Facebook AI Research Expands with New Academic Collaborations
code.fb.com
code.fb.com
1) Much less, if any, grant writing. A standard NSF $500K 3 year grant provides a lab about $100K per year (due to university overhead). This will support 1-2 PhD students on a pittance of a stipend.
2) Less teaching (1 course once a year)
3) Computing resources that are far greater than you could get a grant for. One NVIDIA DGX-1 is about $100K. Very hard to get a grant for something like that. Companies have far greater resources than just one DGX-1
4) Access to software engineering personnel to help with research
5) Access to data, and the ability to create new datasets. Many datasets cost over $100K to make (e.g., those for semantic segmentation). Very hard to get a grant to make a dataset (I've tried)
6) Much better media relations to popularize your research
7) Bigger salary
A big part of this, in my opinion, is just that FB funds their labs. Getting government funding for a lab is difficult and frustrating and must be done continually, especially for these professors that run huge labs. NSF has a 7-15% success rate, and it can take a lot of effort to write a decent proposal. Including teaching regularly, this takes a lot of time from actual research. What these professors, I think, are negotiating is only teaching one semester a year and having a spigot of cash to fund their labs. They also gain access to massive amounts of data and computing resources.
Other companies don't seem to be as open to this arrangement as FB, but it has a lot of appeal to me (although I'm not a lover of FB).
I don't know how it compares in software, I worked for a professor in undergrad who was in microelectronics. I was astonished at how much of their time was taken up by grant writing and applying for funding. The pressure for research professors to support their position was immense.
There's a lot of reasons to hate FB, a lot of reasons to be wary of what comes out of their research. But to castigate researchers that cash-spewing tech titans create a better environment and experience for researchers than the public and universities feels a bit asinine. I've known too many starving grad students, post docs, and non-tenure professors to prefer people subject themselves to the system we have rather than what a company like FB wants to provide.
1. Industry funding comes with its own set of hassles. And overhead is still a thing for industry funding - sometimes the overhead rate is less, but there's been a lot more pushback against this from universities (which I actually think is quite fair).
2. Being able to buy out teaching loads is a function of funding, not necessarily this funding
3. In many institutions, salaries are fixed - what this does is cover a percentage of your salary you don't have to cover from other mechanisms, or let you buy out teaching, as you note. It would be interesting to know if the dual FAIR appointments work like that or not.
While private funding is awesome, and these labs are in a very good position, it should be noted that in most cases in my experience between NSF/NIH-style funding and an industry grant, the government grants last longer, are more flexible, and less of a hassle administratively.
For instance, I assume that grant writing is such a hassle, the average person would accept just to get out of writing grants. But the guys and gals that they picked up are at the top of their field. Would they also have the same constraints? Seems like they would have higher hit rates, less teaching and so forth.
I think the ability to launch the careers of their grad students, get them hired etc might be at the top, but I am just guessing.
There is no free lunch in raising money for research.
I would categorically reject any collaborations with FB as an academic in ML.
> But can they get the huge volumes of data they have tagged out? The real time pipelines and internal tools? The community of people they will be working with?
It's highly problematic when a researcher pushes ethics aside in order to gain access to data and chase "long run ... scientific advancements."
See https://en.wikipedia.org/wiki/Fruit_of_the_poisonous_tree
That’s really the main question, isn’t it?
i'll be waiting for the scientific advancements to be useful to the public. so far they've increased rates of depression, anxiety, etc while enabling totalitarianism.
It's best to think of Silicon Valley as two entities: a mass of technology workers who build software, and a financial extraction function that attempts to extract value from the work they do.
I see Facebook as the least ethical, and least useful from a civilization standpoint of all the big tech firms.
Google is driven by the same ad-clicking incentives, but the one-tricky pony has been developing other extremely societally useful tech, like self-driving cars and other moonshot projects.
Apple and Microsoft sell products, they do not make users the product (on the whole). Together they pioneered computing revolutions, and I'm confident history will judge them for making a positive contribution (on the whole).
Amazon is a leviathan whose societal value I find more difficult to classify, but I genuinely derive lots of value from their service personally. It's good for my lifestyle.
Facebook on the other hand, is a waste of my time, mental energy and a drain on society. As an academic, how can you turn your mind to furthering its goals?
On a parallel line of thought, I prefer FB's React to Google's Angular. In both React and Pytorch I see the same elegant design. TensorFlow and Angular on the other part are unnecessarily complicated.
That's historically true of lots of research innovation though.
War (generally accepted as bad thing!) has advanced technology and civilization repeatedly.
At least the scientists aiding war (on average) had some awareness that killing people is clearly not a good thing, and at absolute best a necessary evil. Can the same be said of the people at Facebook?
The culture of their management is to be in denial about how damaging their service is to the mental health of individuals and society.
When you talk to these guys (they're almost all guys), you realize they're fully aware of what they're doing, and in the back of their minds they know FB is evil. They just like the money too much.
This is 100% about money trumping conscience. FAIR researchers may be millionaires, but they're ethically challenged. I wouldn't trade place with them. These people disgust me.
Interestingly the head of the Pittsburgh lab here isn't even an ML researcher -- she is primarily known for her motion capture work and for running Disney Research.
I highly doubt most of them believe that FB and the work they’re doing is evil. They would have no incentive to work there if they did.
However, the fundamental purpose of our products is to allow people to efficiently communicate with each other. Hard for me to square that with "drain on society." I have many friends who, via Facebook, found a connection that was life changing: from finding a job, a spouse, to a community to deal with the loss of a loved one or support after being diagnosed with a terminal illness.
One of the things that draws AI researchers to come work at Facebook is the opportunity to see their work make a positive impact on billions of people around the world.
The research done by FAIR is helping us do things like deliver billions of translations a day, provide automatic photo captions for people who are visually impaired, and help bring blood donors and people in need together. It also helps us spot when someone is expressing thoughts about self-harm so we can alert first responders.
But we also believe there's even more we can do to help bring the world closer together, to give people a voice, and to open up new opportunities for everyone. AI is a key part of that and we believe pretty deeply in the power of open research to help not just us but the whole industry.
You talk about impact. There's no doubt FB is making a big impact. Unfortunately, it's overwhelmingly destructive impact. As someone in position to change that, it would be great for you not to dismiss out of hand the valid concerns of the people in this thread. Personally, it's because of replies like this (in particular zuck's attitude) that I have zero confidence in FB's potential to fix its products in the future. Bye bye democracy I guess.
One day you may be held responsible for your impact on the world. I hope the talk about making the world a better place will work out then.
All for the purpose of increasing buy-in to an increasingly Orwellian digital surveillance regime.
> But we also believe there's even more we can do to help bring the world closer together...
What brings people closer together is real human interaction and connection. Face to face communication with visible emotion. Vulnerability. FB's video chat is the only thing serving that interest, but that's better served elsewhere with less tracking. Posts that broadcast one-way to an invisible audience are inhuman. Filter bubbles are toxic. Widespread use of FB is cancerous on the social fabric of society.
Tell your PR team the appeal to emotion was a nice touch. If I didn't know anything about your company I might have even been able to get through it without feeling absolutely nauseated.
> the fundamental purpose of our products is to allow people to efficiently communicate with each other
No, the fundamental purpose of your products is to efficiently surveil, profile and manipulate your users on behalf of your customers.
> It also helps us spot when someone is expressing thoughts about self-harm so we can alert first responders
What happens when those first responders bust someone's door down and your "helpful" feature essentially becomes algorithmic swatting?
On a semi-related note, did you guys ever figure out how many of the hundreds of thousands of people you enrolled in an emotional manipulation study without their consent ended up killing themselves as a result? It's a given that the figure isn't zero across that number of people.
But it's an abuse of power to only look at one side of the equation.
Companies are run by people, and at the end of the day, no person would want to dump their money into a technology that couldn't yield any financial gains. Where do all the large companies get most of their revenue from? From figuring out how to trigger dopamine to be released into our brains, and we're starting to see the negative effects it's having on people.
Also, you aren't giving people voices. You're opening up a door to a world where they have no control over. In their outrage, and futility, they focus more and more time trying to fix something that doesn't exist.
Obviously, in my personal opinion and I can understand that a lot of people don't care about it
Mark hasn't taken back his comments about his users being "dumb fucks" for "trust[ing]" him, as far as I know, although he has apparently said that he regrets saying it[0].
[0] https://www.newyorker.com/magazine/2010/09/20/the-face-of-fa...
Source: research and I've asked FB myself
What's the difference between this and taking it back?
Would you take that offer? Would I? Probably not. But I can see the appeal, and plenty of people wouldn't hesitate.
It will keep happening until these companies are a black mark on your CV, like say a tobacco company might be. Will people be so eager if it means they will be shunned by the wider research and engineering communities?
The question as you so rightly point out is whether there is a net value added.
If we took their AI contributions and JS frameworks on one side of the equation, do you really think it balances the other side of the equation?
On that right side lies encouraging general disinformation leading to broken elections and even aiding genocide. Academic studies on happiness show using Facebook and Instagram correlates with poor mental health; that research has been replicated.
What product actively damages those that consume it?
Facebook is the digital equivalent of the Tobacco industry; good business that's bad for people.
They will work part-time for the university, and part-time for Facebook. I don't see why that's so shocking? Presumably the university only funds them part-time for part-time work, so it's not like they're not getting what they're paying for if that's what you're asking. This is a very conventional set up for academics.
And there, there's the clarity of "I'm an employee of the university". With dual appointments...the agreement has to be complex. Either that, or really draconian based on Facebook throwing it's weight around and saying "Accept this, or we'll find another university that will."
> Often, these companies [like FB] are open to research partnerships and things, but it’s always on their terms. If you do research with them, you’re dealing with IP issues, you’re signing over the rights to the research. It has to be reviewed completely and vetted by their legal process. They often handpick researchers that help them and help their purpose and help their cause — they maybe throw in some sprinkles of criticism. I understand why they would be hesitant to want to work with people like me.
[1] http://www.niemanlab.org/2018/02/news-in-a-disintegrating-re...
IDK. Maybe -- just possibly -- there do exist researchers who have their choice of funding spigots and are choosing to work with Facebook. Either that or HN has some damn high standards for what it means to be "worth their salt".
There's nothing intrinsically wrong with industry collaboration, even when it involves companies whose impact on the world you might not like. The big oil companies are, unlike FB, an actual existential threat to humanity. But I wouldn't fault renewable energy researchers for taking research dollars from those companies.
The question is: will the funded research agendas push science forward in the direction it was headed anyways, or will this money distort the type of research being done?
In any case, in a week we'll be back to our regular programming bemoaning the fall of the industry research lab and the paltry salaries offered to phd students...
With them being part-time professors what are their university salaries like?
In total compensation between industry and university I'd imagine this would be a good deal.
Lastly, how does recruiting work here? I assume they don’t make these people go through 5-6 algorithmic interview rounds.
> questions testing that you have basic knowledge of your field
Not directly, but in discussing my research the interviewers drilled down into the contents of my papers, etc.
Reminds me of how on the one hand it sucks that certain talented engineers/designers work for ad-driven companies, but at the end of the day the compensation is simply on another level compared to working for more benevolent causes.
U Montreal has the largest AI program going.