Please create an account to participate in the Slashdot moderation system

 



Forgot your password?
typodupeerror
×
IBM AI

What Ever Happened to IBM's Watson? (nytimes.com) 75

After Watson triumphed on the gameshow Jeopardy in 2011, its star scientist had to convince IBM that it wasn't a magic answer box, and "explained that Watson was engineered to identify word patterns and predict correct answers for the trivia game."

The New York Times looks at what's happened in the decade since: Watson has not remade any industries. And it hasn't lifted IBM's fortunes. The company trails rivals that emerged as the leaders in cloud computing and A.I. — Amazon, Microsoft and Google. While the shares of those three have multiplied in value many times, IBM's stock price is down more than 10 percent since Watson's "Jeopardy!" triumph in 2011.... The company's missteps with Watson began with its early emphasis on big and difficult initiatives intended to generate both acclaim and sizable revenue for the company, according to many of the more than a dozen current and former IBM managers and scientists interviewed for this article... The company's top management, current and former IBM insiders noted, was dominated until recently by executives with backgrounds in services and sales rather than technology product experts. Product people, they say, might have better understood that Watson had been custom-built for a quiz show, a powerful but limited technology...

IBM insists that its revised A.I. strategy — a pared-down, less world-changing ambition — is working... But the grand visions of the past are gone. Today, instead of being a shorthand for technological prowess, Watson stands out as a sobering example of the pitfalls of technological hype and hubris around A.I. The march of artificial intelligence through the mainstream economy, it turns out, will be more step-by-step evolution than cataclysmic revolution.

One example: IBM technologists approached cancer medical centers, but "were frustrated by the complexity, messiness and gaps in the genetic data at the cancer center... At the end of last year, IBM discontinued Watson for Genomics, which grew out of the joint research with the University of North Carolina. It also shelved another cancer offering, Watson for Oncology, developed with another early collaborator, the Memorial Sloan Kettering Cancer Center..." IBM continued to invest in the health industry, including billions on Watson Health, which was created as a separate business in 2015. That includes more than $4 billion to acquire companies with medical data, billing records and diagnostic images on hundreds of millions of patients. Much of that money, it seems clear, they are never going to get back. Now IBM is paring back Watson Health and reviewing the future of the business. One option being explored, according to a report in The Wall Street Journal, is to sell off Watson Health...

Many outside researchers long dismissed Watson as mainly a branding campaign. But recently, some of them say, the technology has made major strides... The business side of Watson also shows signs of life. Now, Watson is a collection of software tools that companies use to build A.I.-based applications — ones that mainly streamline and automate basic tasks in areas like accounting, payments, technology operations, marketing and customer service. It is workhorse artificial intelligence, and that is true of most A.I. in business today. A core Watson capability is natural language processing — the same ability that helped power the "Jeopardy!" win. That technology powers IBM's popular Watson Assistant, used by businesses to automate customer service inquiries...

IBM says it has 40,000 Watson customers across 20 industries worldwide, more than double the number four years ago. Watson products and services are being used 140 million times a month, compared with a monthly rate of about 10 million two years ago, IBM says. Some of the big customers are in health, like Anthem, a large insurer, which uses Watson Assistant to automate customer inquiries.

"Adoption is accelerating," Mr. Thomas said.

This discussion has been archived. No new comments can be posted.

What Ever Happened to IBM's Watson?

Comments Filter:
  • by Lisandro ( 799651 ) on Sunday July 18, 2021 @02:42AM (#61593547)

    ...is that no one, not even the company itself, can seem to properly explain what it even is.

    • It's a fancy search engine attached to a buzzer.

      • It's a bingo.

        • watson.
          it had 2 language interfaces.
          one for c plus plus and one for java.
          the java worked.
          mostly.
          the c plus plus did not.
          it was the crazyest reverse interface one could ever imagine.
          who ever invented it designed it to be just about as user hostile as possible.
          great concept.
          with a user interface designed by a fool.
          the world moved on

      • You just described almost all of what is called "AI". They call it AI in the press releases and the investor prospectus but the engineers know it is just machine learning optimized search algorithms and a multidimensional database.
    • by shanen ( 462549 ) on Sunday July 18, 2021 @03:16AM (#61593589) Homepage Journal

      Substantive comments and personal experiences to follow (time allowing), but in short, I think it's a loss of vision thing, so I'll repeat (and extend) my old (new) joke genre (mostly dug up via websearch):

      Zero Wing
      All your base are belong to us.
      Google
      All your attention is belong to us.
      Amazon
      All your shoppings are belong to us.
      Facebook
      All your engagement is belong to us.
      Apple
      All your sexy toy are belong to us.
      Twitter
      All your brain fart is belong to us.
      Wikipedia
      All your NPV are belong to us.
      Microsoft
      All your computer times is belong to us.
      IBM
      All your... What was the question again?

      I was actually working in the Big Blue food chain in those years. Innovative suggestions were not appreciated so much. Many of my coworkers were deeply involved in the Watson work. Tangentially, I recall a medical records suggestion that was apparently too innovative to receive serious consideration, but probably I'm the only person with medical records, so it's easy to understand why IBM wasn't interested. Overall I never figured out what the company thought it stood for, but it was completely different when I first worked for IBM way back when... Long time passing.

      As regards Watson applications, I think the most threatening one I knew about was a personality analysis system called "Personality Insights". It produced very interesting multidimensional profiles of people. It was still visible online for many years after I left, but recently they decided to hide it. Or lost interest, like the medical thing.

      • https://www.smbc-comics.com/co... [smbc-comics.com]

        But nothing specific to IBM in the joke. Just a general commentary on today's stock market.

      • TIL that apple makes vibrators. The jokes write themselves when you consider all the "i" products you could name in such a vibrant market.

        • by shanen ( 462549 )

          TIL = Today I Learned?

          (But I would not have been surprised. However when I did the search, I only found apps to make the iPhone vibrate, which is not really that impressive and only 'hosted', not 'made', by Apple.)

      • by Anonymous Coward

        Apple

        All your sexy toy are belong to us.

        Sure, the fanbois would cum-and-take-my-money if iJustine helps field test the new 10" iDildo, but you'd have to pay her a LOT more money to be a different kind of attention whore.

        Tim Cook is about as sexy as cottage cheese. Good luck with that.

    • ...is that no one, not even the company itself, can seem to properly explain what it even is.

      Ironically enough, every not-a-product MBA manager at IBM can probably explain what it is.

      A beneficial tax deduction.

      "The company's missteps with Watson began with its early emphasis on big and difficult initiatives intended to generate both acclaim and sizable revenue for the company"

      A technology company spends millions of dollars for four years on the "next-generation" BrainPuter, and then puts it on a fucking game show? And IBM is struggling to understand why the revenue didn't just come pouring in. Perhaps the answer should have been in the form of a question; Why doesn't Alex Trebek earn $500 million a year?

      And we sadly still face those "big and difficult initiativ

    • by technomom ( 444378 ) on Sunday July 18, 2021 @07:11AM (#61593905)

      It's a brand. Nothing more.

      The original code itself was purpose written specifically to play Jeopardy. There was nothing in it that made it appropriate for medical or other applications.

      IBM did this all the time back when I worked for them. Marketing people would sell an idea based on a proof of concept. They would tell a customer the code would do anything the customer asked for, sell it to them, then turn around and tell the Software Division to take Research's barely written hack and turn it into a product that did whatever cockamamie thing they told the customer it would do on some crazy schedule, like maybe a month or two.

      Of course, unlike Research Division code, IBM's Software Division products needed to meet certain standards, like they had to be supported in all the world's spoken languages, be accessible for blind and deaf, pass all manner of performance and scalability tests. So it was almost impossible to do. What really happened is that in general, we threw out the Research code and we wrote it ourselves.

      So it's doubtful that any of the "Watson" products ever had any of the original Jeopardy playing code in it. I know at least the one that I worked on didn't. What IBM really did is to slap the Watson name on anything that had any kind of machine learning code in it at all And then they made their definition of machine learning very loose.

      • by Registered Coward v2 ( 447531 ) on Sunday July 18, 2021 @12:14PM (#61594599)

        IBM did this all the time back when I worked for them. Marketing people would sell an idea based on a proof of concept. They would tell a customer the code would do anything the customer asked for, sell it to them, then turn around and tell the Software Division to take Research's barely written hack and turn it into a product that did whatever cockamamie thing they told the customer it would do on some crazy schedule, like maybe a month or two.

        When I managed involved in several IT deployment projects I said we need to take the sales person, put him or her in a room onsite and let them do whatever they want; but every time the clients says "the salesperson said it would do...:" or "it'll be ready by..." we get to go in and beat them with a clue by four. I once was asked to work a project and when I looked at the scope, timeline and price said "we can't do that. It'll take at least twice as long and cost five times as much." The salesperson said "the client wouldn't agree to that so this is what I sold." I declined to work on the project; because I'd get blamed for delays and budget busting while he'd still get his commission. Now, if his commission depended on actually making a profit and if we didn't not only did he get none here but also a clawback from others, then it's a different story because you've just given me my clue by four.

        • Firstly, it was an astounding piece of AI engineering that solved a problem that most on slashdot would not have a clue how to solve. It was much, much more than simple keyword matching. Sure, it was not really intelligent, when that happens we are all toast. But it was a brilliant and quite frightening example of what can be done with semi-intelligent systems today.

          Secondly, it is a vacuous marketing term. That is why IBM funded it. There is no money to be made just by playing Jeopardy!

          But do not fo

    • by tomhath ( 637240 )
      An even bigger problem was selling it for tens of millions of dollars, then having to admit to the customers that it was mostly vaporware.
    • You can rent processing time on it. There are some services that will transcribe a WAV file into text using watson or you can do it directly yourself. Its very handy for voicemail attachments to send along, in the body of the email, the transcribed text. The ones using watson have a much higher rate of transcription than the ones that do not. Googles transcription for google voice can also be sourced to a 3rd party⦠guess what? Exact same API. I suspect they are reselling watson access.
    • I know what Watson is.. its too expensive and legally burdensome.

      We were looking at adopting some AI tech for a fairly generic requirement that fell directly within one of the Watson variants capabilities. While it had all the bells and whistles we needed the IBM Watson solution was twenty times more expensive than the next bid and it also included signing over all sorts of rights to our sensitive data and data models to IBM where all other prospective solutions had no such requirements. In the end we suc

      • And when you do buy it, you then have to buy-in expensive contractors. So expensive that some Tax Offices have to hide what was paid, what it has cost so far, and what has NOT been delivered. One was looking at security logs - 30 years worth, and tell us who is doing the wrong thing automatically. Suffice to say, a couple of thousand business rules worked on the demo, but feed it a billion records and watch it choke. Just like the EIS (Expert Information Systems) sold the decade before that. Yeah, exec's b
  • Siri (Score:5, Interesting)

    by phantomfive ( 622387 ) on Sunday July 18, 2021 @03:01AM (#61593569) Journal

    In some ways, Siri is the commercial application of Watson. Scaled down, but effective for most people's use cases.

    • In some ways, Siri is the commercial application of Watson. Scaled down, but effective for most people's use cases.

      Which is what, exactly? Asking it to do things that, most of the time, one can do oneself almost as fast, and far more conveniently, while the things one would really like for Siri to be able to do stubbornly remain beyond its capabilities?

      • one can do oneself almost as fast,

        This is just admitting you are slower than Siri. For non-technical people, the gap is even wider.

  • by joshuark ( 6549270 ) on Sunday July 18, 2021 @03:08AM (#61593577)

    The IBM Watson project reminds me of the Japanese "5th generation" project in the 1980's, another "great leap forward" that was well intentioned, but came up short. "Ahead of its time," is the recurrent mantra.

    I always am open-minded but then skeptical as it is always "the wave of the future." Remember those Nissan ads that didn't feature the automobile? It was the future of advertising, etc.

    IBM Watson was great PR for IBM, for Jeopardy, but simply "did not meet expectations."

    JoshK.

    • Advertisements that don't show the product probably aren't all that bad of a concept. If you do it right you can gain a lot of engagement for your product. Likewise with ads that do little to focus on the product itself. I think the superbowl "wazup" commercials were like this and we could name others. Commercials that focus very little on the product or it's quality while building enthusiasm for it and then stamping your brand name at the end.

    • There are multiple kinds of advertisements. When talking about the automobile industry, there are at least three:
      * Ads to sell you a car
      * Ads to make you feel good you bought their car
      * Ads to make you feel good about a brand

      Sometimes there are multiple aspects in a single commercial, sometimes there are not.

      Brands which don't mention a specific product are in the second and/or third category.

    • by mark-t ( 151149 )
      I had always contended that the game would have been more interesting on Jeopardy if Watson had been playing on a level playing field to the other players.

      Evidently at the same moment that the clue is revealed, The text of the clue was transmitted to the computer. This is not a level playing field because a human also needs to *read* the text they are being presented with, the clue is not simply downloaded directly into a human player's brain as pre-parsed words.

      I felt that the game would have shown a

      • Right you are. In fact it should have to read the clue *on the board* using a camera, and/or listen to Alex read it, same as the human players, before it could answer. Also, adding a human reflex delay. The point of the game is not to prove that the electricity sends signals faster than nerve impulses. We knew that in the late 19th Century.

        • It did have to read the clue off the board and press the buzzer.

          • by mark-t ( 151149 )
            No it did not. The clue was electronically transmitted to Watson as plain text. It had to still parse it, of course, but did not have to translate the glyphs in the optical presentation of the clue into recognizable text, an additional step which all the human players had to perform.
            • Sorry, you are right, I misremembered. It had to use the buzzer like a human but did not have to âoereadâ the screen. https://www.ibm.com/blogs/research/2011/01/how-watson-sees-hears-and-speaks-to-play-jeopardy/

        • by mark-t ( 151149 )

          Also, adding a human reflex delay.

          Adding in a human reflex delay would be imposing an actual handicap on Watson. The point of my suggestion was not to make it more difficult for Watson, but for Watson to be operating on the same playing field as the players. The players must optically process the clue and hear it being read as their modes of input, the same should have been true for Watson.

  • IBM has a history of announcing breakthrough tech that will revolutionize an industry, and then doing nothing with it. I cannot figure out if it's the typical big fat dumb happy corporate monster, spoiled by years of fat contracts with big government and/or corporate customers, transformed into a stagnant and choked-on-policy-and-paperwork trying to figure out how to monetize the quirky stuff leaking out of its research labs - or - if perhaps it's corporate managers trying to pad their golden parachutes by

    • IBM is a patent house. Going to market early establishes prior use etc and makes the IBM lawyers jobs easier.
    • IBM has a history of announcing breakthrough tech that will revolutionize an industry, and then doing nothing with it.

      It's one thing to discover a technology, it's another to commercialize it, and it's a third thing to make it make sense within your business. IBM makes over $600M/year licensing patents, to boot.

  • by Sique ( 173459 ) on Sunday July 18, 2021 @03:59AM (#61593671) Homepage
    My first contact with Watson was a text evaluation app. I fed it an erotic horror story involving werewolves and similar stuff, and it told me, that this piece of text was technical, down to the core of the matter and clean off fluff and and decorations.
  • Because it was capable of running a Google search, people believed it was a human-level AI. This should be the new Turning test. If you can run a Google search, you might be a human. If you can't even get that far, well... you might not be a human-level intellect.
    • What about the people who could search google if they were smart enough to go there and try that before asking on a forum or something? Can we just go ahead and declare them to be of sub-human intelligence too? More than half the time I see a question asked I can literally paste it into google and get better answers than people are actually getting.

      • by zlives ( 2009072 )

        Even google search results are biased towards certain word choices and styles. In my experience i have definitely seen some people search google”wrongly”?
        At least sub optimally to a degree that they never get the right answer vs others that hit the nail dead on.

        • Google certainly isn't perfect nor telepathic, but I find that if you make simple natural language queries, it gets what you want most of the time.

          Other times, it's usually pretty obvious which words are screwing it up, and if you have any familiarity with the subject, you can usually tell which words you needed to include in your query. If you don't, you might have to read some of the not-quite-what-you-wanted articles to find out which words they are.

  • Watson, essentially is a search engine. It works well but is expensive to train and operate, it also really enjoys living in an IBM only environment. In the end Watson got eaten by other offerings, relegated to niche markets.
  • He went elementary.

  • IBM should just not have drunk their own Kool-Aid. The reality of AI is that it's stupid as fuck, so it's only useful for data amounts which are impossible for humans to dig through. Anything for which a human is remotely affordable and for which you don't expect random bizarre outcomes (like say driving) is a lost cause until hard AI.

    False positive/negative rates and random bizarre outcomes don't matter so much for targeted advertising and other surveillance. That's what they should have monetized. It's no

  • and became miss cleverbot http://www.cleverbot.com/ [cleverbot.com]
  • Probably the splashscreen says "Based on Sametime".
  • by chx496 ( 6973044 ) on Sunday July 18, 2021 @06:59AM (#61593891)

    I distinctly remember the ads that IBM put out about Watson all those years ago. They made it appear as if Watson were an AI straight out of a science fiction movie. In that sense they over-promised on anything their product could do, on a gargantuan scale. On the other hand their ads were far too vague to give anyone a clear indication what that product could actually do. Here they under-promised, they didn't indicate at all what specific problems Watson would actually solve.

    Contrast that to the marketing strategy of the other cloud providers: they promised specific things that their solutions could do for their customers: "redundancy", "scalability", "cost savings". Regardless of whether these things turned out to be that true when looking at things holistically, the main cloud companies actually delivered on their specific promises.

    It appears to me that IBM tried to copy Oracle and Cisco by appealing to the manager class (instead of the tech people) in their marketing -- but those two companies at least provide products that both solve a specific problem and can be made to work in the end, so that they do provide value to the business. I have yet to see a single customer story that said "yeah, we've used Watson for X and it solved Y for us". I'm sure there'll be some obscure cases on IBM's website of this, but nothing that even remotely lives up to the hype that IBM's marketing campaign generated. Most importantly: tech people simply don't talk about Watson outside of news stories about it specifically.

    • IBM's business strategy was always to appeal to managers, certainly at least as far back as the 60s when I first encountered the company. Oracle and Cisco copied IBM, not the other way around.
  • The questin is: (Score:4, Insightful)

    by nospam007 ( 722110 ) * on Sunday July 18, 2021 @07:11AM (#61593907)

    What happened to IBM?

  • by OneHundredAndTen ( 1523865 ) on Sunday July 18, 2021 @07:48AM (#61593991)
    That's what happened. Overhyping has been pack and parcel with the AI community since its inception, and things are no different these days.
  • No mention of Nuance (Score:5, Interesting)

    by DavenH ( 1065780 ) on Sunday July 18, 2021 @08:12AM (#61594061)
    The story isn't complete without noting IBM sold a license to Watson's source code to Nuance for $400m [source [techmonitor.ai]], and then realized that they had to re-implement it in full. This alone put them 2 years behind. I got that from a NLP contractor who worked for IBM at the time.
    My own speculation: it used classical machine learning and a ton of hacks, so was brittle and not easy to deploy in multiple environments, i.e. it couldn't be easily retrained to work in German, or legalese.
  • by AmazingRuss ( 555076 ) on Sunday July 18, 2021 @10:24AM (#61594331)
    They've spent the last 30 years demonstrating how to run a tech company wrong.
  • Watson for Crossword Puzzles

  • The IBM tech can debate at the college level with no pre-training. The subject is announced less than an hour before the debates begin, and the AI is not attached to the Net. https://www.research.ibm.com/a... [ibm.com]

  • I feel like part of Watson's problem, from a marketing point of view, is that instead of naming it after the famously smart and observant character, they named it after the slightly slow on the uptake straight-man who follows the smart guy around and takes notes. (Yes I am aware of the Watson it was actually named for. And its cute that someone thought that people would hear "Watson" and think of anything but "It's elementary my dear...")
  • The hard part of AI isn't the algorithm, it's the quality data to train it with.

    I was invited to sit in on a sales pitch from the Watson team to my company's archivists. They had nearly 50 years of hand-drawn designs of products in boxes in a warehouse. The Watson team said it would let them learn all kinds of things about the history of their products, and better yet, use that to predict the qualities of products that will be profitable in the future.

    I asked a couple questions that pretty much sunk the w

  • There is this notion, partly perpetuated by the likes of IBM, that it's possible to build an AI that is "hands-off"--that an AI can be "the expert" without having to deal with human engineering.

    The reality is, IBM spent millions customizing Watson to get it to the point where it could play Jeopardy. Any implementation, such as in the medical field, would require equal or greater amounts of custom engineering (training) to get it to do what IBM hopes it would do. Few customers will pay for that, partly becau

Don't tell me how hard you work. Tell me how much you get done. -- James J. Ling

Working...