Keeping Top AI Talent in the United States [pdf]
cset.georgetown.edu
cset.georgetown.edu
The current situation of making it difficult for recent graduates to get green cards could put pressure on retention (having students staying in the US). At my last job I tried for about a year to replace a deep learning engineer after someone on my team left for a dream job.
Currently many white collar jobs (like Wall Street analysts) are getting eliminated by deep learning and other models automation. I predict that it will be 5 to 7 years before most routine deep learning development will be largely automated, but until we achieve this, it is really important to have trained people to do the work. That said, top level AI research will not be automated in the near future.
The best deep learning engineers with infinite hardware built Amazon’s recommendation algorithm, and it’s largely trash. I can’t imagine it’s magically thousands of times more capable for these other tasks.
What is trash for you may not actually be reflective of how good the algorithm is. My experience is that Amazon's algorithm actually works somewhat well.
Recommendation algorithms usually work well for people who are closer to average (i.e. their preferences being closer to the aggregate). People who deviate from the average tend to get poorer recommendations. It does even worse for segments that deviate so far from average where there are few sample points.
Recommendation algorithms optimize for aggregate conversion -- if it can get the most number of people to convert, it is good (for the merchant) even if it looks like your personal recommendations suck. It's all a matter of perspective on what "good" is.
Is this "reading lots of company reports and extracting ten key features" or is there something else ?
The reasons for this should be quite clear. It isn't appropriate for this kind of work. Pretty much all financial problems are insanely noisy, and DL is shit at dealing with any kind of noise, both in labeled, unlabeled and reinforcement configurations. Trading problems on short time horizons can harvest innovations vastly faster than DL can keep up using what amounts to Kalman filter like ideas. Long time horizons don't have enough data to fit a meaningful DL model even if 90% of the spectrum weren't noise. We're not talking something like automatic translation where you basically have all the clean data on the internet to fit to. The only place I could imagine it being used at the two-sigmas of the world is in natural language processing, but since you are asserting something completely different, that's irrelevant.
R&D departments, of course, fool around with it. And fintech companies are definitely implying they're using it because ... everyone else is. IRL they're all doing some kind of Markowitz MPT with factor models spooge. If you have any actual evidence that there is some large scale automation of wall street analysts facilitated by deep learning: I'd love to see it. Marketing submarines riding the DL hype chuckwagon are not evidence.
This idea may have some merit, but I want to be very blunt about this - PhD programs aren't doing a good job attracting and retaining good students, at least not the ones who aren't bound by US residency and work rights in their choice about what sort of degree to pursue. Elite Law and Medical schools typically have an attrition rate of below one half of one percent, <0.5%. For PhD programs, it's typically 35%-50%, and while there's less data for MS programs, we're still looking at 25%. Sorry no cite, I documented this a few times in comments, I may go back and get it later.
On a personal level, I was a doctoral student in an engineering program at Berkeley, and I will say with full confidence they couldn't give less of a fuck if a student fails to get the degree. I don't swear on HN much, and if you look at my history, I hope you'll see I don't like to troll. But in this case, I'll repeat it, they could. not. give. a. fuck.
Give these people power to bestow the right to reside in the US, and I assure you, the'll give even less of a fuck if anyone is happy in their program, or if their PhD or even MS programs are competitive with MBA, Law, Medicine, and other programs that require attracting and retaining people who already have the right to live and work in the US.
I still think it’s good to try to keep people who get PhDs, especially in areas I think are important, like machine learning research.
But I agree that making a PhD an even more useful credential is unlikely to increase the quality of the PhD experience itself.
I an in favor of immigration. I am not in favor of giving PhD programs power over the system so they can remain uncompetitive with the options free people have when choosing a degree or path in life.
I think we have general tendency to underestimate what hard working people are capable of. All we need to do is guide them in the right direction and make sure the majors required are available.
>Or, in other words: there were as many American citizens getting CS/EE grad degrees in 2016 as there were in 1990.
The push for STEM has been incessant for extremely real concerns. Our population is just. not. entering. the STEM graduate courses. Yes, this needs to change, but it's a cultural problem, not an availability problem.
I'll never forget the day in lab meeting when it came out that I was a US citizen. I thought my accent and appearance would have made it obvious, but evidently my coworkers had assumed otherwise, and the response was an astonished "Then what the hell are you still doing in academia?" followed by looks of pity when I admitted to loving research. This was a biomedical lab at Carnegie Mellon about a decade ago and I was the only one doing ML work, so their astonishment was in part due to an underestimation of the market value of my degree, but on the other hand their intuition was likely accurate with regard to the STEM PhD market outside of AI.
Those PhDs are worth $pittance + green card, but not worth $pittance.
I trippled my compensation by deciding to work in industry.
And not only that, but the push for STEM imo has more to do with lowering wages and getting more software engineers (especially for S and M, I don’t think there is any real critical lack of PhDs. In fact a very large amount of physics and math phds just go into data science or software engineering). It’s entirely an economic problem, if it’s even a problem at all.
I was an undergrad in the late 90s, and even then there was no point. It made sense if you wanted to pursue an academic career (barely) and it made sense for immigration purposes. Personally, I’m an advocate for very open immigration, instead of creating weird niche paths to citizenship.
Probably because they import so much cheap labor it's a lot easier and more sane/lucrative to do something else.
Check out the list of names under departments with publications in top AI conferences in the last three years. A large percentage, maybe even most, were not born in the US.
http://csrankings.org/#/fromyear/2017/toyear/2019/index?ai&v...
(Although the number of publications in top conferences is not a perfect measure, it is a pretty good one and correlates with most other measures one would think of as important.)
The world is also catching up. Among the “top 20” departments, according to this measure between 2017-2019, 10 are in the US, 9 in Asia (7 China, 2 Singapore), 1 in Israel.
Also note how many top researchers in sciences and engineering, not only AI, who helped build the US into the technological powerhouse in many fields, over the last century, were born in another country. It is not a coincidence. World population is 7.8 billion or > 20 times US population.
.
Another major issue is math skills. Given that there are many technological specialties that require excellent math training and aptitude, it is possible that there are too few people growing up in the US who received that training and supportive environment during their formative years.
A conversation among many Asian students is how easy American math curriculum is. It is estimated to be 2 or 3 years behind the same grade level of many countries in Asia. The PISA 2018 results also show some concerning evidence:
https://www.oecd.org/pisa/PISA-results_ENGLISH.png
https://hechingerreport.org/what-2018-pisa-international-ran...
“In math, the U.S. ranks 36th out of the 79 countries and regions that participate in the test.”
“Today’s [US] scores in reading and math aren’t statistically different from when PISA started testing the subjects in the early 2000’s.”
I would find 10% to be wayy over-estimated.
If there's some geopolitical reason that it's a good idea for the government or someone to offer them free training programs, I'm all for it, but I'm not understanding the moral argument.
You want a network effect to collect talent from anywhere & then tax it. Children growing up in this culture will be more prone to tap into this. Immigration laws insulate this process. Eventually the talent will be collected somewhere less insulated, & then the American talent will be moving there
If I want to steal your jobs, I don't need a visa. I do it well enough from Vancouver, along with all the other employees who are in between visa issues. Canada will accept our taxes in the meantime
Solution is to improve teaching of STEM subjects at primary and high school level. The problem will sort itself. Second option is protectionist in nature. Favour US residents ahead of international students, even when international students are way better. Second option will only drop the quality of PhD graduates coming from those colleges.
How did you come up with this number and what reasoning and data can you provide to rationalize it?
Actually, training and allowing these people to work (in potentially very advanced technologies) is ANTI-security.
But it is monetarily beneficial for the shareholders of companies who will benefit from the research conducted by "cheap but smart Asians". Fundamentally, this article should be considered treasonous because it is anything but pro-American.
We have a massive issue in this country of certain elites feeling entirely negated from any responsibility towards their fellow citizens.
> That appears to be how we built this country.
You do realize that the Native Americans would consider this a very racist remark. There did exist advanced civilizations in the Americas before "immigrants" came to try to rule it all.
Also, the slaves that helped build this country were not brought here because they were well-educated abroad and driven.
And the poor Chinese that helped build this country were not well-educated abroad and driven to build railroads elsewhere in the world.
This country was built by driven people, many of whom were very intelligent, but this simplistic view you are proposing is not only not true it is actually degrading to all those in the history of this country that contributed a lot without being considered "smart" by the elite.
Perhaps, we should avoid overly simplistic conclusions and viewpoints?
Ok, at what salary?
At minimum wage, lots of demand, not much supply. At $500-$750k a year, lots of supply, considerably less demand.
I would propose highest salaries in the world as that incentive. There is nothing fundamentally superior about American culture or politics to many other countries, so that seems like the most reasonable conclusion to me.
Thus, do you really need to worry about retaining talent after you have assessed the true underlying incentive? The simple answer is no. Americans getting paid more than anyone else, game-theoretically, guarantees the smartest people will not only try but absolutely find a way to live and thrive in America.
Speaking of incentives, if we know that is the main driver of smart immigration then one can wonder what is the "true" incentive behind published "research" like this.
For example, conclusions like this:
>U.S. policymakers can ... control whether the United States remains an attractive and welcoming destination for international talent.
are simply not true (at least in attracting the truly best and most desirable candidates).
US essentially has an H1B visa quota of 85k per year, with 65k being the regular pool and 20k being the advanced degree pool. It is always advantageous to file for the advanced pool if you can, because if the quota for the advanced degrees got filled for the year, the spillover people (i.e., the ones with advanced degrees who didn't get their visa approved yet) get put in the regular pool for that specific year.
https://redbus2us.com/all-about-h1b-visa-cap-basics-regular-...
“A person who has obtained a U.S. Master’s degree or higher qualifies for this cap quota.”
A foreign advanced degree doesn’t qualify.
People always talk about how much they hate corporate ladders, but honestly most will happily accept a new position with a shinny title
In the absence of further information I hesitate to dedicate funding preferably to "AI" based on faith alone. I might as well spend the money on cheaper and more reliable power turbines for remote wooded areas in California or finding a clean way to mine and convert coal to energy. Maybe the money would be better spent upgrading the health-care system or replacing our Navy with something more likely to survive a flurry of cheap cruise missiles.
There are an infinite number of ways to spend money and "keeping top AI talent" is far down my list.
Top AI talent seems to be too narrow in this example.
China? They seem to get it. Here it's a bloodbath to raise a dime. For money is our new God. I'm not sure China is at that point yet.
Learning Mandarin. :-)
I work with a PhD in mechanical engineering, who shares my views about the US not being a Long term place to stay at ( at least based on current status).
The rest of the world used to see the US as a leader, as a peace keeper. Now, not at all. :(
https://immiguides.com/immigration-guides/singapore/
- Singaporean citizens of Chinese heritage tend to enjoy preferred treatment over their Malay and Indian counterparts. Examples of overt discrimination include employment ads specifying that someone must be of Chinese origin to apply.
- the Singaporean government censors sexual, political, religious, and racially sensitive material. Lawsuits are often used to quiet dissidents. Journalists often self-censor, to limit their legal liability. This also limits the development and prevalence of non-commercial art
- members of the LGBT community face discrimination in Singapore. Male on male sex is illegal. It is difficult for members of the LGBT community to find housing (landlord discrimination).
- punishments commonly involve caning and long prison sentences.
Obviously the U.S. has problems, but on the whole it is pretty great. I’ve traveled quite a bit and lived in other countries, and some are better in some respects, worse in others by a large margin.
People complain about police at schools in the U.S. (and it’s fair) but I don’t feel comfortable in Paris walking by a Jewish school that has soldiers with rifles standing outside either. YMMV. ️
At some point in the past few decades it became de rigeur among a certain fraction (eg those in this thread) of the US population to shit on their own country at every opportunity, to distinguish themselves from those flag-waving yahoos in the other tribe. Patriotism became heavily tribalised, and saying good things about America in certain parts of the country gets you ostracised as one of those uneducated Republican types who is too stupid to see that absolutely everything is so much better in every other country and if it isn’t then it doesn’t count.
Combine this with the US saturation of global media, and you can turn on the TV in every country and see Americans shitting on America. Eventually the message starts to sink in.
People in America need to get more perspective, and quit letting every single thing that comes out of their mouth be determined by some boring red vs blue conflict.
https://edition.cnn.com/2019/12/17/us/north-carolina-student...
Agreed that it would be better if it did not take soldiers standing outside of schools to protect the Jewish community but at the same time doing nothing wasn't an option and there has - as far as I know - so far not been any harm to bystanders on their behalf.
Most of the people who think the current political or social state of the USA is horrible have never opened a single book about asian, african or eastern europe policies.
While everyone is formatted to be shocked by what is going on in the USA, they refuse to even see that torture (including citizens) and hardcorde censorship is routine and legal in a lot of asian countries.
To quote John F Kennedy:
“I have been guided by the standard John Winthrop set before his shipmates on the flagship Arabella (sic) three hundred and thirty-one years ago, as they, too, faced the task of building a new government on a perilous frontier. "We must always consider", he said, "that we shall be as a city upon a hill—the eyes of all people are upon us". Today the eyes of all people are truly upon us—and our governments, in every branch, at every level, national, state and local, must be as a city upon a hill—constructed and inhabited by men aware of their great trust and their great responsibilities.”
Not to defend this, but I expect to see this among smaller employers in US tech hubs in the future.
Anecdotally, I know two PRC nationals who spent two years in Singapore working for employers that discriminated. They had nothing but bad things to say about their workplaces when it came to code quality, leadership, and how women were treated (both of them are female).
I know it's a sample size of two, but it makes me think that you wouldn't want to work for an employer that favors Chinese applicants as anything but a last resort.
Keep in mind that these companies are often fulfilling three things simultaneously: allowing a PRC national to gain citizenship in another country through an avenue that involves significant financial investment, laundering/protecting RMB (often the same thing in China), and, finally, functioning as what most of us consider a viable business (turning money into more money). These are listed in order of prioritization.
It's ok if the business fails completely and never turns a profit, as long as the first priority is accomplished. Accomplishing the first and second is ideal. Ending up with a viable business and a financial return on your initial investment is just a nice-to-have.
The stories in the media are always tilted towards outrage, fear, or hate because those base emotions are what drive clicks. The reality on the ground is much different. I often travel for work so I spend time in lots of different major US cities, they are clean, have low crime, and American's are generally very friendly.
Granted I don't travel to the worst parts of major US cities, but I've travelled enough places in the world to know that in some cities all the parts are dangerous, and we're a long way from that.
Not sure what made USA change for the worse or if it has just been a slow decline.
Some of those countries are attracting previously unheard-of talent and subsequent funding from the US. It's hard to imagine how future US startups can ever compete.
I also mean entire marketplaces, solutions and systems that the US tech industry is now having to copy from somewhere else.
Meanwhile there are many, many ways other than AI to spend our dollars for both research and education: healthcare, energy, climate change, a more modern military, etc.
We tend to treat the term "AI" as if the technology were already true artificial intelligence but it is little more than a set of pretty good recognition and classification algorithms. Investing billions in a teechology prematurely is always a mistake since it will divert those billions from other uses. But I see no technology remotely approaching the intelligence of a 6-year-old child even on the horizon today. Now I'm not saying that we couldn't use a zillion 6-year-olds (after all, Victorian England did so with their workhouse system) to do useful work, I'm just saying that "AI" (the poor version, not artificial intelligence) is very likely to become a financial sinkhole and cause possibly decades of lost effort if we get carried away by mistaking a term for the real thing:
ARTIFICIAL INTELLIGENCE MEETS NATURAL STUPIDITY Drew McDermoH MIT At Lab Cambridge, Mass 02139
https://homepage.univie.ac.at/nicole.rossmanith/concepts/pap...