the algorithms).
The point I'm trying to get at is: it seems that most ethical objections are to the result of applying algorithms to ("bad") data, not the algorithms themselves.
since the 90s everything that involved a computer was not prosecuted because lawefforcement/regulation didn't have the skills, not for lack of laws. But new laws were created spelling out computers.
A.I. is the same, it is not nothing different from enforcing existing laws and regulations, but because the agencies lack the skills, we will see a new wave of stupid laws that only misses the point instead of proper training and funding of what we already have.
There is zero need for new laws if a company doesn't hire or sell to a minority group, or their products explode, or whatever. No matter how much they scream it was the bad-brain-in-the-computer fault. Sadly, lots of companies will get away with this silly defense for a while.
laws should not care about the means. only the results you want to promote/punish.
What if laws about murder had to be updated every time a new bullet caliber or knife length become popular?
yet that is exactly what happens with money laundering laws and computers. or media regulation. or investment schemes. Every new technology and everyone forget scams via email can fall under scams via courier mail. go figure.
This is because there is an increasing proportion of people that is not going to the conferences for the science and instead to try to hijack the conferences with their own agenda. It's of course their right to make a fuss, but it would be nice to still have venues that are focused on the science and not the other political stuff.
For example, basically any financial evaluations of US citizens will likely result in an inherent bias against Black people due to institutional biases such as redlining that have long lasting socioeconomic and demographic repercussions. This might mean that something as simple as incorporating the zip code of a home into a mortgage pricing AI can end up with a racial bias.
In your financial evaluations example, many biases and disadvantages would remain even if you solely reduce the decision to relevant financial facts for a specific individual, because a poorer individual with lower socioeconomic status and less opportunities actually has a higher risk of non-payment, and a disadvantaged group will have disproportionally more such individuals. Should we accept that? Should we require the other groups to subsidize their non-payment? Both options are unfair in some aspect and fair in another, you can't have your cake and eat it too, and it's not the fault of the model you use - the only difference with a human is that they can better hide the factors they use, lie about the influences (perhaps also lie to themselves) and rationalize/invent factors to justify their decision.
For this topic, perhaps this talk "AI Ethics, Impossibility Theorems and Tradeoffs" https://www.youtube.com/watch?v=Zn7oWIhFffs or its slides https://www.chrisstucchio.com/pubs/slides/crunchconf_2018/sl... might be interesting for you, it has some flaws but is a decent exploration of the problem space.
Let's say you have a racially-neutral observation of lower income, maybe disability status or a criminal rap in the past. That looks like a bad bet for a loan regardless of color, it just so happens that our society's created a statistical imbalance in those metrics.
I mean it is never truly the "AI's fault". It is the fault of the systems that resulted in the biased data and the people who ignored that bias while still delegating the decision making to that AI allowing it to be a tool that propagates those biases. You end up with a dangerous and self-perpetuating system like this:
1. Data is biased against Black people due to historic racism.
2. AI is built of this biased data.
3. AI results in a biased system that disadvantages Black people.
4. Black people get discriminated against due to the results from the AI.
5. New data is now even more biased against Black people.
6. Go to step 1.
The ethical concern would be, "don't build a racist AI system that assumes black people don't pay back loans". This could be mitigated by removing race and clear proxies such as zip code from training data, but it doesn't necessarily break your cycle, because the poverty and social problems are real and exist regardless of whether the AI considers race.
The anti-racist approach, on the other hand, would specifically require incorporating race so you can perform affirmative action. Maybe that's good for society, maybe there's a debate about how much and in what way you do it, who pays for it, etc. But it's left "ethics" far behind and become "politics" at that point IMO.
It isn't this simple. Not all these proxies are clear. Also what are we even left with once all these proxies are removed? The zip code is hugely important when trying to establish the value of a home. Does removing all data that is potentially racially biased leave us with a model that is nearly worthless?
>But it's left "ethics" far behind and become "politics" at that point IMO.
It depends entirely on what branch of moral philosophy to which you subscribe. For a lot of people their personal politics is just applied ethics meaning not acting would be ethically wrong in their opinion.
AI shouldn't take the blame. Blame the folks collecting biased data, or those making biased decisions encoded in the data. The data is known to be tainted. Blame those using that data to train models, and sell/rent/apply those models for profit. Blame the researchers who know, or should know, better but make breathless claims about how their AI can be used without regard for the impact if people follow their advice
Is it, just don't do data, same interest rate for everyone, no denials and amortize defaults across higher rates for lower-risk borrowers?
You'd need a law, the first bank to do that would be crushed by other banks that can attract the lower-risk borrowers with lower rates.
Curious. Is your answer to the title "we need laws to regulate AI?" I'm not specifically agreeing or disagreeing.
To remove this discourse from models ironically means applying a strong bias imho.
I'm more concerned about billionaires with an agenda than no-name researchers giving a talk on cyber-intersectionality.
Enough said.
Well forgive me for being so intolerant to state such an offensive opinion, but I'm really getting fed up with transgender issues taking front and center with every possible place they can insert themselves, as if they're our top priority when in reality they're a 1% problem. Ironic when you think of what groups actually push this.
I personally have never seen such an issue be so flogged to death by a willing majority of self-proclaimed liberal society who probably have never even spoken directly in a conversation to a transgender person.
Worse, it makes it seem like we don't have equally important yet bigger civil rights issues that have not even yet been resolved. Personal pronouns, ambiguous "they" speech, policing algorithms for transgender hate, contorting ourselves as if we're majority transgender and being persecuted left and right.
Look, I'm not saying anyone of any group should be made to feel disadvantaged. Or that a minority just because few in numbers should be overlooked.
But for fucks sake, enough with transgender agenda being made to seem like it's the top concern of society, ahead of those who have been patiently waiting in line for their fair treatment.
/end
For example, right now I personally am considering a particular ML research project in cooperation with our local police agency. I am aware that in many places around the world there is an antagonistic relationship between the local communities or the people and their police system, where helping the police can be reasonably considered unethical or even outright evil in certain areas for totalitarian regimes. But I'm happy (or privileged?) to live in a place where I do trust my law enforcement to be on the side of our community, and I do consider helping our police to be more effective as not only ethically permissible but as a social good, as helping the people and country. With that being said, should I be worried about the research results being tainted and unpublishable? I can certainly imagine a perspective or validation rules that could label this as "line of research that shouldn't exist" simply because it does enhance the power of police that could also be misused against the community - but, as I said, at least for our community I consider improving police capabilities as ethically permissible and even desirable and see no need to import the "government is a threat to us" mentality from places which do have this (sad) problem.
Perhaps the example in the original article about an NSA booth at NeurIPS conference is relevant - like it or not, the voters generally have supported the concept that NSA should exist and should get funded, that their mission in general is ethically justified and necessary. While it has had specific projects and actions that are arguably unethical, it does not mean that it's reasonable to label everyone working for NSA as unethical and treat helping NSA as taboo - the society apparently considers NSA as a whole as net positive, and there is no serious push with widespread support to dismantle it and its functions. Similar arguments apply to military (is it ethical to be a soldier, to train to more effectively kill people at the service of your country?) and arms development. There's also the potential asymmetry of location - perhaps you might consider it ethical for a USA citizen to work for NSA and unethical for them to work for Russian or Chinese intelligence, and vice versa. Which ethics and whose ethics, indeed.
May I ask where that is? Because it could not be any location in the US.
In mid-sized cities and large urban areas, the police are generally extremely corrupt and integrate with high level politicians and corporate interests for bribes, carrying out policing actions to serve cronyism, etc.
In small towns, police typically align with small town politics and specific locals with political power, ranging from judges to attorneys to business owners. If they decide you’re on the wrong side, they’ll find reasons to harass you, use speeding tickets, littering, frivolous fines, etc. Of course this often affects underrepresented groups, but can easily affect people in the majority if they just don’t like you.
Most communities see the police as a group to strictly stay away from. You hope that in a real emergency they will do their job, but more often they are just authority theater looking for a way to jam you up or exploit you. They absolutely are not good natured community members by default, and even if some specific cop is trying to do right by the community, nobody can assume that is trustworthy and everyone has to just stay away.
Policing, especially the procedural mechanics of fining or arresting people, is just fundamentally broken from first principles in the US, and most police systems exploit this for corrupt gains and abuses of power. Exceptions are so rare as to be ignored and assumed irrelevant.
Now machine translation is starting to make it possible for some communication and cooperation to take place between people who are completely monolingual. So should children still be required to study foreign languages? Should international students be allowed to use MT when writing papers for university classes? Should university applicants be required to state whether or not they used MT (or GPT-3!) when preparing their admission essays?
Compared to problems like surveillance and profiling, these issues are trivial. But the fact that they already exist, and the fact that the people involved—educators in this case—have barely begun trying to understand their implications, suggest to me that discussion of ethical issues surrounding AI is vitally important.
Convexity trumps understanding.
Method: Randomize exposure to autoplay, associate with other user data.
Now you just have to decide what labels you want to optimize for. Do you value clicks? Complexity of linguistic output? Calories burned? Rate of visitation to white supremicist websites?
All the other data (including autoplay) are festures.
I wonder if a good solution would be simply to have a rating system for accepted papers, with ratings applied by the reviewers. Basically, you'd have two numbers:
Useful rating: On a scale of 1 to 10, is this a useful system that solves a real problem to make life generally better?
Problematic rating: on a scale of 1 to 10, can this system be used in ways that create new problems or exacerbate old ones?
So, let's use the example from the article of a system that predicts what a person looks like from their voice. Maybe the author submits it to the conference, and the reviewers accept the paper, and give it a useful rating of 3/10 (it could be used to identify criminal suspects with low probability, maybe?), and give it a problematic rating of 8/10 (could be used to discriminate against minorities or automatically apply cultural stereotypes; could be used for surveillance, could falsely implicate innocent people for crimes, etc...).
The author could then decide whether they want to continue publishing the research, or drop it. For all published papers, the ratings are listed in the table of contents of the proceedings, and the papers in each section are sorted with the most useful first and the most problematic last.
I genuinely don't understand why we've given such influence and power to boorish people like Alex over the past decade.
It doesn't matter if Neo can dodge bullets if the AI wields the Ban Hammer.
It’s just a hype vehicle to cram in whatever social justice outrage du jour that some liberal source wants to push as an agenda. I say this as a liberal who is sympathetic to most of those issues, but finds their representation in AI ethics or fairness debates to be ignorant and juvenile power grabs.
The whole Timnit Gebru affair is a direct example. The paper Google didn’t want to approve for publication was really poor. Lots of specious arguments. Lots of declarations of opinions as if they were fact (eg. nobody is required to agree with or accept woke vocabulary, let alone ensure NLP models are kept updated with it), and lots of sanctimonious assertions about various agendas that nobody has to agree “are right” and have zero appropriateness within a science publication.
If I'm interpreting your argument correctly, I think you're criticizing either section 4.1 Size Doesn’t Guarantee Diversity, or section 4.2 Static Data/Changing Social Views. But your criticisms don't track with either of those sections. 4.2 seems the best fit (since they discuss keeping models up to date with modes of communication). The claim made in the paper seems to be that an LM trained today will fail to keep up with shifts in language (and these can happen quickly, in months, not years in many contexts). This is true whether the context is "woke vocabulary" whatever that means, reclaimed slurs (such as "queer"), or memetic slang/shibboleths ("kek" or "do you listen to girl in red"). I don't see anyone saying that language models or designers are required to agree with "woke vocabulary". The closest I can come is the authors suggesting that models which fail to stay up to date with shifts in vocabulary (including, I guess, woke vocabulary) will be less effective. That seems trivially true. Can you clarify?
[0]: https://faculty.washington.edu/ebender/papers/Stochastic_Par...
The other major problematic section is on energy use and unqualified comparison of units of large model training with carbon emissions of other activities (with no attempt to account for societal value of ML model research or training, nor economies of scale as training technology democratizes it and makes it cheaper). It is similar to the incredibly specious diatribes from Stephen Diehl lately balking at the raw energy numbers of Bitcoin mining, as if those numbers mean anything or relate to any conclusions on their own.
I wrote up a lot more of my thoughts on Gebru’s bad article when it was first leaked. Here’s that comment:
- https://news.ycombinator.com/item?id=25314824
I also feel you are very commonly involved in lengthy downvote / flagging threads with disingenuous takes on others’ comments and deliberate unwillingness to apply charitable takes or avoid taking things out of context. I am sure from your POV you don’t feel that way and I’m not asking anyone to agree with my assessment; I have zero goal of citing things to back it up. I am just going to preemptively drop off from replying to you any further.
I think this is a completely fair summary of the paper.
But you described the paper as suggesting that people are "required to agree with or accept woke vocabulary" or previously your criticism was "Nobody is required to accept “wokeness” vocabulary".
I don't see how you can justify those criticisms from the paper. They aren't there. There isn't a place where they paper says that anyone needs to use or agree with a particular kind of vocabulary.
I don't know if it's intentional or not, but the claims you're making about what the paper says just aren't in the actual document. Hence my question: where are you getting that criticism from? Where does the paper suggest that people need to agree with woke vocabulary? I think it's uncharitable for you to criticize the paper based on a mischaracterization, and so I think it's completely reasonable for me to clarify by what reasoning you came to make that criticism. Because again, despite trying I can't find an interpretation of the paper where your criticism makes sense.
I'd implore anyone else reading to look at sections 4.1 and 4.2 as well. They just simply don't say this.
I hope people will read it, because a direct reading of it does support my position and does not support your position.
The premise is: why do we care about some forms of inequality (financial inequality among households in one nation, or protected class inequality within employment law in one nation), but we don’t care about other things (inequality between socially lonely people and social butterflies, inequality across country boundaries, inequality of physical attractiveness).
Hanson suggests it just so happens that we develop “morality” around some forms of inequality because it offers easy ways to politically coordinate to grab money or power and punish your rivals.
Other forms of inequality that involve just as much harm, but where it’s not politically easy to drum up support for physically taking money or power, just happen to conveniently not turn out to be big moral concerns, or may even have other competing morals invented over time that make it “wrong” to even treat it as inequality (like the hallowed status of a “right” to control what happens to your body counting for more than inequality based harm of variation in genetic sexual attractiveness outcomes - which is a massive double standard compared with no such “right” to control what happens to the fruits of your labor or intelligence, which undergo forced redistribution via taxes because we just so happen to believe inequality of financial outcomes is not acceptable).
For example, most people don’t give a shit about animal suffering on the scale of agribusiness. Most people don’t give a shit if ugly people suffer loneliness or unhealthy lack of sex or job discrimination from their appearance. Most people don’t care much about structural inequality in countries besides their own.
These are just convenient moral omissions because in each situation it is not easy to see how you could “redistribute” attractiveness, social companionship, or power structures of some other country. Plus animals don’t have any money or status to give you in exchange for allyship, while big agribusiness has plenty of lobbying power to crush you with.
I wish more people appreciated this. We don’t, as a society, advocate for race, gender, or pay equality because it’s “right.” We just know that aging white patriarchy is full of convenient targets we can take money and status from. Conveniently for other domains of inequality, like animal welfare, where no such easy targets exist to take from, it just somehow doesn’t qualify as a moral issue worthy of advocacy.
[0]: https://www.overcomingbias.com/2013/08/inequality-is-about-g...
Therefore, AI research is unacceptable and they are trying to cancel AI research and censor and deplatform AI researchers.
The most contentious aspect of AI seems to be whether you're trying to generate porn with it, somewhat amusingly. Almost 100% of people who are, are terrified of being found.
https://quillette.com/2021/01/27/beating-back-cancel-culture...
I need to be clear that I don't believe in censorship and it's fine if this is what the author wants to write, but it should be called out how irresponsible it is. On a place like HN, the audience is generally able to read such things critically and form their own opinions, but for an audience that is not tech savvy (like that of the New Yorker) making false or hyperbolic claims about ethics and bias in research is both misleading and has the potential to really interfere with peoples livelihoods, funding, and legitimate research progress.
There needs to be more calling out of this kind of nonsense, the same as if people are posting about vaccine conspiracies or something.
If you create a true intelligence, it will necessarily be different from us.
To create a true intelligence, and make it your tool is unethical, at least to humans. Its slavery.
Perhaps equally importantly, I'm skeptical that you can limit ethics rules in this way. If we clearly establish that suchandsuch research would be unethical at NeurIPS, I don't expect sociology conferences or think tanks would be eager to continue it, and I'm not convinced Alex Hanna would want them to.
I'd, in some ways, expect Dr. Hanna to ascribe to the axiom that predictive policing is unethical. As such, applied research into "improving" predictive policing would also sit squarely in the unethical space.
And I expect that's all you're going to get at an ML conference (outside of explicitly ethics/critique papers): "Our model predicts crime 4.2% better than competing models"
However, I'm not convinced that research into the meta-question "Can predictive policing be done ethically" would be considered equally unethical. ML is only a very small part of that meta question though. You're looking at much more complex socio-political changes that don't belong at NeurIPS ("do we need to change policing so that predictive policing can be done ethically").
> ethics saying they can't study minimum wages or tax increases
In general, we have far more oversight into the elected officials who manage tax and minimum wage policy than police forces. These structural differences in influence and oversight play into the need for self-regulation.
> Relevant expertise matters.
There's nothing that prevents a think tank from working with domain experts (in fact, iiuc this is often what happens: a think tank will contract out to domain experts to write a report on some subject). Now there's all kinds of other conflict of interest issues with think tanks, but we all know that already.
One issue here is that the pool of domain experts depends on the different incentives faced by different researchers. If the top machine learning conferences all decide to reject papers about predictive policing, then this will reduce the incentive for machine learning researchers (who mostly want to publish in these conferences) to work in the area at all.
Maybe critics of predictive policing think this is good, or think that it's better for the field to restrict itself to the few computer science researchers who are prepared to seriously devote themselves to it? But there don't seem to be very many of them. And, as someone who thinks there is a lot of room for police departments to in general do a better job --- and I think most of the people arguing against predictive policing would believe the same! --- it's disappointing to me that the end result of this argument seems to be disincentivizing technical research on it.