Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 



Forgot your password?
typodupeerror
AI IBM Businesses Education Software The Almighty Buck Technology

IBM To Invest $240 Million To Develop AI Research Lab With MIT (bloomberg.com) 39

IBM will spend $240 million over 10 years to develop an artificial intelligence research lab with the Massachusetts Institute of Technology, pooling the organizations' resources as competition intensifies to produce breakthroughs in the field. Bloomberg reports: The MIT-IBM Watson AI Lab will fund projects in four broad areas, including creating better hardware to handle complex computations and figuring out applications of AI in specific industries, the Armonk, New York-based company said Thursday in a statement. While IBM has always conducted long-term research internally, it decided AI was such a vast field that it needed to reach out for talent and ideas, said John Kelly, the head of International Business Machines Corp.'s research and cognitive solutions groups, which includes Watson products. While researchers will focus on long-term innovations in artificial intelligence, IBM will also be looking for developments -- a new medical imaging algorithm, say -- that it can immediately plug into its existing products. Big Blue expects to see results that boost its Watson-branded AI business in the next year or two, Kelly said. The plan is to change the focus and number of teams as needed to produce results, he said. The partnership underscores IBM's focus on building a business selling AI software, a strategy that requires clients to adopt such products and the company to develop offerings that add actual business value and are competitive with juggernauts in artificial intelligence, including Microsoft Corp. and Alphabet. IBM and MIT will jointly own the intellectual property that results from the projects conducted together. The company also has the option to buy out MIT for full ownership, Kelly said.
This discussion has been archived. No new comments can be posted.

IBM To Invest $240 Million To Develop AI Research Lab With MIT

Comments Filter:
  • by TimothyHollins ( 4720957 ) on Friday September 08, 2017 @05:25AM (#55157937)

    How about spending some time and money figuring out how to make a human-interpretable learning system? You know, reinvent rule-based classification, reducts, and Rough Set theory stuff. That's something that will actually advance the field instead of another Alexa or Siri for easier ad-targeting,

    That way, when Skynet finally decides to assassine Kanye, at least we can go to the databanks and see exactly how it reasoned.

    • Especially as it concerns safety, I wholeheartedly agree.
    • The problem with many learning systems is once they are capable of making decisions like that, going back and trying to see how they reasoned is nearly impossible.
      • Yes, that is why I specifically recommended something like rule-based learning, a classification technique built on 1) a ruleset that specifies an outcome given a specific set of observations and 2) a set of metarules that specify how unclear cases should be handled and any upper/lower bounds on certainty/quality of data etc required.
        With such a system you can pinpoint exactly which rules were used for every decision made, and how altering the base metarules would affect the given decisions. It also allows

  • Awesome! (Score:5, Funny)

    by Pseudonym ( 62607 ) on Friday September 08, 2017 @06:08AM (#55158013)

    I always thought that MIT could do with an AI lab. Nice of IBM to finally help them get one.

  • To tweak a list of threshold values between 0 and 1.
  • by Baron_Yam ( 643147 ) on Friday September 08, 2017 @06:55AM (#55158121)

    IIRC, IBM has made memristors, a basic electronic component only recently produced (and I believe also difficult to produce which is why they're still in the lab).

    Memristors are nice little toys that effectively combine memory and processing, are non-volatile, fast-switching, low-power, small, and can be non-binary. In other words, they sound very much like a non-biological implementation of a neuron.

    I'd love to see a hardware implementation of an artificial neural network, as getting it out of software should immediately massively improve the potential speed and complexity.

    https://en.wikipedia.org/wiki/... [wikipedia.org]

  • Last year they funded the Center for Cognitive Computing Systems Research at the University of Illinois Urbana-Champaign. http://www-03.ibm.com/press/us... [ibm.com]
  • IBM and MIT will jointly own the intellectual property that results from the projects conducted together. The company also has the option to buy out MIT for full ownership, Kelly said.

    With Watson being completely underwhelming in most or all of its commercial applications, IBM needs something better. Something useful.

    Hopefully MIT gets royalties as a part of that option, as I assume IBM is planning to make a lot of money from this endeavor. IBM gets access to some of the best minds in the field, and they would have no worthwhile product otherwise.

  • no surprise (Score:3, Insightful)

    by Anonymous Coward on Friday September 08, 2017 @08:41AM (#55158483)
    Having met with top execs at IBM Big Insights, they've had a very hard time getting clients to use their product. As an Apache committer, it's no surprise. IBM has been trying to sell Watson through their big data initiative for years without any luck. The project they did with Wellpoint using Watson went no where and isn't really useful. It was nice for dog & pony show, but that's about it. The challenge with DNN has always been subject matter expertise. The engineers that built the frameworks don't have expertise in the customer's subject matter. Finding people in each field with expertise is challenging and convincing them to spend 3-5 years training DNN is even tougher.

    If you're asking "why doesn't IBM find 10 doctors and pay them for 5 years?" That's just not practical. A doctor isn't interested enough to stop being a doctor and spend 3 years teaching a programmer the subject matter. You need programmers that are experts in DNN to learn enough of a subject to produce something useful. You need to teach a doctor enough about DNN so they can help produce the training dataset. There's a lot of trial and error.

    IBM needs MIT to figure out how to solve the subject matter expertise problem or at minimum reduce the cost/need for subject matter expertise. 240 million is a drop in the bucket when you consider IBM has dumped billions into watson without making any profits.

  • Glad to see Gore and Romney working together on something. Yeah, that was bad.
  • MIT's Project MAC is 54 years old. Who's fooling whom?
  • Wait, didn't IBM just lay off like 70,000 employees? How do they still have $240M just laying around to throw at outside research?

"Remember, extremism in the nondefense of moderation is not a virtue." -- Peter Neumann, about usenet

Working...