This sounds like a direct response to two things (mostly): 1) CMU establishing a 'department' of AI (where MIT had merely an AI track within the CS program), and 2) Kai-Fu Lee's recent book on the rise and inevitable domination by China of all things AI-related -- inviting a new 'space race' between the superpowers. (A 'brain race'?)
Yet I can't imagine why an entire 'college' of AI is needed. AI simply isn't a field that's deep or broad enough to warrant an entire college with a handful of distinct majors, like an engineering college or medical school. Each of this college's AI degrees will span distinct problem or solution spaces? Not likely.
Maybe this was the only way to ensure the gift of all $350 million. Or to build multiple new buildings...
MIT's news article [1] has more concrete information. The NYT article is a bit misleading.
It is a College of Computing, not College of AI. Also note that MIT has Schools (School of Engineering, School of Science, etc.), not Colleges; so this will be something different.
For example, according to MIT's FAQ [2], EECS Department will likely continue to be part of School of Engineering, even as it becomes part of College of Computing.
In particular:
> The College will reorient MIT to bring the power of computing and AI to all fields of study at MIT, allowing the future of computing and AI to be shaped by insights from all other disciplines;
> Q: Why is this a college, rather than a school? What is the difference?
> A: The MIT Schwarzman College of Computing will work with and across all five of MIT’s existing schools. Its naming as a college differentiates it from the five schools, and signals that it is an Institute-wide entity: The College is designed with cross-cutting education and research as its primary missions.
> Q: What kinds of new joint academic programs or degrees are envisioned?
> A: MIT has been making progress in this direction for some time; for example, we already offer undergraduate majors that pair computer science with economics, biology, mathematics, and urban planning. The MIT Schwarzman College of Computing will allow MIT to respond to the student demand the Institute is seeing in course and major/minor selection more effectively and creatively. It will enable MIT to pursue this vision with unprecedented depth and ambition, and will give MIT’s five schools a shared structure for collaborative education, research, and innovation in computing and AI.
AI isn't a method, it's a field of trying to solve problems through an almost pure computational mean. People tend to think of it as machine learning, but there's a whole range of mathematics and computer science related to solving these kinds of problems. In this sense, PDE-constrained optimal control of power grids, parameter estimation of pharmacometric models, robotics and automation, and bioinformatic advances like automated cell detection for enhanced next-generation sequencing techniques can easily fit under an AI umbrella. That makes it a huge interdisciplinary effort that can easily span a whole college, and something MIT is very well primed to succeed in.
MIT was one of the earliest leaders in sharing course material online for free.
I wish they would take that $1 billion & invest it in modernizing online education. If you want to be a leader in AI, I argue you should open your doors to as many applicants as possible from all over the world at an affordable price. Discover a better way to improve collaboration & online learning. EdX & Coursera are nice but they seem to be halfhearted attempts. UNC's online MBA requires video conferencing for discussion & is much more engaging.
I guess $1 billion of server infrastructure & employees doesn't look as pretty though.
I think if you want to be a leader in AI research you want to attract and nurture the best graduate students and researchers. I don't really know how much doing good research has to do with good online education. I am sure online education is useful but these issues don't really have much in common.
I believe that there are a few universities whose education is by far superior. If you can increase the amount of people receiving top education, you can increase those that can then go on to do research.
Getting into MIT is not easy. I imagine it is especially hard for those outside the US. The gatekeeper effect lowers the amount of people who can go on to become graduate students & researchers.
ehhhhhhh I am not so sure the main limiting factor in going on to grad school is access to quality undergraduate education. a diligent student at the top public school in every state is probably qualified to go do research in graduate school.
I'd imagine a more limiting factor is the a) willingness to work really hard for 6+ years for uncertain rewards for the joy of research with very low wages. especially when you can go into industry and make 100k+ b) student loans-see low wages as an academic.
As an academic in ML at least-I think there are more than enough academics in the field or trying to get here....look at NIPs submissions!
Agreed, collaborations and publishing innovative research/experimentation go much further than trying to do a course simplified to the level that it can be taken online.
I view the online education/MOOC type stuff being for teaching the basics. For example, have the undergrad curriculum be online and basically free. If people want to go to grad school at MIT or wherever, then have them take an extremely hard exam in the subject. I think overall, you would get higher caliber candidates this way just because of how big online education can scale.
The current education system is highly exclusionary based on characteristics that are obtained in high school and most of the characteristics are directly linked to income. The problem is many people can still go on to get these skills later in life but can't really get into MIT once they're adults. You can, in theory, but we all know in practice it is not realistic.
The big universities will NEVER give out high quality equivalents of their courses online. The problem is the undergrad tuition is helping to subsidize the higher-level research that MIT and other big name schools are known for. How many times have you heard the stories about how undergrads are being taught by TAs or graduate students because the professor is off doing some research.
MIT and the other schools know if they gave away their core curriculum's online that are basically the same thing then less people would want to go to the school at the current over inflated rates. If you guys are looking for a revolution in online education, I hate to say it, but don't look at traditional institutions to do it.
The revolution in online education will be done by someone similar to a Steve jobs or Jimmy Wales (wikipedia) that basically has no connections to the traditional education industry but is just very motivated to change the world for the better. We all know that teaching is not rocket science (many people can teach it) and most of the significant human knowledge is being written down in books. The only exception is for the stuff that is cutting edge latest research but of course people aren't going to learn those topics until they've learned all of the other known subject matter on the topic which is recorded in books. So overall, the goal here is to simply take those books, produce free equivalents of them (the knowledge in the books are NOT subject to copyright) and then create some sort of online self-study system where people can learn the material. Tools need to be created that help people to learn the knowledge on their own and offer innovate tools to self-study.
> The problem is the undergrad tuition is helping to subsidize the higher-level research that MIT and other big name schools are known for.
Actually MIT undergrad tuition is about 14% of MIT's revenue and about 16% of expenses (all in, not just the teaching portion), numbers which have been remarkably stable over the last 30 years. Undergrads simply aren't that important to a major research institution like MIT which is better thought of as a big research lab with a small school attached (undergrads make up less than 20% of the personnel on campus).
> The big universities will NEVER give out high quality equivalents of their courses online... MIT and the other schools know if they gave away their core curriculum's online that are basically the same thing then less people would want to go to the school at the current over inflated rates.
Except Open Courseware (thank you Hal Abelson) is exactly that: typically everything handed out by the prof including syllabus, lecture notes, problem sets, clarification notes...everything! And videos of lectures in some cases. And the motivation was precisely the opposite of what you say: "we assembled this stuff; perhaps it's useful for you to make your own course too."
> If you guys are looking for a revolution in online education, I hate to say it, but don't look at traditional institutions to do it....The revolution in online education will be done by someone ... that basically has no connections to the traditional education industry but is just very motivated to change the world for the better.
Umm, maybe. Sadly, a big part of higher education is credentialism, and for that you need to tie back to institutions. And the big institutions have an interest in such experimentation for the standard big institutional reasons that are not specific to universities (the "satellite campus" system has worked for some big institutions like NYU, and their students in, say, Abu Dabi who never go to NY at all) and there's no reason to think that similar classes of experiments could happen via linkups like U of Il + Coursera).
But I agree that new entrants like Kahn are doing interesting experiments that might have a huge, benefit effect in the long run.
I think we will be able to disrupt higher education credentialism. I disagree with your idea that it needs to tie back to institutions. It does not. It simply needs to be able to show to employers that the credential has value. That's it!
You can’t get a full undergrad or masters at either school based on your ability.
You can get a “certificate” or some asterisked form of diploma, or you can enter the traditional applicant lotto where a significant number are rejected yet go on to do great work.
The old lotto model is based on the legacy of having enough seats to put students into.
Some newer programs, including one from MIT are experimenting with a scalable online model.
You want a degree from us? Take some classes for a while, prove your ability, you could get in.
The lotto application process besides being limited is imperfect in so many ways. The GMAT if I recall correctly correlates to success only around 65% of the time.
It’s time for these elite schools to decide how important an issue brand dilution (maybe) is for them, and come out and be straight about how much they factor it into their strategy vs. limiting how many diplomas they grant based purely on scalability while maintaining quality.
Ones a logistical problem. One is profit (endownmenrm prestige) motivated.
The issue you have identified is finding scalable method of accreditation. Other schools have certainly tested online-only degree programs and produced many graduates. As an alum, I do struggle with the question of, "would an online-only graduate be 'real' alum"? That's my own personal bias. I imagine the institute does think about brand dilution to some extent.
That said, while colleges may be gatekeepers to degrees, it is employers that require the degrees to get jobs. Why bother with degrees in the first place if the candidate can prove they have the necessary skills for a job despite not holding a degree?
I realize I'm deflecting, but it's worth pointing out that there are multiple parties in play here, not just universities—MIT or otherwise.
On the other hand if talented teachers can get paid for doing what they do best then they can offer a personal experience as opposed to a diploma factory. I strongly suspect that there will always be a market and mechanism to support that.
* High-quality courses (or at least as good as the original few!)
* Open-licensed courses (as originally promised and intended)
* Real checks-and-balances and not-for-profit structures
* Investment in research in improving teaching-and-learning
* Commitment to integrity in results presented to the public, in respect for student privacy, and in general, a strong set of core values and to keeping what's working in education
The collaboration component of MOOCs ranges from mediocre to god-awful. And it's hard to see how it could be otherwise at scale.
A lot of courses are run asynchronously which blows a lot of meaningful collaboration out of the water right there. And even when they're run like a real-time course (which a lot of people who have other schedules/travel/etc. tend to hate), you have such a wide range of skill/language/etc. levels that it's hard to have sensible discussions.
Courses that try to be explicitly discussion-focused are even worse.
Autograding for coding assignments is nice when it works. But I'm honestly not sure the average MOOC is really any better than just reading a book and doing some related exercises.
I do think you need to have deadlines. They can be more flexible but deadlines help at least keep groups of the class at the same pace. The more people participating, the more relaxed the deadlines can be. I've seen some courses that have so many people, you could honestly take the class at your own time & always have people to discuss the current lecture with.
In the case of a real MIT online degree, I would support a schedule that mimics the campus schedule. If you have other schedules/travel/etc., then sign up only for 1 course at a time & understand what you're committing to.
I get scaling is hard the more "real" you make the course. I feel you can have a nice balance between hiring assistants to help with grading & discussions by increasing the cost somewhere in between on-campus & average MOOC prices.
>I feel you can have a nice balance between hiring assistants to help with grading & discussions by increasing the cost somewhere in between on-campus & average MOOC prices.
Blended models have a lot of promise--at least in theory. My understanding is that post-pivot Udacity does some things along these lines. And, of course, there are more traditional degree programs that have a large online component.
One of the nice things about CS/programming is that, in many cases, you don't really need the physical resources of a university campus. And even if you can't handle 100% of a full degree program, "nanodegrees" and the like are a big win. It's also nice that computer systems can handle a lot of the grading of problem sets--and, as you say, it's not super-expensive to have TAs handle the rest. (Source: I remember what I was paid to be a grader for a few courses in grad school :-))
I would say that most if not all subjects taught at the undergrad level can be implemented in online platforms. You can have in-person seminar courses for all the subjects that actually need hands on practice. Even in those situations, you can still have a significant impact by having simulations using AR/VR or similar. There is a lot of potential in this space that no one really taps into. I see some people doing x and others doing y. It is all good but there is no one doing a unified approach to this so it just ends up going no where. People want the whole picture, not bits and pieces.
> 1) CMU establishing a 'department' of AI (where MIT had merely an AI track within the CS program)
CMU's Machine Learning Department was founded in 2006, before the current AI hype cycle started. If this is MIT's response to a 'department' of AI, then MIT has been asleep for >10 years.
Perhaps you mean CMU's new undergraduate AI degree. But that is not a new department. It's merely a separate major within the existing school. And not really comparable to MIT's recent announcement, which is much more focused on research than new undergraduate majors.
> Yet I can't imagine why an entire 'college' of AI is needed. AI simply isn't a field that's deep or broad enough to warrant an entire college with a handful of distinct majors, like an engineering college or medical school.
The article discusses this point.
Let's start from the premise that MIT is going to focus a lot its hiring efforts on "Computational X" for all X in which it hires. There are basically three advantages to introducing a new academic unit instead of hiring Computational X people into the X department:
1. Collaboration
2. The "X" department might be ossified and unwelcoming to "Computational X". So from a P&T incentive structure perspective, starting a new department/college can make sense.
3. Naming rights => $$$
> Each of this college's AI degrees will span distinct problem or solution spaces? Not likely.
They're hiring 50 faculty, and half those lines will be dedicated to non computer scientists. That's larger than many R1 CS departments. So, they're certainly hiring enough manpower to run several innovative educational programs.
This also explains why it's a college instead of a department. Hiring historians, philosophers, MD/PhDs, biologists, engineers, and computer scientists into the any single pre-existing college would be pretty awkward.
> AI simply isn't a field that's deep or broad enough to warrant an entire college with a handful of distinct majors, like an engineering college or medical school.
The current field of AI might not yet be as broad or deep as the fields you mentioned.
The science and engineering of intelligence, however, holds potentials to be even deeper and more impactful than those fields. Intelligence underlies most human endeavors. Civilization itself would not be possible without it. It is the greatest distinction between us and other animals.
>Yet I can't imagine why an entire 'college' of AI is needed. AI simply isn't a field that's deep or broad enough to warrant an entire college with a handful of distinct majors, like an engineering college or medical school. Each of this college's AI degrees will span distinct problem or solution spaces? Not likely.
Depends which departments/courses they're assimilating. Course 6 is computer science that holds CSAIL, course 9 is Brain and Cognitive Sciences that holds cognitive science, cognitive psychology, and neuroscience. CBMM encompasses everything from probabilistic programming to deep neural nets to classical computer vision.
I think that if they take some of the more exciting but empirically rooted stuff from CBMM and build up a department that can actually train students for it in-depth, that will be a significant improvement in training tracks available to people now. Computational neuroscience, theoretical neuroscience, computational cognitive science, machine learning, statistical learning theory, etc all remain small specialties within larger fields when taken alone, but when put together really deserve to have their connections considered as potentially forming the foundations for a single field.
"...AI simply isn't a field that's deep or broad enough..."
Its ultimate subject is how to make machines think as well as brains think. Arguably, brains are the thing that distinguish homo sapiens from other animals, and we don't know how brains can be engineered or how such devices can be made efficient and accurate.
If you conceptualize that math/engineering/biology/philosophy question holistically, it is definitely worth an independent college.
>> AI simply isn't a field that's deep or broad enough to warrant an entire college with a handful of distinct majors, like an engineering college or medical school.
AI is a field that's over 60 years old, that includes at least half a dozen sub-fields, by my counting (say, NLP, speech processing, machine vision, robotics iiish, game-playing, information retrieval, knowledge representation, inference and reasoning, etc, and, of course, machine learning), with, oh I don't know, out of the top of my head, 50 or so conferences, and about a thousand journals? Thereabouts.
That's broad enough and deep enough to warrant a couple of colleges alright. And if you take into account the age and breadth of subjects encompassed by many of the sub-fields of AI, you probably need a college for each.
> Yet I can't imagine why an entire 'college' of AI is needed
There was a point in time where this statement would sound true when describing computer science as a degree. People quite reasonably thought that CS should just stay under the math department. Yet look where we are today.
Who's knows how much bigger the field will become in a couple decades.
>Maybe this was the only way to ensure the gift of all $350 million. Or to build multiple new buildings...
If it’s anything like the funding that goes to climate research, chunks of the gift will get funneled into projects that are tangentially associated with AI, for the reasons you mentioned.
>> AI simply isn't a field that's deep or broad enough to warrant an entire college with a handful of distinct majors, like an engineering college or medical school.
I guess by this move their goal is to now make it deep enough.
As a newbie to AI and ML, what puzzles me most is that what is really to ultimately it to work is a very well optimized and trained model - but once you have that model, who do you need?
It seems like, to many people AI = ML = Deep Learning. Depending on cluelessness level the executive you talk to they may add CV to the equality, too.
Back in the day we called ML "Pattern Recognition", I remember taking the course from Keinosuke Fukunaga at Purdue (good memories, I found out recently that his son is Gen Fukunaga: https://en.wikipedia.org/wiki/Gen_Fukunaga).
The success of statistical machine learning is more limited than may be obvious. Try to think of problems these methods can't solve, then try to think of problems they can solve. Which stack is thicker? Then ask yourself, for the problems they can solve--can they really solve them as well as humans? Machine Translation, I'm looking at you, Image Recongition--I'm also looking at you. If they can solve them as well as humans, ask 'how much human intelligence is imprinted into this machine artifact?' Yes, AlphaGo, I'm looking right at you.
There are plenty of solutions to get around those problems though. Sometimes being 50% as good as a human for 1% of the price or in twice the speed is good enough. Or if it can be correct 50% of the time, but can tell when its wrong, it can be correct 100% of the time with twice the work. If the work is still cheaper/faster than humans...
>> The two things people dislike the most, the way things are, and when they
change.
Look here, there is no excuse for a reseacher to be ignorant of the history of
his or her field. A researcher, after all, is expected to be a world-class
expert in his or her chosen subject. Joining a field with a history of ~70
years and remaining clueless about 9/10s of it, is not being an expert in anything.
But to address your comment directly, and frankly- what I'm mostly afraid of
is repeating the mistakes of the past, and being lost in a sea of
cookie-cutter papers that repeat the mistakes of the past.
And that's modern machine learning research in a nutshell.
Oh wow, that article explains it's a much bigger change than the headline says — all of Course 6, plus CSAIL and several others will move into the College.
But it's also not a "College for Artificial Intelligence" — it's a "College of Computing".
> A: The founding of the MIT Schwarzman College of Computing is the most significant structural change since 1950, when MIT established the Sloan School of Management and the School of Humanities, Arts, and Social Sciences. But this is much more than a restructuring: With this change, MIT seeks to position itself as a key player in the responsible and ethical evolution of technologies that will fundamentally transform society.
Looks like it's the other way around — Course 6 will become part of the Schwartzman College of Computing. Does that make it an organizational peer with the Sloan School of Management?
The EECS department is moving, but nothing I've read has indicated a change in numbers.
Is what a peer with Sloan? The College of Computing? Here are the current schools and departments: http://www.mit.edu/education/#schools-and-departments. Considering the EECS department is moving, it seems to me that, yes, the College of Computing is an organizational peer of Sloan. The term "college" is strange, given that everything else is a "school".
I have been and continue to remain very skeptical about all this hoopla about AI.
Five years ago, I thought the hoopla about AI is just a fashion and it's all going to quickly pass.
Two years ago, I thought it's a bubble that will eventually burst.
At this point, I'm wondering whether what's happening is pure re-branding where we'll stop using the term CS and instead use the term AI.
If I think of AI as CS, I'm ok with it although I don't think graphics, comp. architecture, networking, and OS are AI. But if the rest of the world wants to call them AI, then let them.
I think you are assuming that weak AI is being touted as strong AI.
What has happened is that methods have been discovered to do many very useful things using weak AI that have significant useful applications.
However, since it's weak AI the building blocks aren't that different from the AI we've had for decades. What makes it novel and worth its own academic specialization is that understanding how to combine and utilize those building blocks is not trivial.
I'd argue that it's currently a bit more of an art than a science, but surely it will eventually become a science.
I think that AI is just a rebranding of all automation, especially in the case where it's data driven automation. So I don't think it is something that will just pass, but will become a term for the interdisciplinary field of solving interdisciplinary problems with computers and data.
There are reasons to think some isn't hoopla, mostly that hardware is getting to brain level capabilities in processing power so you can do cool things you couldn't do before.
My understanding is that you can't really compare processing powers between brains and computers. They work differently and are better at different things.
Nobody is going to be better at raw number crunching than a computer, but there's also no computer that can recognise patterns as well as your brain can. At this point in time, the computer vs brains argument is very situational.
> My understanding is that you can't really compare processing powers between brains and computers.
I agree – I guess I papered over this and basically interpreted GP as saying "computers are getting fast enough to be able to emulate the brain's pattern recognition skills", which seems way too strong – hence my question.
Unfortunately, I think statistics will continue to be under-appreciated. All of my mathematics professors in undergraduate said that statistics and linear algebra are the two most useful fields of math to know and that dedicating time to studying them will pay dividends. It still surprises me when I apply for jobs today how few places, even financial or scientific firms, distinguish between statistics and mathematics in their application forms.
People want to be become machine learning engineers because it's the sexy thing right now, but they don't want to learn the necessary statistics/linear algebra/optimization necessary for the roles. In my experience, these "AI" and "datascience" programs are largely just cash-grabs at most universities. I don't doubt that M.I.T.'s will be rigorous, but I'm largely skeptical of how useful these programs actually are.
I don't know. I studied a lot of Machine Learning in UC Berkeley (even though my specialization was on systems since I like it more) and it was all very rigorous linear algebra, probability theory, optimization, signal processing, information theory, statistics, algorithm analysis etc... Sure we also took classes about designing heuristics, data visualization etc but they were no where near as serious/hard as other classes, so students focused on other classes. Pretty much all students who were serious about ML took upperdivision Linear Algebra, Abstract Algebra and/or Analysis classes. We all took EE, Stats, CS, Data Science etc... and saw ML from bunch of different aspects (e.g. EE perspective being more signal/information -esque, or CS perspective more computational (kernels!) etc...). I have no reason to believe MIT will be any less rigorous. I think most (almost all?) random ML intros online are filler courses without much relation "Actual" (?) ML, but I have no reason believe something from MIT, Berkeley, Stanford, CMU etc will be like that.
Yup. Statistics does continue to be under-appreciated, nowhere more than among dudebros who read up about ML for no reason except that's where the money seems to be.
MIT alum here. I expect this new department to be a gross embarassment compared to the rest of campus initially. Then they'll enforce standards to bring it into line, which will suppress enrollment, and then the department will be merged into the CS department. Won't be the first time.
Generally, the traditional university model has been adding "<insert job title> program" for a long time, whether or not they have a useful curriculum to put students through.
There are office management and msft liscencing degrees. The most popular degree (at least where I live) for both undergraduate and graduates are various generic "business degrees."
In practice, universities have curriculums for accounting, finance and ecocomics.
There is no curriculum for social media or digital growth hacking. There are job titles. There are students who want to enroll. Employers are asking for these graduates (in theory, graf salaries for these are low). Politicians are willing to fund them...
Does a bachelors of social media businessing serve a student 10, 20, 30 years later?
So, will these ml/ai these programs really produce better graduates than maths, statistics or CS programs? Dunno. I'll wager that they're a whole lot better than business stuff degrees.
I'd actually wager that MIT will put students through decent statistics classes. Hopefully they'll also have them write a decent amount of code too.
It has always seemed to me that it is correct distinction. Although there exists a branch of mathematics called mathematical statistics, by its very nature statistics is more like physics, in that it is trying to develop efficient methods of getting information about the objective world based on a certain kind of observations and measurements; and, of course, just as theoretical physics, statistics is highly mathematical, which often creates confusion regarding the actual subject of investigation.
I don't know what job postings you are seeing but the vast majority of Data Scientist/ML Engineer postings at large tech companies that I've seen explicitly mention Statistics as a requisite skill.
Sure, and most of time they do not vet the candidate during the interview. Even if they do, they ask basic questions about normal distributions and other basic concepts.
How important is optimization? Like a lot of engineers I know, I took a few required courses in undergrad on linear algebra and stats. But I've never studied optimization. And I see it come up all the time on lists of central theory for ML & AI. . .
Is there a classic textbook on the subject? Are there any free online courses that are considered good?
Optimization as a tool is important and widely used, but... almost everything grabbing headlines uses some form of SGD + Momentum. Very little of the actual progress comes from better optimization.
Optimization can be pretty useful. Most stats / ML problems are posed as minimizing a function subject to some constraints.
Depending on your problem, you might be able to exploit special structures to solve problems faster than just doing gradient descent. If you know linear algebra and stats, you'll be fine getting through an optimization book.
Boyd's book is canonical at this point, but might be hard to get through. Before you get to actually optimizing anything, you need to make your way through some chapters on convex analysis with little application.
>>People want to be become machine learning engineers because it's the ... thing right now, but they don't want to learn the necessary statistics/linear algebra/optimization necessary for the roles.
Most people who will do these jobs, will stitch libraries into producing an application, like every other programming job. They will need passing knowledge of things, but only make things work.
This is for the same reasons why anyone build an web app is not writing their own TCP/IP stack and their own operating system.
As a matter of fact I wouldn't be surprised, If most people who are claiming to do AI are just writing SQL queries to get Averages and means.
In the hey days of Big data craze, people were using Hadoop and Pig to deal with files a few kilobytes in size, and calling it 'Big Data'.
I don't doubt it and I don't doubt that M.I.T. will create an intensive AI college. My point was that a lot of universities, even distinguished ones, are recognizing that there's a real demand and hype for "AI/Data Science" degrees and in an effort to maximize enrollment and appeal they often minimize the mathematical and statistical requirements.
I don't believe that you need an advanced degree to become a component ML engineer, but the math/stats is necessary pre-requisite and these pre-reqs are often poorly defined. At my college, the only pre-req to the graduate-level ML course was the freshman level intro to stats class and multivariable calculus. About 50% of the class dropped when they realized they didn't know how to construct Gaussian models or perform convex optimization.
Maybe it's just me, but having gone through all the stats and maths behind ML, it seems like ultimately the less interesting part (though to be fair, algorithm design is similarly uninteresting for similar reasons). We're talking about a lot of very long-in-the-tooth concepts that are still the basis of many, many approaches. They're important, but it's well-worn territory.
The underappreciated parts of AI, in my experience, are more philosophical; about the nature of reasoning and approximating or beating human thought. About autonomous agents, non zero-sum games and ethical, non-maximizing functions. There's a huge overlap with logic (philosophical and mathematical) here, and I haven't seen that really broached at any of these big programs.
It definitely is not just you. I spent my first two years of PhD wrapping my head around the stats and maths commonly used in ML, and realized that mathematically (as "theoretical" ML is practiced today), most answers are already provided in classical work of statisticians and probabilists. There are many fascinating questions of probability theory and statistics, but most have little to do with AI. In fact, in terms of the biggest empirical success story (deep neural networks), there are essentially no theorems providing a solid conceptual leap of understanding. Mikhail Gromov goes one step further regarding the lack of theory for neural networks (https://www.youtube.com/watch?v=g4Wl3Ggho6k), and provides a fascinating overview of his thoughts in:
https://www.ihes.fr/~gromov/category/ergosystems/
I am interested in the points you raise, but also realized that I would not find a good environment for it at MIT in EECS, for reasons that are rather obvious from the article's subtext. As such, the last year or so has been spent in a search for good alternatives in terms of research, and I am slowly finding answers. I am happy to discuss more over email.
Long story short: you are certainly not the only one who thinks that way.
EDIT: added a video link to Mikhail Gromov's actual views for better accuracy.
I see a lot of graduate students focusing on practical uses of ML algorithms as a result of this. A lot of people don't realize that a good portion of the math is already figured out, and that it's in the implementation of these algorithms that they can find more interesting results.
It must be noted, though, that "approximating human thought" is just one direction of investigation - and not the most important one at that; as interesting as it may be, it makes almost as much sense as trying to have computers resemble human brains. In other words, the true AI, when it arrives, will not think like us humans (even if at some level it might pretend that it does).
> ethical
The AI will be just as "ethical" as a computer or an assault rifle.
Sort of my point. Current (by that I mean post-early 20th century) approaches were to mimic what we believe to be human reasoning. That's clearly limited.
> The AI will be just as "ethical" as a computer or an assault rifle.
I think that's reductive. Reasoning is not entirely analytical. There are other implications and concerns to artificial sentience.
You can have linear algebra and stats for the ML engineer classes as coursework. Most sciences have an applied linear algebra class which doesn't require the proof heavy math version. And most colleges require a year of stats. I think though while linear algebra for the sake of linear algebra is good, it isn't strictly necessary for AI and you can learn what you need. If you want to go deeper then go deeper. You could argue you need a class in ODEs and stochastic optimal control theory to understand RL, but you don't really. Maybe in grad school, maybe in some research area inside of RL but not in undergrad. Of course the linear algebra will help you. The best thing would be that the RL, control systems, stats, physicists and other related folks would start speaking the same language.
That's much of the math behind AI, but there's much more to building real things with it. It's a cross-disciplinary field touching:
- Math, as you say
- Software engineering, especially data engineering
- Design - since the the math and engineering enable new kinds of problems to be solved by computers, there's a lot of unexplored design territory
- The domains of all the input and output modalities it touches, like linguistics, computer vision, etc.
- Increasingly, ethics
Sure, each of these topics are already covered by existing university departments, but the boundaries between departments are arbitrary and often limiting anyway. Why not establish a new locus that brings much of the above under one physical and administrative roof?
>It is expected that the Department of Electrical Engineering and Computer Science (EECS), the Computer Science and Artificial Intelligence Laboratory (CSAIL), the Institute for Data, Systems, and Society (IDSS), and the MIT Quest for Intelligence will all become part of the new College; other units may join the College.
I agree. I hope that M.I.T. is able to consolidate these domains in a cohesive manner. The mathematics of ML is super unfriendly because it borrows from so many inter-related domains. You'll end up with notation that uses sigma to represent both summation and covariance within a single expression, or inconsistencies with whether vectors are column vectors by default or row vectors. It makes your understanding of the material dependent on whether you had the same background as the author.
In mathematics, good and consistent notation does help learning stuff; but I noticed that it can also have some sort of "parasitic" effect and interfere with the true understanding of things, and so I find it useful to try and see if I can understand something given a less elaborate notation (or without one at all).
In general, one has to remember that mathematical notation was invented to make calculations (on paper) more efficient and not with the goal of making it easier to understand things.
I wonder if this attitude was a thing when CS was first introduced as a major. At the time it must have looked like someone basically took a tiny part of electrical engineering and made it into it's own field of study.
Historically, it was as likely to be part of the math department as it was to be part of electrical engineering. In part, this was because at the time electrical engineering was much more about analog circuits, power systems, etc.
in germany, you can still see the impact on whether CS emerged out of math or EE in the curriculums. When you have a lot of low-level programming, operating-systems, designing hardware etc. the departement was born in EE. If the emphasis is more on math, they teach you an abstract view on "computers" (the turing machine/lambda calculus as the foundation of computing) and view processors as just an application, it came from the math-departement. It's probably the same in the rest of the world.
This type of condescension is precisely why I am glad I was introduced to the field of ML/pattern recognition/statistical learning (whatever you want to call) through a course from the CS department rather than the statistics department.
I have always felt course offerings from CS are more approachable/amenable to beginners in this area. Maybe anecdotal, but statistics depts. have a gate-keeping attitude , a sense of 'oh you don't know the math already? too bad, we are not for you'.
P.S: Approachable/amenable does not mean it cannot be rigorous or you have to cut short the math. You just built up gradually rather than throwing math books in people's face from the very beginning.
There is no subject under the name "Statistics and Nonlinear Function Approximation" that describes the algorithms involved in NLP. No subject or concept in the name "Statistics and Nonlinear Function Approximation" that describes the history and codification of breaking down an image into thousands of additional images and then performing the "statistics" - there is much more in ML/AI than what you'd like to call that college.
Neural networks are all "nonlinear function approximation". Last I checked, the state-of-the-art in all the fields you mention is some variation of deep neural network.
Not to be despicable. On the contrary, function approximation is a very rich and deep topic.
I found the cross-discipline focus on the staff interesting:
> The goal of the college, said L. Rafael Reif, the president of M.I.T., is to “educate the bilinguals of the future.” He defines bilinguals as people in fields like biology, chemistry, politics, history and linguistics who are also skilled in the techniques of modern computing that can be applied to them.
> [...]
> Traditionally, departments hold sway in hiring and tenure decisions at universities. So, for example, a researcher who applied A.I.-based text analysis tools in a field like history might be regarded as too much a computer scientist by the humanities department and not sufficiently technical by the computer science department.
They're not just talking about streamlining "learn these statistical models" but also expanding humanities studies.
The obvious wins from this that I see are:
a) more applications of A.I. in areas that C.S. students are less interested in.
b) more people who are knowledgeable about A.I. outside of C.S.
> The college, Ms. Nobles said, offers the possibility of a renewal for humanities studies at M.I.T., where students flock to computer science and engineering.
Do MIT students major in humanities subjects, or is the department's purpose to make engineering students well-rounded through taking electives in their department?
There are also a fair number of other majors that aren't in the School of Humanities but also aren't what most people would consider STEM like the fairly large School of Architecture, which includes things like Urban Studies and Planning.
The overall number also includes a fairly large number of degrees, mostly masters, from the Sloan School of Management.
Both. Undergrads have certain humanities requirements to fulfill - obviously some are natural fans, others treat it like an "eat your broccoli" requirement. Independent of requirements, MIT also has a fairly good reputation for philosophy, political science, linguistics, media studies, and a few other areas.
Interesting that you choose to remark on Chomsky's political views rather than the fact that he is one of the most influential (and highly cited) academics in modern history.
Both. Students are required to take humanities classes as well as declare a concentration. Many often choose to minor or major in humanities alongside a science or engineering degree. A handful end up graduating with only a humanities degree. My high school physics professor actually graduated MIT with a writing degree before returning for a PhD in physics.
I myself double majored in Music Composition and Computer Science and have spent my career so far working at their intersection. The strength of their music department was a big part of the reason I applied to and attended MIT.
I'm an MIT alumnus ('92). A friend of mine specifically went to MIT to study political science, because of the high-quality faculty and the very small number of students majoring in it. He was one of four majors our year, and got to know the professors very well. He had a blast.
I'm quite skeptical of the interdisciplinary approach. We already know that it's possible to create useful neural networks without any "preprogramming", i.e., creating AlphaGo zero without beginning with any knowledge of Go. An interdisciplinary approach seems backward in that it premises that we need to know something about the problem before we can build the tools to find the solution. That's increasingly not the case, and I think what will ultimately happen at this College is that CS researchers will wall themselves off from the rest, who will just slow them down. If you want to bring AI to other fields, educate the people in those fields or hire an ML engineer.
This boondoggle will have everyone from department heads and tenured professors to Boston building contractors climbing over each other to get to the feeding trough. In the end the billion will be soaked up like a sponge, everyone will be looking around saying "What happened?", and little will show as a result of the expenditure.
Better to have selectively (and quietly) researched and invested in more specific efforts. But yeah, it's hard to find an easy way to spend a billion dollars.
On the one hand I think its great that the humanities are getting increased support in general, and a sort of "upgrade" with more focus on integrating more statistical and analysis techniques / technologies.
But on the downside this seems to only fuel the hype bubble around "AI". I'd rather see existing departments and courses get updated with the technologies and techniques they are trying to integrate rather than a new "College of AI".
> I'd rather see existing departments and courses get updated with the technologies and techniques they are trying to integrate rather than a new "College of AI".
That seems very close to the goal of this new College of Computing (it's not "College of AI") stated in MIT's FAQ [1]:
> As MIT’s senior leaders have engaged with faculty and departments across campus, many have spoken of how their fields are being transformed by modern computational methods — specifically, by access to large data sets and the tools to learn from them. Some of the most exciting new work in fields like political science, economics, linguistics, anthropology, and urban studies — as well as in various disciplines in science and engineering — is being made possible when advanced computational capabilities are brought to these fields.
> The key connector of the College to MIT’s five schools with be the 25 “bridge” faculty: joint faculty appointments linking the College with departments across MIT. With this new structure, MIT aims to educate students who are “bilingual” — adept in computing, as well as in their primary field. The College will also connect with the rest of MIT through its work to develop shared computing resources — infrastructure, instrumentation, and technical staffing.
The FAQ says that this unit is named College instead of School (i.e., in contrast to MIT's School of Engineering, School of Science, etc.) precisely because it is meant to "work with and across all five of MIT’s existing schools".
ever searched for a job in AI? They ALL (at least in Europe) want senior developers with 5-8 years experience in all sort of areas and no junior AI devs. Anybody is offering AI jobs (international) for people who got out of college and specialized in that? Looking for my brother, he is specialized in AI and looking for months for an AI job in Germany.
It's funny how much of a different world Europe seems. In Silicon Valley there are all sorts of AI jobs. I wonder if there are more AI jobs in San Francisco than in all of Europe.
it's not funny if you search for a job. They want developer or admin experts with ai knowledge but nobody is interested to invest in a newcomer (which means university degree with knowledge) in ai.
I'm conflicted on this. On one hand, academic donations are always great. On the other hand, Stephen Schwarzman has a record of giving large donations in exchange for naming rights. He gave $150 million to Yale to renovate the main campus center, called "Commons," into the "Schwarzman Center." He gave $25 million to his high school in exchange for naming rights bordering on autocracy, including having his portrait appear "prominently" throughout the school. These conditions were scaled back after they were made public, previously kept private as a condition of the donation. (source: https://www.washingtonpost.com/amphtml/news/answer-sheet/wp/...) He made a scholarship program called the "Schwarzman Scholars" modeled after the Rhodes. Now, MIT.
Look, don't get me wrong. I think donations to academic institutions are fantastic and he should be lauded for his generous giving. However I think it is worthwhile as a society for us to inspect these kind of actions a bit more critically. In my view, Schwarzman, who has no prior record of public interest, giving, or service prior to the last 10 years, is embarking on an aggressive campaign to formulate a positive legacy of his name with his money before he dies. It is artificial, transparent, and revisionist. 100 years from now, people won't remember Schwarzman for being a Trump supporter/friend/advisor and a wealthy Republican. As he has made certain with these donations, Schwarzman will be remembered as a benevolent philanthropist.
He has done an extremely clever thing. Even I can't deny that he has done a wonderful thing by giving away so much money. So who can justifiably criticize the intent behind his actions? No one, really.
To me, Schwarzman's donations reveal just how much of culture and history is straight up bought and paid for. If you have enough money, no matter how you actually live your life and what you do, you can just pay the right people or institutions, and you will be forever remembered as a good person. Remember that.
It's not as if this is an unusual thing. Stanford University is named for Leland Stanford. Carnegie Mellon University is named for Andrew Carnegie and Andrew&Richard Mellon. Duke University is named for the father of James Buchanan Duke. Purdue University is named for John Purdue.
Those people got to name universities by being rich and donating huge quantities of money. They weren't scholars, nor so far as I know were they especially virtuous people. They wanted their name on things. If Schwarzman does the same, it'll be nothing new.
We can go a lot further back, of course. Consider, for instance, King's College at the University of Cambridge. It's named for King Henry VI, who was no scholar and doesn't even seem to have been a particularly effective king. Balliol College at Oxford was named for John de Balliol, notably mostly for being extremely rich; it seems he didn't even particularly want to found a college but was told to do it as penance after some sort of dispute with the Bishop of Durham. That was in the thirteenth century.
This is normal. The Rhodes scholarships you compare with were themselves funded by Cecil Rhodes’ dirty fortune made in Southern Africa: Rhodesia, now Zimbabwe.
> If you have enough money, no matter how you actually live your life and what you do, you can just pay the right people or institutions, and you will be forever remembered as a good person.
Yep. The Nobel prize was created quite explicitly to assuage critiques of Alfred Nobel's legacy [1].
With 100 years to look back on, did the ends justify the means?
>To me, Schwarzman's donations reveal just how much of culture and history is straight up bought and paid for. If you have enough money, no matter how you actually live your life and what you do, you can just pay the right people or institutions, and you will be forever remembered as a good person. Remember that.
Excellent point, I never think twice when people insist on the original source. The motive behind the article matters and secondary independent / unbiased source can provide a more balanced and truthful perspective.
None of those are accredited programs and employers don't care at all if you have them completed. They are treated on the level of MS certified beginner in something.
Can you be more specific? What's wrong with GT's OMS CS? I've heard it's tough and students are flooded with projects all the time, and some bleeding edge ones.
Yet I can't imagine why an entire 'college' of AI is needed. AI simply isn't a field that's deep or broad enough to warrant an entire college with a handful of distinct majors, like an engineering college or medical school. Each of this college's AI degrees will span distinct problem or solution spaces? Not likely.
Maybe this was the only way to ensure the gift of all $350 million. Or to build multiple new buildings...