The machine learning community has a toxicity problem
reddit.com
reddit.com
It can be said from every single field in science (I am in the natural/health ones):
- peer review is broken - check
- reproducibility problem - check (try to reproduce any miracle cancer cure paper)
- worshipping problem - check, there are kings and no one can take them down.
- diversity problem - check. In my department 80% of the professors and PhD and postdocs are women.
- moral and ethics are set arbitrarily - check. Morals? in Genetics? give me a brake
- there is a cut-throat publish-or-perish mentality - check, tell me something new
- discussions have become disrespectful - check. The time I saw Pavel Pezner giving a keynote lecture at ISMB and instead of showing his work he spend 70% of the time dressing down other people and trashing their work, Science died to me. And this was early 2000’s.
Machine learning won’t be the first field to notice it, and won’t be the last. Science is not scientific anymore.
Academia? Humans in general?
I'm not sure it's the answer, but I think there is something seriously amiss with our institutions that causes them to all have a similar set of self silencing momentum. The amount of corruption that has been exposed in institutions over the last 20 years that has gone largely ignored in traditional media and inside this institutions themselves is staggering. Every time we hear institution X has a problem with Y, the story quietly disappears and there is no institutional change enacted.
https://bigthink.com/culture-religion/eric-weinstein-intelle...
I blame it on exceeding Dunbar's Number and creating different groups inside of an org, which now have different priorities and tribal affiliations, e.g. Ops vs. Engineering vs. Finance, or sub-teams inside of Engineering fighting holy wars about Tool [X].
I think the Dunbars number idea is interesting in terms of possibly describing how our organizational scaling has outpaced our biology
Ultimately, scientists want to be understood and accepted by their peers and make friends, they are people after all.
A culture of abuse can start when the notion of exclusivity is accepted and outsiders want to join the group, whether that be in a single research group or within a field of study.
The problem is exacerbated by greedy individuals, and limited pools of funding to carry out research.
In turn, the greediest are the most exclusive, and the most prone to toxic behavior and chronic hazing.
Max Planck had this famous quote:
> A new scientific truth does not triumph by convincing its opponents and making them see the light, but rather because its opponents eventually die and a new generation grows up that is familiar with it. . . . An important scientific innovation rarely makes its way by gradually winning over and converting its opponents: it rarely happens that Saul becomes Paul. What does happen is that its opponents gradually die out, and that the growing generation is familiarized with the ideas from the beginning: another instance of the fact that the future lies with the youth.
This was an unexpected twist! It's rare to read such an honest, unbiased opinion on this issue.
It really shows just how systemic bias can be.
Admittedly this makes me uncertain what studies have been done to prove or at least anyhropologically report on effective movements at increasing diversity. We already know (or have evidence that) more diverse teams are more effective teams.
1. The expected working hours and travel schedule
2. Poorly-compensated, unstable unemployment through the mid-thirties as de riguer
I'm a man, but my experience is relevant in that I left early for industry from a Physics PhD that was dragging on, in large part because I wanted to start a family. I'm now, according to the feedback I receive, a high-performing C++ developer in the CAD / CAM space, and I didn't have to move around the country doing temp work for below-market compensation, work crazy hours, or suffer through the degrading administrative bullshit one must endure to convince a committee. And I have a seven-month-old and a partner getting a master's on my salary, and we get to live near my parents.
It was unpleasant to nod along when "enduring the company of gross assholes like you" was presented (rather than the above) as the dominant discriminatory mechanism, but, you know. There were plenty of gross assholes and it was a learning experience and I guess I'm now probably better for it.
Also, when you have a bunch of very intelligent and hard-working physics grad students, how on earth can "working hours" not be a discriminatory mechanism between them, when the coveted faculty positions are so few and far between? Does it not stand to reason that a person who devotes more of his day to the work will come out ahead, on average?
It isn't good or bad, it simply is.
These are human-created systems, though, and would be built upon human ethics and morals. Stating academia as an institution just “is” isn’t much different than asserting the military-industrial-congressional complex just “is”
Doesn't it? If you have two lions of the same physical ability and one of them hunts for six hours a day and the other one hunts for 12 hours a day, which one would you expect to be more successful?
> Stating academia as an institution just “is” isn’t much different than asserting the military-industrial-congressional complex just “is”
It's not about academia specifically though, it's about any and all systems where individuals compete. Given similar talent, the one who puts in more time/effort will likely win.
We build our institutions to guard against this. That’s why we don’t live in an anarcho-capitalist society and have things like anti-trust laws.
Obviously, but if you consider those to be very similar at the top of some field, wouldn't you expect work-hours to be important in predicting the outcome? I'm sure you can draw clear lines between two individuals, but when you look at many people of similar ability, how much time they put in will most likely predict their results pretty well.
> We build our institutions to guard against this.
We do? Is there a thing where you can explain that you're as smart as other people and therefore it shouldn't matter that they work 5 days a week in some area and you only work one day, and that it's unfair that they produce more results than you and get the promotion/grant/whatever you compete for? I've never heard of it.
Sports is another example. Leagues put salary caps as well as minimum salaries to artificially limit compensation.
Both sports and economics put these in place to provide a bulwark against such "natural laws" in favor of human defined ethics like fairness.
> Both sports and economics put these in place to provide a bulwark against such "natural laws" in favor of human defined ethics like fairness.
That's not at all why we have anti-trust laws. They are harmful to the economy, that's why. If they weren't, that is, if monopolists usually were more efficient in lowering prices/increasing quality/innovating than companies that had competitors, we'd have no issue with monopolies at all.
It's a strange moving of the goal-posts to posit that sports doesn't count due to endorsements. There are also examples of leagues that limit these as well (UFC being one, although the case can be made this is of the managers interest). It becomes more pronounced in amateur levels (e.g., Olympics).
The military is an institution that sometimes put numerous limits on one's professional career regardless of contribution.
Someone above brought up E. Weinstein. He's been a vocal critic of how academic institutions use immigration institutions to artificially drive down labor rates. There are seemingly boundless examples of institutions either biasing or leveling outcomes.
The larger point being institutions do put artificial limits on all kinds of interactions. These interactions have societal ethics forced upon them and don't exist in a libertarian vacuum consisting of only natural laws.
> It's a strange moving of the goal-posts to posit that sports doesn't count due to endorsements.
If you go for total comp, endorsements is part of it. If you don't, you'll just see that the average salary in highly lucrative and competitive fields will sink and the stock options, bonuses etc for the top performers will increase. In both cases, the result is the same: the top performers get substantially more than the rest. The different is only whether you're trying to hide that fact by making them nominally earn close to the same, only to then give some of them the rest of the money in a different way.
> The military is an institution that sometimes put numerous limits on one's professional career regardless of contribution.
And they are also a very special case, with lots of intricacies and paranoia and not something that comes to mind when you ask people about rewarding merit and efficiency.
Nobody is arguing that anything and everything must "only consist of natural laws". But to claim that something as basic as "more effort = more results" shouldn't be allowed to be real because it's too much of a natural law and has no place in civilized society sound ridiculous.
It's fine if you want to argue that we shouldn't promote the top 1% of each year and murder the rest. I agree, and we don't. But to not consider the actual output of people competing for something because they might have put in different amount of hours?
Should we have the Olympics add more classes where only people may compete against each other that were born on the same day and had their training supervised and limited to a reasonable amount achievable by every hobbyist for their whole life, to make sure that none of them had an "unfair advantage" by training harder or longer than the others?
My intent is not to rail against the idea that “more work = more results”, all else being equal.
My point is that this is not something that should just be expected to run its course because it’s a “natural law”, and is thus some evidence of some fundamental truth. Society places limits on how far these “natural laws” can extend. The “intricacies” of special cases are exactly what I was inferring when I stated what I felt was too simplistic of a model; namely, reducing the systemic effects to a single correlation like “more work = more results”. In the real world I think those rare situations that can be boiled down to such a simplistic relationship are the exception rather than the rule.
We do add rules to many (most?) sporting events, largely out of ethics. Combat sports have weight classes, others have age restrictions, Olympians have pay restrictions etc. Whether or not your examples would be adopted is a matter of social convention about what is “fair enough”, so they seem to illustrate the point about society setting ethical boundaries.
I don’t think there’s anything wrong with the meritocratic goal of “more work = more results”. To get there, though, all else must be equal which is too idealized to work in the real world. So society creates ethical rules, outside of natural laws, to get closer to that level playing field. Insinuating differences influenced by institutional convention is evidence of some foundational truth is naive and potentially dangerous.
What I'm missing is what alternative you see. Sure, we could turn the relationship between input and output on its head and watch what happens when whoever crosses the finish line last wins the race. But what does that achieve?
> We do add rules to many (most?) sporting events, largely out of ethics.
We add constraints so we can compare the abilities, and that gets too hard to be useful when we don't focus on something. If we had a new sport, not unlike a decathlon, but testing every skill imaginable, I'm sure we'd find the field much closer together, and we'd probably have quite a few surprises, but it would take forever and wouldn't really tell us anything. Constraining it with rules allows for comparison.
Yes, of course, we could constrain researchers as well by how much time they were allowed to work, or which books they were allowed to read, and how often they are allowed to look something up etc, but we're not really interested in some very narrow, hyper-specific "research" skill, it's not a hobby, we want results. Ergo we look at who gets the most results, or gets them the fastest, or cracks the hardest puzzles, whatever you may use to compare researchers.
I think those three areas combine to create students with massive debt and limited prospects, ripe to be taken advantage of under the guise of meritocratic competition. What I’m seeing is that people may interpret success in this area as a natural outcome of what “just is” rather than the byproduct of a broken system.
> And they are also a very special case, with lots of intricacies and paranoia and not something that comes to mind when you ask people about rewarding merit and efficiency.
The elephant in the room is however, what is merit and efficiency in academia. I would argue the attempt to somehow make scientific output measurable has led to the big problems plaguing academia today, salami publishing, churning out more and more papers with less and less results, making papers purposefully difficult to reproduce, time spend on applying for funding instead of research, crazy work hours... In many ways academia is similar to the military in that short term incentives are actually counterproductive to the end goal that you try to achieve.
> Nobody is arguing that anything and everything must "only consist of natural laws". But to claim that something as basic as "more effort = more results" shouldn't be allowed to be real because it's too much of a natural law and has no place in civilized society sound ridiculous.
Again the difficulty being what does more results mean and is it necessarily good.
That being said I don't think artificially capping work hours (how even) would alleviate the situation, work hours are a symptom not the cause.
My IRL analogy:
Working in election integrity, I emphatically encouraged my peers to pointedly ignore intention, to frame problems as "errors", not "fraud".
From the systems view, all fraud are technically errors.
And the conversation is full stop over the moment someone says "fraud", "chicanery", "electioneering".
But if you can keep the convo in the "safe space" of "errors", avoid partisanship and blame and whatever, then it's easier to build coalitions, persuade others, neutralize opposition.
Just my two cents.
An election process can fail to provide validation against errors. A person, however, commits fraud.
You bring up the military, and I can draw another analogy - development of a military technology. Sure, you can decide nukes are amoral, but the more actors there are in the world, the higher the chance that someone else will develop them anyway, and will wipe the floor with you, amoral as though they may be. So do you really have a choice? Generall,y when a technological advance is possible and brings about significant advantages for the entity that develops it, it will eventually happen. It won't be good or bad, it just will be. Agonize about it or don't, develop it or let yourself be overpowered by the one who will - your moral choice entirely.
Your second paragraph I’m less sure about. It comes across as realpolitik (that may have been your point) which can be used as a rationalization to continue immoral behavior. My bringing up the military is that we can create systems that incentivize certain behavior (lobbying for a military need) and confuse that with natural law when in reality it can be mitigated through a differently oriented system (in this case, one that more clearly separates money and politics)
Now, you can guess based on my emotional language that I think it is good to pursue policies that make careers more open to people who want families. And, as it happens, the US government agrees, in so far as it designates wanting a family a protected class for the purpose of employment discrimination.
My argument, devoid of moral content, is:
If it's your aim to avoid excluding people who want families from a career, you need to regulate the amount of hours people in that career are expected to work, the amount they're expected to travel, and the distribution of lifetime expected earnings (don't unnecessarily gate a prosperous 40s behind penurious child-bearing years). Since this particular career is overwhelmingly government-funded, free-market analyses miss the point - the government decides how many Physics Grad Students there are, what they're paid, and largely what their future career trajectories are. Given that it holds the purse strings it also holds substantially more power to regulate working conditions than it otherwise would.
It's empirically obvious that people who want families are excluded from science, and I'm positing that the above is why.
However, my proposed solutions are quite different, and much more direct - simply address the main issues directly - the main issue IMO currently being housing (at least in major cities, near major colleges, in major Western countries). So you could make a simple rule, giving (either literally, or in a form of long-term (10+ years) loan that automatically gets written of under some conditions) young people apartments when they e.g. enroll in a difficult STEM PhD program. Once this problem is "sorted" (housing represents a major factor in how "secure" you feel about your future - you and your kids never going homeless) I think that would encourage many young highly-educated high-achiever couples to start families much sooner.
For several years an adult needs to be present 24/7 and if the adult is someone other than the parent that ain't cheap. If the adult is someone other than the parent and it's not business hours that really ain't cheap.
What's a better use of the NSF's dollars? Funding night-nurses for grad students so professors can keep working them through the night? Or just funding day-care and regulating their hours?
You definitely don't want to just give grad students assets that can be converted into liquidity after some fixed term, because then you're just exacerbating the "you get rich eventually, after you're too old for kids" issue. A better asset would be something like a seven-year voucher for day-care that is:
1. Granted after a two-year vesting period (so that people receiving this benefit get at least a Master's degree)
2. Non-transferrable (so students don't just sell it)
3. Not conditional on continued enrollment (so it doesn't tie you to an abusive professor / department)
Again, I think this only really works with regulations on hours (possibly with a shift system of some kind), but once the federal government begins funding such a benefit there's no reason to restrict it to grad students (I'm sure military service-people could use such a benefit as well, and the more government professions receive it the more flexible the voucher is).
So it's not exactly "poor compensation," it's more "poor compensation through the years when most educated people have children."
Why should the vast majority, or even a sizable fraction, of fields not have a gender or race balance even remotely approximating the balance of the general population? Is it not a 'should' but merely 'it's okay if they do'? If the balance was in the other direction and men or white people were being intentionally denied jobs, would that be okay?
Should it be OK for other things to be unbalanced? Should it be OK if 60% of black people can't get houses but only 10% of white people have trouble getting homes? Should it be OK if 90% of asians are turned away from emergency rooms but only 20% of mexicans are?
You can argue that race/gender genetic differences play a role but it's kind of hard to explain away the widespread imbalances here just based off DNA.
https://www.mercurynews.com/2019/06/07/google-discrimination...
"Now the plaintiffs can request access to internal Google documents to try to support their allegations, which also include some people being “denied employment because of their actual and perceived conservative political activities and affiliations, and their status as actual or perceived Asian or Caucasian male job applicants,”"
There are myriad examples of this sort of allegation from the last decade, it's not hard to find them. They've been discussed here on HN.
You're implying – and your argument rests on – that right now, women and non-white people are being intentionally denied jobs due to their gender or the color of their skin [1], and that this is a widespread phenomenon [2]. I would really like to see some hard evidence for that claim.
[1] As opposed to their qualification or other directly relevant criteria.
[2] I'm assuming you mean in the US, or at least in the Western world.
If you believe that racial discrimination in hiring (or otherwise) is made up, you need merely perform a google search to find many reputable articles and papers on the subject. Feel free to educate yourself.
https://www.google.com/search?q=race+discrimination+in+hirin...
https://www.beckershospitalreview.com/hr/gender-ratio-of-nur...
>Show me one field (that does rely on the person's ability not looks) where there are more women in the higher ranked positions and more men in the lower ranked positions.
Nursing
https://www.theatlantic.com/politics/archive/2015/01/liberal...
Did your teacher tell you stories of Marie Curie passionately? Did your teacher tell you stories of Noether and her amazing theorems with great admiration? Did your teacher tell you the inspirational stories of Ada Lovelace? Did your teacher and parents tell students that everyone can achieve the same level of mastery on STEM? Did your country treat engineers and scientists as heroes? I guess not.
And now a bunch of lefties are complaining there are fewer women scientists and engineers in the pipeline and attribute it solely on discrimination or bigotry?
Get your priority straight.
> Parents mock geeks.
I was a kid nerd. I was mocked. (And bullied too, but I’ll take that one for the team.) Trust me, there was absolutely no encouragement provided for me, except by a few geeky teachers. If girls receive(d) approximately zero encouragement, that sounds like equality to me (though not how things should be!). The idea of geeks being cool, rich and popular is very recent, brought to you by the likes of Zuckerberg, Elon Musk, and Big Bang Theory.
You're right. I somehow fell into the framework set by the progressives. Any gender can do STEM, and the country I came from ensures that such message is loud and clear. Having female role models helps, but that's beside the point.
Nobody said that they can't be, that's a straw man. But there is a difference between not being inspired by male scientists and never having a female scientist mentioned to you.
> > Parents mock geeks.
> I was a kid nerd. I was mocked. (And bullied too, but I’ll take that one for the team.) Trust me, there was absolutely no encouragement provided for me, except by a few geeky teachers. If girls receive(d) approximately zero encouragement, that sounds like equality to me (though not how things should be!). The idea of geeks being cool, rich and popular is very recent, brought to you by the likes of Zuckerberg, Elon Musk, and Big Bang Theory.
I don't think the OP was only talking about female geeks but in general. Regarding bullying, I agree it's a big problem, however girls tend to have another issue to fight with, which is being told (by teachers, parents, and other geeks...) from very early on that they are not good at science/math. So imagine being bullied for geekiness and at the same time being told that you can't really do that think you like.
I know quite a lot of women (both in science fields and ones who did not go into science because of this) who told me that they were told this.
I have read it repeatedly on reddit/HN and others by people (in particular those who call themselves geeks/nerds) who argued women and girls are not good at science.
Also one thing I learnt when talking to female colleagues and friends, is that some things we just don't see/hear, because they are said in private, e.g. in teacher meetings...
Moreover, I know several women who were told by women, e.g. their mothers, that girls are not good at STEM. Just to say the discouragement doesn't just come from men.
_People are becoming afraid_
People are already afraid. Every other discipline outside STEM is completely rotten, it's clear STEM is next.
The quick publication cycle creates an environment that is always just about 'beating' the state of the art, but if you look closer into the reported values, you will often find a lot of questionable experimental choices. In one of my main application areas, viz. graph classification, almost none of the papers holds up (with respected to the reported performance gains) if subjected to a thorough experimental setup.
This creates a dangerous environment; in the worst case, we might miss some interesting contributions because they are drowned by the noise of reviewers (here we go again!) claiming that 'It does not beat the state of the art, so it must be crap'.
Could you give an example? Just being curious :)
For the GIN-ε (https://arxiv.org/pdf/1810.00826.pdf), for example, the authors report a classification accuracy of 75.9±3.8 on the PROTEINS data set (classical graph benchmark data set). If you run it with a cross-validation setup that is repeated to account for effects of chance, performance drops to 73.1±0.7.
Notice the drop—the second accuracy value is at least within the standard deviation of the first one, but you can see that a different experimental setup shrinks the gains quite a lot...
Same goes for different data sets. Since the gains are not super large for most papers, these changes matter a lot. But of course, the paper is now published, so no one is going to go back and change it.
FWIW: I like the GIN paper and think the authors did a good job. It's just that their experimental setup is insufficiently thorough for the data sets they are considering, thus leading to overoptimistic estimates. This is a problem because the next 'state of the art' paper has to find a way to get a slightly higher mean accuracy, at the expense of an even larger standard deviation, etc.
I agree the trend you point out is worrying. I suppose if this continues at some point the benchmarks are beaten, but that still tells us nothing about the true abilities of the tested systems or algorithms.
(this sounds more ominous than I intended it to sound; the reason is plain and simple that the publication is still under review and we have no preprint)
I got a research interest in GNNs and the datasets used in papers like the one you link, but I have so far only dipped a toe- because I have other priorities right now. But, more if I ping you :)
Insensitive according to who? The most sensitive 5% of people? All statements will be deemed insensitive by at least one person somewhere. It's silly to allow the most extremely (often unreasonably) sensitive people to set the threshold for what is sensitive or insensitive speech.
Well, that's one of the drawbacks of social media. Offended people can band together, amplify their voices, and spark nation-wide outrage. Whether the outrage is "real" or just "perceived" (i.e. the media says everyone is outraged so it must be so) is a different debate.
You can't plug your ears and say "it's just your training set" as a response to unfairness in ML algorithms. Real life is biased. Any real life data in our world is going to be biased. If you train algorithms on this data, they will cement any existing divides in society. So, with the understanding that researchers need to be more circumspect about ML algorithms than worrying about just the training data, consider that the upsampling algorithm in question only worked for white people because they fed it a huge amount of white faces. Claiming "it's just the training data" is one of those "well yes, but actually no" situations where ML researchers tend to miss the broader picture of how ML algorithms are used in real life, and just makes Yann look ignorant.
A model's only link to the real world is the training data, so saying it's sufficient to "worry about the training data" captures all the concerns we may have about bias, because from the model's POV there is no other relevant interface with the real world.
Saying "we need to do more" is devoid of meaning when by addressing the training data we are truly doing all we can as model builders and trainers.
A huge problem in the field is that we must use the previous benchmarks. This is because how do you know if the needle moves or not if you just change your data constantly?
So. In order to tackle this problem, someone with more resources than me needs to create training sets that are less biased. THEN, new academic papers need to benchmarked against the old biased sets, and also the new "less biased" (I don't think it's possible to ever get 0% bias, the world just isn't that clean) sets. And progress needs to be eventually transitioned to be measured on the new less biased sets.
The upsampling algorithm used pictures of celebrities. And the researchers put a blurb in their paper that was basically a "We know this is biased but everyone uses it so we must also". I feel like this is less useful science than an algorithm trained on more of a mix of actual real-world humans.
I admit it's quite challenging and probably impossible to do in some areas. I mean, how do you make a field whose end algorithmic goal is generalization, not use real world data to generalize people? But I think the issue can be worked on, and the need to use celebrity photos to train a set is a good place to start.
We'll probably go back to the 2000s model where you have to email the authors for code and data. The authors will delay by saying they are preparing it and then release it a few years later when it becomes irrelevant for public discourse.
Sure. Your comment's language equivalent is something along the lines of "Care to explain how words are racist?" Which yes, they are just a collection of words. They possess no consciousness and cannot be racist by themselves.
Similarly, a gradient is just a collection of vectors. It's just numbers. However, like language, it's what they represent that matters.
For example, I can create a machine learning algorithm to determine who should get a home loan. I create a gradient to optimize the algorithm to give loans to people who I think are unqualified.
The gradient can easily be racist if it optimizes heavily on something like race. Minorities tend to be lower income and so can be seen as less qualified as higher income individuals. However that's the easy argument, and also quite illegal. If you exclude race, there's 2nd degree variables that are proxies for race. Things like zip codes, job titles, whether they rent or buy. These are not explicitly illegal to filter on, though the end result is illegal if they exclude certain protected statuses. It can even be no fault of the researchers who implement the algorithm, because controlling for bias using real world data is extremely difficult. But we must do it, since it is the ethical thing to do.
And so, it's easy to see that one can optimize ML algorithms to exclude certain protected statues, which is what can make the algorithms racist.
There are English words that as pieces, they don't mean anything except their face value. I can string words together that mean bad things that are harmful to real humans.
Gradients are not racist by themselves, they're just math. It's like saying multiplication is racist.
But I can use multiplication as a tool in a chain to create weighted averages to create a naive Bayesean classifier to reject people for home loans.
And so too can I misapply gradient descent as a part of a larger ML model that is racially biased. For instance, I could choose a loss function that when minimized, gives biased output despite less biased input. Or, I could accidentally settle on a local minimum on the gradient in my model. There's many naive implementations of an algorithm that will just be biased no matter the unbiased inputs.
So in summary, a gradient is just math and is not racist by itself. It's being used in an algorithmic tool chain that researchers are frequently using which potentially will always produce biased output no matter the inputs (but more often than not also with biased input).
Even if you insist that a gradient or mathematical function is unbiased and can never have negative impact based on race or gender or other demographics, you have to explain any resulting negative impact somehow. Saying that the function or gradient is racially biased is a generous interpretation of the situation because it allows the creators to deflect blame towards an error in their mathematics or training set. If you insist on claiming that the training set and mathematics are infallible, one of the only remaining explanations is that the creator intended to discriminate. I'd rather not assume that!
If it was an experiment, then let it be. Perhaps the researcher was looking for something else, circumscribing the data, model, whatever to the experiment itself.
> researchers need to be more circumspect about ML algorithms
What does entitle you to tell what to study or how?
If the researchers created a toy, then great, it's a cool project and is a neat algorithm. But they didn't create a toy. It's an academic paper to attempt to move the needle forward in ML academia. And they are doing the exact same thing as a lot of other researchers, which is basing their research on old biased benchmarks. If the bedrock of the field is based on biased data and everyone builds on top of that, your research down the line will skew more and more in favor of the bias.
>What does entitle you to tell what to study or how?
Nothing entitles me. It is my opinion based on the facts in front of me. The ML field has a bias problem, researchers toss a "oh this is biased" blurb in their papers, and then continue using the biased data. Everyone looks at the cool demos, and then the research gets slurped up and implemented without regard to the science. More algorithms get based on previous biased algorithms.
They might have a reason. I can understand if they want to compare the result of the model with a past experiment. That's normal.
> attempt to move the needle forward
Completely agree, so just let them work.
By the way, I don't see "evil" in these experiments and I want a 100% free from bias model too, but I wouldn't dare to attribute the result to lazyness, stupidity or racism. If I come with something completely new then I would try to compare it with something that already exists too.
All very dispiriting, indicating real problems making progress in AGI beyond the clear lack of ideas.
Of course, just because these toxicity claims are being made, it doesn't mean they are accurate, but they certainly ring true.
If they are, it is good to know they are being talked about in some quarters.
In many respects I think it is more important that we have no idea what is true about any industry. When I dwell too much about my impression of any part of the tech field I usually end up demotivated to do anything that might connect me with the social group involved. I think there is a general shallowness and overcompensation in our industry that makes it impossible to figure out what real life is like with a group until you are way too invested and they have already hazed you.
it seem to me that a majority of the performance gains in ML are a result of using better hardware to run brute-force statistics with larger more complex models but the algorithms themselves have been improving at a nominal rate.
Going by the hype articles (which may be unrepresentative), we just seem to be moving faster and faster on an impressively powerful, but AGI-irrelevant train along a machine "learning" railway track and although I suspect plenty of people on the train would like to get off, the drivers and momentum are making that very difficult, as indicated in the OP article.
I'm completely optimistic about AGI, just think we are allowing the excitement of the advances in Artificial Unintelligence over the last few years erroneously dominate our thinking about it - at least in the sort of papers that turn up in tech-related feeds. Again, this may be unrepresentative of the top thinkers in computer science (machine-learning/whatever).
My own (layman!) opinion is that the good ideas have and will continue to come from external (or intersecting) fields, philosophy, neuroscience, etc; not computer scientists raving about the power of DeepWhatever using cloud-enabled networks.
Range: How Generalists Triumph in a Specialized World by David Epstein.
I'll check it out - thank you.
https://www.reddit.com/r/MachineLearning/comments/hkc697/d_r...
Yes, and the deep sea fishing, oil drilling and logging industries also aren't super diverse and have a severe lack of women. I don't see anyone complaining about that. Different groups have different preferences. I've met several women who have gone into science and IT only to find it immensely unfulfilling, as it's often a very socially isolating job by nature, and leave to switch careers.
If there's a rule or law preventing women from getting into the industry, then let us know and we'll change that. But don't criticize an entire industry because women on average chose to pursue other passions.
>At this very moment, thousands of Uyghurs are put into concentration camps based on computer vision algorithms invented by this community, and nobody seems even remotely to care.
What does he propose be done about this? Tell Chinese government bureaucrats to stop "stealing" publicly accessible research papers and code to implement tools that help commit genocide? Sue the Chinese government for violating licensing agreements that require "no violation of human rights"? Not everything should, or can be, about global politics. We should let people researching machine learning worry about machine learning, and leave the broader socio-political effects to political pundits and sociologists.
I personally don't really like this trend but I think our society is not ready to handle freedom yet.
If government money is being distributed, things become more interesting. But I don't think everybody is entitled to a career funded by tax payer money. So it should probably remain "cut throat" to get a good job based on government money. Whether governments have the best criteria for handing out money is another question (number of papers published might not be the best metric).
Sixthly, moral and ethics are set arbitrarily. The U.S. domestic politics dominate every discussion. At this very moment, thousands of Uyghurs are put into concentration camps based on computer vision algorithms invented by this community, and nobody seems even remotely to care. Adding a "broader impact" section at the end of every people will not make this stop. There are huge shitstorms because a researcher wasn't mentioned in an article. Meanwhile, the 1-billion+ people continent of Africa is virtually excluded from any meaningful ML discussion (besides a few Indaba workshops).
Edit: I think it makes sense to have a stance on Machine Learning moral issues in a machine learning group. I just don't think you need to have a stance on every issue ever.
I say this as someone always a bit frustrated that HN and Reddit etc are very US centric.
There's a lot to that and it goes beyond the scope of what researchers can handle by themselves. It's an effort that involves more than just the researchers, though they will take a central role.
For one thing, how do you evaluate whether a usage is "immoral"? How do you enforce correct usage? These are difficult questions with multifaceted answers which no one has yet elucidated (yes, it has to be more than "I'll know it when I see it").
Because its wrong to write-off the mistreatment of other human beings as an external issue. It never is: we're all human and we all share the same planet.
Eventually, but this is a unique time right now. People are paying attention to injustice.
I would say that much of these "stances" are inauthentic and done just for PR reasons. We'll see who is being genuine and who is phony soon enough.
Getting 10 people to agree a stance on the 10 issues facing you is much much easier than getting 7bn people to agree on the billions of possible issues in the known universe.
That's sort of my whole point: eithe limit the issues you have to take a stance on, or watch the amount of time taken up with stances grown exponentially.
Some communities choose to do so. Einstein, Szilard, Pauling (and others) formed the Emergency Committee of Atomic Scientists to argue against nuclear weapons at the dawn of the Atomic era. And many scientists have taken a stand against nuclear weapons since then.
Their actions have, at least, influenced the proliferation of these weapons and informed the public about their dangers. Would you argue that it wasn't "their business?"
One could make a case that machine learning has negative consequences/possibilities that are similar in scope to nuclear weapons. The people who work with and understand machine learning are in the best position to take a stand against abuse of these technologies. I would say it's their responsibility to speak out, otherwise a despot will do it for them.
How? The number of nuclear weapons was growing rapidly until the 80s, to the point where dozens of bombs were aimed at single targets.
> informed the public about their dangers
Seeing the effects of bombs informed the public, not a committee
It's hard to assess their impact, but that doesn't mean it was insignificant. Could we have been in a worse position? Definitely. Someone might have "pushed the button" by now were it not for decades of work from organizations (some of whom consisted of scientists) lobbying against nuclear weapons and their proliferation.
> Seeing the effects of bombs informed the public, not a committee
The nuclear tests demonstrated that the weapons work. There are other effects such as "nuclear winter" that have not been (and hopefully never will be) demonstrated. Sadly, because of Chernobyl, we do have examples of what happens to contaminated areas. Understanding and communicating the effects beyond "blast radius" is best done by scientists and health experts.
Technology is the ultimate medium now. Technologies like machine learning shapes discussions like money and GDP figures continues to shape discussions about the state of societies the world over.
Ignore the strategically important position that technology holds now and in the future of our society at your own peril. To be so willfully blind to the changing times, I can’t imagine a less engaged worldview.
Note I do not support tossing ml researchers into the sea and lots of industries are worse.
I saw this repeated frequently in the last few days, and I think I've missed some important info. I know that China uses face recognition for surveillance purpose, and that it detains a substantial share of the Uighur population for supposed violations of the law. I miss a link between the two: is China using AI to detect who is Uighur and arrest him/ her for this very fact? Does anyone have more info?
Despite the scale- which we considered worrying in itself and for understandable reasons- this seems to be an automated version of "I suddenly see a lot of mafioso-looking guys congregating here, let's keep an eye to see if there's anything dodgy going on"- which has been practiced with skilled eyes since forever. And it's a whole different thing from "if I see anyone looking Jewish I'll round them up and send them to a concentration camp".
That's a weird suggestion. It's undisputed that they have put millions into detention camps, so are you suggesting they all have been engaging in "terrorist activities"? So why suggest ("suppose" ) that they have been engaging in terrorist activities?
> Despite the scale- which we considered worrying in itself and for understandable reasons- this seems to be an automated version of "I suddenly see a lot of mafioso-looking guys congregating here, let's keep an eye to see if there's anything dodgy going on"- which has been practiced with skilled eyes since forever. And it's a whole different thing from "if I see anyone looking Jewish I'll round them up and send them to a concentration camp".
The article literally said they are looking for (tracking) Uighurs because of their ethnic appearance. That is looking for "Jewish" , not looking "mafioso-looking" in your example above
No, the article talks about "tracking" Uighurs, not rounding them up. The Uighurs are 25 million and mostly live in the Xinjiang region, so there is no real need of AI to round them up. If what you want is to find them and imprison them, you can just go in any street of Kashgar: they make up 5/6 of the population.
What the linked article says is:
"The facial recognition technology, which is integrated into China’s rapidly expanding networks of surveillance cameras, looks exclusively for Uighurs based on their appearance and keeps records of their comings and goings for search and review. " (Italic mine)
> The article literally said they are looking for (tracking) Uighurs because of their ethnic appearance.
No, the article says they're tracking Uighurs through their ethnic appearance. Not "because" of it.
> That is looking for "Jewish" , not looking "mafioso-looking" in your example above
The point is whether you're looking for a visible trait because that is in and by itself the fault (and so after detecting it you immediately proceed with an arrest), or because it is somewhat predictive of something else, and it merely helps in narrowing your search.
1. People getting into positions for the wrong reasons (diversity hires), 2. No one has any kind of backbone any more... victim culture and molycoddling shows it's ugly face, 3. Real smart people that don't' want to content with stupid getting rightfully angry with their subpar collegues who are desperatly trying to being somewhat relevant.
Let's face it... scientists are most of the time not people persons. Most smart people don't care for your snowflake persona getting hurt by the truth. It's basically impossible to do antying in an environment that treats microaggressions as anything else but the completely infantile bullshit that they are.
GROW A FUCKING SPINE AND GROW THE FUCK UP WHINERS.