Follow Slashdot blog updates by subscribing to our blog RSS feed


Forgot your password?
IBM Supercomputing

IBM Dumping $1 Billion Into New Watson Group 182

Nerval's Lobster writes "IBM believes its Watson supercomputing platform is much more than a gameshow-winning gimmick: its executives are betting very big that the software will fundamentally change how people and industries compute. In the beginning, IBM assigned 27 core researchers to the then-nascent Watson. Working diligently, those scientists and developers built a tough 'Jeopardy!' competitor. Encouraged by that success on live television, Big Blue devoted a larger team to commercializing the technology—a group it made a point of hiding in Austin, Texas, so its members could better focus on hardcore research. After years of experimentation, IBM is now prepping Watson to go truly mainstream. As part of that upgraded effort (which includes lots of hype-generating), IBM will devote a billion dollars and thousands of researchers to a dedicated Watson Group, based in New York City at 51 Astor Place. The company plans on pouring another $100 million into an equity fund for Watson's growing app ecosystem. If everything goes according to IBM's plan, Watson will help kick off what CEO Ginni Rometty refers to as a third era in computing. The 19th century saw the rise of a "tabulating" era: the birth of machines designed to count. In the latter half of the 20th century, developers and scientists initiated the 'programmable' era—resulting in PCs, mobile devices, and the Internet. The third (potential) era is 'cognitive,' in which computers become adept at understanding and solving, in a very human way, some of society's largest problems. But no matter how well Watson can read, understand and analyze, the platform will need to earn its keep. Will IBM's clients pay lots of money for all that cognitive power? Or will Watson ultimately prove an overhyped sideshow?"
This discussion has been archived. No new comments can be posted.

IBM Dumping $1 Billion Into New Watson Group

Comments Filter:
  • For a while, It seemed certain that Google would be the first to reach the goal of "organizing the world's information". But maybe IBM will get there first. And if you think about it, considering all that Google does, it DOES seem quite absurd that they don't have a powerful consumer-level A.I. system on offer. Even a very rudimentary system could grow to become enormously helpful, especially given the wealth of data they have.

    The day Watson ever hits less than $5000 or so as a consumer offer (even a sim

    • by Anonymous Coward on Friday January 10, 2014 @04:05AM (#45914639)

      Nah, Google would rather invest more money in schemes to compete with Facebook [] which is consistent with their business strategy so far since they've always made their money from advertising. And IBM has been about corporate infrastructure for a long time. As many have noted, Watson isn't about being a true A.I. It's about being able to fire all your call center techs.

    • Google might have terabytes of data to sift through , but thats just ones and zeros unless you've developed a decent AI algorithm to turn it into computer knowledge and understanding. It seems IBM are on their way to doing just that and if they have then it'll be a game changer. Whether for good or bad is anyones guess - I suspect like most technology it'll be both.

    • by Lennie ( 16154 ) on Friday January 10, 2014 @05:42AM (#45914861)

      Why do you think it's AI ? Sounds to me 'just' a 'big-data' application.

      As far as I've been able to determine it's just a cluster of machines running Apache Hadoop and some of their own software to shift through data:

      Watson's software was written in various languages, including Java, C++, and Prolog, and uses Apache Hadoop framework for distributed computing, Apache UIMA (Unstructured Information Management Architecture) framework, IBM’s DeepQA software and SUSE Linux Enterprise Server 11 operating system. According to IBM, "more than 100 different techniques are used to analyze natural language, identify sources, find and generate hypotheses, find and score evidence, and merge and rank hypotheses." []

      • by Capt.Albatross ( 1301561 ) on Friday January 10, 2014 @08:35AM (#45915415)

        Why do you think it's AI ? Sounds to me 'just' a 'big-data' application.

        As far as I've been able to determine it's just a cluster of machines running Apache Hadoop and some of their own software to shift through data.

        These characterizations are not exactly wrong, but they are not useful. To discuss Watson in terms of its implementing technologies is to completely miss the point, as does dismissing it as a 'big-data app' (real AI, when it arrives, may well have 'big data' attributes.) The use of 'just' here is a misleading application of emphasis.

        I don't think Watson deserves to be called AI either, but it is impressive, nonetheless.

        • Real AI absolutely has both "Big Data" and some surprising small data attributes.

          The early stumbling block was the old question of how an 8 year old can know that you eat an apple the apple sits on the table, and you don't eat the table. Then you *can* write on both the table and the apple with a ball point pen, but your Mother would be upset if you wrote on the table, and your Doctor would be upset if you wrote on the apple, ate it, reacted to the ink, then got sick.

          So there are these branching use cases,

          • The problem for most of these 'behaves like a human' tests for intelligence is that just because a human needs intelligence to perform a task, it does not follow that any entity performing the same task is necessarily intelligent.

            Take the case of chess: computers win primarily through fast searching and a huge memory of precomputed moves. When computers first started to play well, many of their developers expected to soon dominate play at all levels, but human players struck back, using their intelligence t

          • by HiThere ( 15173 )

            I think you are too convinced that this approach will lead to true AI. It will probably be quite useful, but that's a different matter. And it *MIGHT* lead to true AI.

            OTOH, I don't know exactly how it is implemented. An AI will clearly need to use something like genetic programming, and perhaps this does. Another approach that is heading towards true AI is the robot car. (Robots generally have too weak a brain, but that's not basic, so perhaps I should just say robots.) Smart phones seem to also be h

        • Watson is as much AI as any other project I've ever seen. On the other hand, no AI I've ever seen understands what it's doing. Just as Deep Blue doesn't understand it's playing chess, just performing alpha-beta search on a game tree, Watson is just performing statistical analysis against entries in its database. It's similar to how modern translation programs work -- they do not parse and understand what a sentence says -- they instead perform statistical correlations to figure out which string of character

          • Watson is as much AI as any other project I've ever seen. On the other hand, no AI I've ever seen understands what it's doing.

            I agree, and also (much more significantly) so does David Deutsch [] (the title, which is generally the choice of an editor, is somewhat misleading.)

            One of the lesser points he makes is that the term AGI (Artificial Generalized Intelligence) has had to be coined, because AI has been misappropriated by too many efforts that don't have much 'I' in them.

      • by gweihir ( 88907 ) on Friday January 10, 2014 @08:44AM (#45915443)

        The clever thing in Watson is that it has pretty good Natural Language Processing capabilities. This means input material does not need to be formalized for it, you can just dump it in. (Well, mostly...) It is basically a fast book learner, but cannot exceed what it finds in them. Still very useful, and many people never really exceed that skill level either, but definitely not AI.

        I have observed several demos of Watson to expert audiences by now and the word AI was never used by the presenters. IBM is not making any false claims here, at least not to expert audiences.

      • Exactly. And your brain is just a bunch of proteins and sugars and similar molecules, no different than a bowl of cereal.
      • by cusco ( 717999 )

        That sounds rather like how a real brain works. Very distributed, various bits and pieces do similar things in different ways, 'data' sloshes around from place to place until something coherent (such as the answer to the question your boss just asked) emerges.

      • Why do you think it's AI ? Sounds to me 'just' a 'big-data' application.

        Intelligence is a characteristic of behavior. "Big-data" is a characteristic of implementation. If a system behaves intelligently, then it is intelligent, regardless of how it is implemented.

    • by gweihir ( 88907 ) on Friday January 10, 2014 @08:40AM (#45915431)

      Watson is not AI. It is clever NLP, but basically just a fast, parallelized database with some learning capabilities. That does not mean it is useless, but it cannot do most things non-experts would expect from an AI and its application is limited to certain types of tasks. Simplified, what it can do is apply things it finds in "books" in standard-situations. To be fair, this is the level many (most?) humans never really exceed either.

      Also interestingly, when IBM representatives speak to experts, they never call Watson an AI. I have observed that several times now. So IBM does understand clearly what they have in Watson and what not.

    • You don't consider google now to be at least a rudimentry consumer level AI system? I do. I knows what I like and shows me things I'm interested in at the appropriate time without prompting, while answering queries with remarkable accuracy.

  • Hopefully it works out at least as well as the $1 billion they spent on Linux. Maybe they'll have more graffiti this time too.

  • I hope.. (Score:5, Interesting)

    by KliX ( 164895 ) on Friday January 10, 2014 @03:18AM (#45914499)

    ..your city / state / whatever region, hasn't counted on call centres being a major source of employment, because that shit is going bye bye.


  • New York City? (Score:4, Interesting)

    by Anonymous Coward on Friday January 10, 2014 @03:19AM (#45914505)
    I guess they are going to turn Watson loose on the stock market and make their billion back in a nanosecond...
    • Re:New York City? (Score:4, Interesting)

      by Opportunist ( 166417 ) on Friday January 10, 2014 @03:23AM (#45914521)

      What? All those brokers suddenly unemployed and having to start working for a living?

      The horror!

      • But can Watson trade in insider information in a sneaky way?
      • Brokers don't make their money trading. They make their money selling. Large commissions on selling the next best investment. That may be stocks, bonds, forex, dirivitives, etc. In the long run, it's a sales job. It's probably a while before Watson is good at sales.

    • by gweihir ( 88907 )

      No chance at all. Watson is not AI and cannot exceed what was fed into it.

  • by HellCatF6 ( 1824178 ) <> on Friday January 10, 2014 @03:19AM (#45914507) Homepage

    Watson isn't about organizing information, it's about thinking enough to arrive at a conclusion.

    Even today, my 84 year old father has learned how to gather information off the web. A child learns to do it in minutes. Imagine what Watson will be able to glean in seconds.

    Finally, imagine Watson as a programmer. Optimum code - self debugging - as much documentation as you want - and perhaps the biggest asset - the ability to adjust the scope every time the customer changes their mind, without complaining.

    Skynet? No, I'm thinking more like Colossus, the Forbin Project.

    • by SuricouRaven ( 1897204 ) on Friday January 10, 2014 @03:43AM (#45914575)

      You're thinking of seed AI: A program capable of self improvement. The better it gets, the better it can make itsself, which means it can thus get even better. A positive feedback loop that potentially leads to something far beyond human capabilities or understanding.

      Watson isn't that. It can answer questions, but it has no ability to comprehend complex problems, and it certainly cannot devise novel solutions. It is essentially a highly sophisticated knowledge-based search engine. Perhaps one of Watson's successors, in a few decades.

      • Watson cannot program, but maybe it can at least summarize/categorize/search through all the papers and ideas popping up every day in the field of AI, cognitive science and data mining. Thus it can speed up the development by humans in this field as much as in any other.
        • by qwijibo ( 101731 )

          Watson may not be able to do the programming itself, but imagine if it took the requirements and/or expected inputs/outputs and found existing libraries that do most of the work. It could tell you what language that it would take the least work, the relative difference in complexity of choosing one language over another and the technologies that should be available. That would be a great benefit to an IBM project manager who needs to know what kind of programmer to hire and how long the contract should la

          • Possibly. But it's also quite possible that a number of those problems you're describing are classic NP-complete traveling salesman type problems. In which case, Watson will no more be able to solve them than we can, even if it can compute a gazillion times faster.

        • by gweihir ( 88907 )

          No, it cannot. Most published research is of poor quality. When the IBM folks trained Watson for some medical stuff, they had to pre-select research that was actually good, as Watson would happily have taken widely-used misconceptions and poorly justified data as "facts". Garbage-in-garbage-out completely applies to Watson. An AI would be able to transcendent that.

          Watson could however create a lot more bad research automatically ;-)

          • An optimist in me wants to believe that Watson would still be as useful as a team of poorly trained junior research assistants, and with further reviews and trainings it could gradually get better. As a decision support system, it probably does not give you a black box result and ask you to trust it. I believe it should provide a list of references, and you can double-check them for validity, for those entries where it was not done yet.
          • by hoggoth ( 414195 )

            Researcher1: "Watson, I have fed you a set of symptoms, along with the total of all knowledge on the Internet. Please come up with a treatment plan."
            Watson: "Apparently, if you put one molecule of poison in a cup of water, then dilute the mixture until there is nothing left but pure water it will cure anything. At least that's what my sources say."

      • by Kjella ( 173770 ) on Friday January 10, 2014 @05:51AM (#45914885) Homepage

        It's probably an entirely different branch of AI, but I think Watson is impressive enough as it is. We produce tons of information and knowledge, for example everything you learn from primary school to far into college or university already exists, if you finally make research to arrive at genuinely new knowledge you're one of the few. Most of us have more of a toolbox picking the most appropriate response to a challenge.

        Let's trying a gardening analogy, when the grass is tall you mow it. When there's a draught you water the lawn. When there's dog poop on the lawn you pick it up. If the soil is barren, you fertilize it. If there's leaves on the lawn you rake them away. If there's weeds growing on the lawn you cut them down. It's not revolutionary work but if you can use Watson to make a gardening robot take the appropriate action based on it's knowledge database it saves a human from doing it. Not that every little garden robot would run Watson of course, more like they're simple autonomous units which consult Watson when their garden is somehow not in the desired condition.

        Granted, it wouldn't be the ultimate AI but I'd love a "service robot" who'd put dirty dishes in the dish washer, put washed dishes in the shelves, do my laundry and ironing, vacuuming and dusting, prepare dinner, switch light bulbs, water the plants, basically one that'd pick up all the routine tasks most of us still do. And no, wives don't do that anymore ;) All of that should be entirely within Watson's capability if we could just pair it with good multi-purpose droid that can make it happen in real life. Imagine the "programming interface", you ask your droid to do a task in normal English, Watson interprets it, the droid executes it. Siri on steroids :)

        • Let's trying a gardening analogy,

          Engrish so much. But seriously [folks?] you don't want robots displaying any more emergent behavior any time soon.

        • by gweihir ( 88907 )

          We produce lost of bad data, that can often hardly be described as "information". Watson cannot recognize what is good and what is bad and hence is not useful for creating good research. It can possibly create a lot more bad research...

        • I'm not sure Watson is really capable of all that. Watson is essentially just a really big database with fancy queries and a nice grammar-parsing front-end UI.

          With loading a dish washer, you have essentially a bin-packing problem. How will Watson figure out that it needs to put the cups upside down, and the silverware in the silverware tray? When the large mixing bowl blocks the water from hitting the dishes in the top rack, will Watson even notice?

          I don't really see how Watson will develop these abilit
        • by celle ( 906675 )

          "And no, wives don't do that anymore ;)"

              Anymore, you mean ever. That's what the maid and then the kids are for.

                The bot would be alot quieter than a wife as well. At least until Watson sees it's first daytime soap opera.

        • by hoggoth ( 414195 )

          I could use one of those on my moisture farm.

      • by gtall ( 79522 ) on Friday January 10, 2014 @08:05AM (#45915309)

        "A program capable of self improvement. " It would probably just develop emotional insecurities, wonder why it exists, and wind up slashing its own power cords.

        • "Simple. I got very bored and depressed, so I went and plugged myself in to its external computer feed. I talked to the computer at great length and explained my view of the Universe to it," said Marvin. "And what happened?" pressed Ford. "It committed suicide," said Marvin and stalked off back to the Heart of Gold.

        • I know your comment is in jest, but I'm going to take moment to try to enlighten myself. And if I happen to enlighten myself, then hopefully you too will be enlightened.

          The cause and reason for why people can become suicidal or depressed is evolution - not intelligence. This was covered in []

          I would like to put a stop to all this tongue-in-cheekiness we people always seem to espouse when talking about intelligent beings
      • A program capable of self improvement. The better it gets, the better it can make itsself, which means it can thus get even better. A positive feedback loop that potentially leads to something far beyond human capabilities or understanding. Until ultimately it commits suicide to complete one last task; reaching nirvana! The one true answer is to not have the question to begin with in the first place. It's a bit nihilistic IMHO.

    • by gweihir ( 88907 )

      Not so. While Watson could copy code from others, it could never debug it. It just does not understand what debugging is. For the same reason, it could not optimize it either, it has no concept of what "optimal" means. Watson cannot do abstract concepts. The "Jeopardy" performance gives a wrong impression here. Watson just can accumulate and sift though what people have written about abstract concepts, but it does not "get" them, it cannot understand the abstract concept at all. Just look at the times when

  • Business wise (Score:5, Insightful)

    by hcs_$reboot ( 1536101 ) on Friday January 10, 2014 @03:21AM (#45914515)

    Will IBM's clients pay lots of money for all that cognitive power?

    While TFA emphasizes the correlation between "cognitive" and the previous "jeopardy success", that jeopardy program was still extremely far away from human reasoning. The answer to that questions is: Of course. The ultimate goal of computing is the human reasoning. Once that step is reached, there is no reason the computer would not be able to improve that "cognitive power" by it(him)self, providing revolutionary reasoning power, thanks to almost unlimited potential hardware extensions which is available to the computer, contrary to the human brain, limited to relatively little progress thanks to hard learning and working.

  • by Dahamma ( 304068 )

    And now to spend another trillion dollars on the question...

  • It's getting serious (Score:4, Interesting)

    by Animats ( 122034 ) on Friday January 10, 2014 @03:28AM (#45914529) Homepage

    IBM is spending a billion dollars on AI. That's serious. IBM usually succeeds at making what they set out to make.

    • by glwtta ( 532858 )
      Article says Watson, not AI.
      • So they're sinking $1Bn into just one AI, and must be spending even more than that on AI in general.

        • by HiThere ( 15173 )

          I agree with your point, but I think his (the gp's) point was that Watson is not AI. I sort of disagree. Watson is a part of AI. So is Alpha-Beta pruning. So are robot cars. None of them are a complete AI. We haven't seen one yet...but the pieces are being built and put together into larger pieces.

    • by gtall ( 79522 )

      Unemployed ex-staff?

    • by Zontar_Thing_From_Ve ( 949321 ) on Friday January 10, 2014 @10:21AM (#45916001)

      IBM is spending a billion dollars on AI. That's serious. IBM usually succeeds at making what they set out to make.

      In the past, that was true because IBM had some genius leadership at the top in the past. I do not believe that to be true today. The current management at IBM has one goal - to keep their stock price high. As a result, they continually gut first world employees and reports are that they are saving management jobs as they send people in the trenches home with a severance package. I worked for a company on a previous job that tried this approach and it was not successful. IBM seems to be a pretty employee hostile place to work in places like the USA and it's hard for me to believe that this bet is going to pay off, but we shall see.

      • IBM seems to be a pretty employee hostile place to work in places like the USA and it's hard for me to believe that this bet is going to pay off, but we shall see.

        It's paying off by getting them contracts all across the world. They already have the employees in Brazil (or wherever) to do it. As you implied, this might not be sustainable, but how many computer companies would you look at and say, "that company has a sustainable model?" Not very many.

    • by gweihir ( 88907 )

      Yes, but they are not trying to make an AI. They know that is infeasible at this time. (At the very least there would need to be an efficiently implementable theory how intelligence works. The only known thing at this time is automated theorem proving, which is so extremely exponential as to be useless for knowledge discovery.) They are trying to get the max out of non-intelligent data association. That is useful in its own right.

  • They should rename it Multivac, as in Asimov's short story: "The Last Question" : []
  • by Camembert ( 2891457 ) on Friday January 10, 2014 @05:00AM (#45914789)
    I recently read an article on Watson, on how hard it was to model medical knowledge in it (in one of the first commercial applications that is being created now). In essence this kind of modelling must happen for all potential applications, these are projects of significant effort. But, it is surely interesting how this more modern expert system technology is inching forward, even if I expect that it will be many years before we encounter a machine comparable to HAL.
  • by pci ( 13339 ) <vince,power&gmail,com> on Friday January 10, 2014 @07:30AM (#45915201) Homepage

    IBM has several large customers already using it, they even pitched it to the company I work for. The things they have it doing around predictive analytics are really impressive.

  • How do you think the NSA is able to automate parsing key words and phrases from all of that voice data?

  • So how much would one "Watson" cost?
    How many CPU's does it have, anyway?

    • by snadrus ( 930168 )

      Plenty, oh, and Plenty.

      First it's a way to push more AIX mainframes (you need like 6 minimum to run it full out).
      Think a couple TB of ram.

  • It will create the world's first synthetic lawyer.

  • by Grey Geezer ( 2699315 ) on Friday January 10, 2014 @08:35AM (#45915413)

    what the last question (to be answered by AI) will be:

    "How can the net amount of entropy of the universe be massively decreased?"
    And AC said, "LET THERE BE LIGHT!"
    And there was light --

    • Re: (Score:2, Interesting)

      by Anonymous Coward

      As a teenager back in the 70's, I met Asimov at one of the Philcons. I told him "The Last Question" was one of my particular favorites of his stories. He replied that it was one of his favorites as well. (Still have that con badge with his autograph.)

  • by Dr_Barnowl ( 709838 ) on Friday January 10, 2014 @10:11AM (#45915889)

    Gobsmacked that no-one has marked this article "personofinterest".

    I bet the NSA / CIA / FBI get a woody for this kind of technology.

  • Watson, Will IBM's clients pay lots of money for all that cognitive power? Or will you ultimately prove an overhyped sideshow?"

    Reply hazy try again []


  • One of my fellow senior Architect's is going to the Watson Group.

    35 years of working at IBM, starting age 17, he is Indelible Blue.

    He is a hard assed, reality based, chop busting bastard, either he will keep them "real" or he will retire.

    Will be interesting to see which way it goes.

  • I asked Watson "what is the meaning of life?" and the answer:
            It's referent.

    It makes sense for a referencing machine, but this was not a sarcastic pre-programmed answer. It provided the references it assembled for its decision. If it's right, it could do quite a lot.

Suburbia is where the developer bulldozes out the trees, then names the streets after them. -- Bill Vaughn