http://noticingnewyork.blogspot.com/2014/10/plutocratic-clas...
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...
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.
[1] http://news.mit.edu/2018/mit-reshapes-itself-stephen-schwarz...
[2] http://news.mit.edu/2018/faq-mit-stephen-schwarzman-college-...
Do you have a citation for this advance?
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.
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.
* I'm an MIT alum and former edX engineer.
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.
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.
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.
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 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.
I don't see why not? We are a long way away from general AI.
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.
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.
I guess by this move their goal is to now make it deep enough.
Am I wrong in this understanding?
But I guess we all have our pet peeves, eh?
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).
http://news.mit.edu/2018/faq-mit-stephen-schwarzman-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.
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.
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.
is it? not picking a fight, but i was under the impression that we're nowhere near. could you expand on this?
> 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.
I'm not sure why they need an entirely new college. Doesn't that just increase administrative overhead out of proportion to any perceivable benefit?
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?
https://web.mit.edu/facts/financial.html
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".
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".
[1] http://news.mit.edu/2018/faq-mit-stephen-schwarzman-college-...
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.
https://news.mit.edu/2018/mit-reshapes-itself-stephen-schwar...
Why link to the nytimes when you can link to the actual source?
- Someone circa 18-hundred-something
shrugs
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.
If this was slightly reworded I would love this quote. Is this from somewhere?
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.
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.
Is there a classic textbook on the subject? Are there any free online courses that are considered good?
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'.
- 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?
I wonder how this is going to play with the new MIT Schwarzman College of Computing, and what each entity will choose to focus on.
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.
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.
Not to be despicable. On the contrary, function approximation is a very rich and deep topic.
College of General Purpose Computational Heuristics.
Medical College of Frontier Brain Prosthetics.
The largest subcategory was Economics, which is considered a humanity subject at MIT.
The overall number also includes a fairly large number of degrees, mostly masters, from the Sloan School of Management.
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.
https://news.mit.edu/2017/mit-creates-new-major-computer-sci...
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.
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?
MIT already has a Charles Koch building.
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.
http://news.mit.edu/2018/faq-mit-stephen-schwarzman-college-...
Story time?
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.
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.
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.
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.
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".
Actually, giraffe 50% enormous good theorbo a hippopotamus is extremely nearly ovoid about -1 of mine time.
That's 50% of the sentence:
Actually, being 50% as good as a human is not nearly enough about 100% of the time.
And 50% garbage.
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.
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!
I am currently a CS grad student, I got started with computer science by watching MIT Open Courseware lectures.
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.
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.
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.
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.
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.
Did you mean 'competent ML engineer'
That being said, his LAD and UG theories have mostly not been useful, as we’ve seen to have done better with....ML (ironically enough).
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 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.
Pick a side for the future.
MIT has done a lot to expand access to content in the form of OpenCourseWare (https://ocw.mit.edu/index.htm) and edX (https://www.edx.org/).
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.
Author of edX here. At the time I left edX, the vast majority of courses used long videos and multiple choice questions.
* 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
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.
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 :-))