Carnegie Mellon Reels After Uber Lures Away Researchers | Hacker News Reader

Carnegie Mellon Reels After Uber Lures Away Researchers

wsj.com

27 pointskjhughes15 comments

tl;dr

uber hired 6 PIs, 34 engineers, and the institute's director by, in some cases, doubling salaries and/or paying $x00k bonuses

Apparently systematically underpaying people (for science!) isn't as good a strategy as one may naively believe it to be...

>doubling salaries and/or paying $x00k bonuses

Holy...

One serious problem with universities, and the explicit reason Danny Hillis had to create Thinking Machines to get his Connection Machine (CM, first version) developed, is that they can't pay researcher more than the pay professors. The CM needed 1-2 top notch analog engineers there were wires in the interconnect that had to hit each of the processor chips, which could be as many as 65,536.

And, yeah, the people who pay for this sort of thing have been on a campaign since the '80s to drive down salaries for STEM researchers; the only reason the fields are doing half-way well nowadays is the Great Recession has made other field suck, and of course lots of private money from companies like Google and Uber.

Sounds like the institution moved and left Carnegie Melon behind.

This echos the story of how Symbolics/Lisp Machines effectively wiped out the MIT AI Lab in the 1980s.

Can anyone else think of similar parallels?

Is there a parallel?

The Lisp Machine was developed at MIT with DARPA money. When the technology was commercialized, DARPA wanted two companies - not just one. LMI and Symbolics were founded by MIT people - the same people who already worked for DARPA money at the lab, were now founding two companies financed by more DARPA money. They took more people from the AI Lab, where the machines were originally developed.

The parallels aren't perfect. The analogy I see is that there's a large number of people working on a topic at a university. That topic is of likely commercial value. A company/two companies hire out most of the people who work on the topic. The research at the original university is severely curtailed.

(Quoting Stallman about that event in the 1980s, "Nobody had envisioned that the AI lab's hacker group would be wiped out, but it was".)

Another example might be when Netscape hired many of the original NCSA Mosaic developers. However, that was not a major research component at NCSA, so it's a much worse analogy.

> The analogy I see is that there's a large number of people working on a topic at a university. That topic is of likely commercial value.

The MIT AI Lab projets like the Lisp Machine were financed by DARPA. D stands for Defense and ARPA for Advanced Research Projects Agency. It was always expected that the research was paying back with real applications and improved defense capabilities. Remember, it was cold war money.

The Lisp Machine companies LMI and Symbolics were basically startups founded by people of the MIT AI Lab.

Stallman was at best naive. What did he expect? That government/DARPA would finance his hacker paradise without actual results indefinitely? The MIT AI Lab cost huge amount of money and was a part of a DARPA research and budget plan. Commercialization of the research was a part of the plan. The companies then should work on bringing research results into deployment or enabling more specific research - DEFENSE stuff. For example many of the early Lisp Machines went straight into SDI (Strategic Defense Initiative) research.

> Nobody had envisioned that the AI lab's hacker group would be wiped out, but it was"

Maybe he should have talked to the lab director, checked with the DARPA director, should have looked at the business plans of LMI and Symbolics, etc. etc. Then he would have known what was going on.

Did he expect that Greenblatt founded LMI, Inc. to create a new computer company and did not need any people? Where should the Lisp hard- and software experts be coming from? The market? Which one? There was none. This was all new stuff, straight out of university research - financed by DARPA for DARPA. Thus LMI people were coming from the AI Lab.

Symbolics had at it best time almost 1000 employees with over 250 software engineers. Where would these people come from?

Background can be found in these two books:

http://mitpress.mit.edu/books/strategic-computing

http://www.newquistbooks.com/brainmakers/brainmakers.html

Uber and CMU had a partnership deal where Uber funded faculty chairs and graduate fellowships.

Granted, that was only since February. But did CMU expect that Uber would find their research indefinitely and without actual results? Where did they expect that Uber would get their research staff?

Again, I know it's not perfectly parallel. But I don't see why it isn't an historical example of something similar.

It's also not like all of the other DARPA projects had the same results. (Eg, MIT’s Architecture Machine Group wasn't decimated as a result of DARPA funding for the Aspen Movie Map, no?)

Symbolics & LMI were founded by MIT people (and a few others) to commercialized the stuff they developed at the MIT AI Lab. The commercialization of the Lisp Machine technology was a joint decision of the sponsor (DARPA), the university (which licensed the stuff) AND the original developers.

Other than Stallman, most people were happy with the decision.

The problem with that analogy is that these companies didn't gut the MIT AI lab, especially at the level of actual AI researchers vs. the systems people who built and maintained ITS, MACLISP, the Lisp Machine and the Chaosnet (MIT's first LAN). Per https://web.archive.org/web/20140401022259/http://danweinreb... for Symbolics and my memory as an LMI employee:

  Tom Knight (from MIT AI Lab)
  Jack Holloway (from MIT AI Lab)
  David Moon (half-time as MIT AI Lab)
  Howard Cannon (from MIT AI Lab)
  Mike McMahon (from MIT AI Lab)

Plus Richard Greenblatt for LMI.

While some of these people did some actual AI work (and that included Richard Stallman), they were primarily systems people who developed the infrastructure AI research needed prior to industry adopting the these sorts of things like large physical memories (one million dollars ... er, 9-bit bytes), personal workstations, LANs, etc.

ADDED: People really wanted and needed Lisp Machines (Danny Hillis was quite distressed LMI's weren't available when he needed them for Thinking Machines), way beyond what the AI Lab could build, let alone how improper it would have been for them to get into that business, so the commercialization of it was pretty much inevitable, be it on a shoestring, later rescued by TI for LMI, or via the healthy VC market back then for Symbolics.

Thanks for correcting my history!

WSJ commenters speculate that CMU screwed up their contracts and could've protected themselves with better.

Thoughts? Do academic researcher contracts ever include non-compete clauses?

Given how easily these positions go "poof" when a grant ends or doesn't get renewed, that would make any such hiring contract amazingly unattractive (poor pay, no equity, a resume that may not make it clear you can cut it in industry), and very much against the general academic ethos.

As a matter of law, they would only be enforceable against industry sponsored research where the companies involved maintain trade secrets, which again is against the ethos of academia, especially nowadays. (The only legal basis for non-competes is that you will learn trade secrets and need to be prevented from getting the "first bite from the apple* by initially disclosing them to another outfit before legal action can be initiated).

They lured staff away from NREC. I don't think this will affect the academic programs of Carnegie Mellon at all. NREC works on technology commercialization not fundamental research.

Reply on news.ycombinator.com