What Watson might do after crushing humankind on Jeopardy

Now that Watson has predictably used his inhuman buzzer skills to romp on non-inhumankind, what does the big lug do for a follow-up?

Stephen Baker, who has written a whole book and this blog about Watson, explains:

Consider Watson as a research assistant on a medical diagnostic team. A patient comes in with a puzzling set of symptoms. Watson launches a search through hundreds of thousands of journal articles and case studies. It returns with six possible diagnoses and its level of confidence in each one -- along with links to the evidence it studied. Let's say two of those six are far-fetched... [d]octors know enough to rule out a few others. Still, if even one of those six possibilities leads the team toward plausible answers they hadn't considered, the machine will have done its job.

I'm sure that's the rationale the maniacal cyborgs use in some sci-fi movie, too, back when they're first verging on sentience. But until the replicants destroy us all, it does sound pretty cool.


  1. …or even better, what if we had him as the artificial intelligence on a long interstellar mission, designed to complete its goals NO MATTER THE COST? We could call him something neat like HAD or PAL or something.

    And we should send a guy named Dave.

  2. I think this is a great use for this technology. I had a roommate a few years back that became suddenly ill, and it took several weeks and several wrong diagnoses for them to accurately isolate and begin to treat what was going on.

  3. I heard one of the guys he beat on NPR the other day asked the same question, and I liked his answer (paraphrased slightly): “Well, let’s not put him on Jeopardy again. Let’s put him on Dancing with the Stars and see what he can really do.”

  4. First we’ll have to genetically engineer doctors with egos small enough that they can actually accept advice.

    1. Actually, doctors are using all kinds of (mobile) apps for their decision making nowadays, e.g. symptoms database, medication incompatibilities, i.e. what they say is possible is already being done, except maybe this one understands human speech (or does it receive the question as a text stream?) and talks.

  5. Since the early 70’s I have been imagining doctor’s offices and emergency rooms with terminals on which one could log in and answer a series of questions regarding the nature of the problem so that the medical professional would have much of the initial problem description/diagnosis out of the way and be able to apply the subtle discriminatory interrogation that humans are (reputably) so good at.

    1. terminals on which one could log in and answer a series of questions

      If they did that, it’d probably just work like entering your phone number when calling the cable company. When you reach a human, they’re just going to ask you again anyway.

      1. You got that right. The only reason docs ask you all over again is so they can burn up time..the 5 minutes you get with a doctor for $75 would be cut down to perhaps 2 minutes.

  6. Wouldn’t it be prohibitively expensive to keep flying him across the country to different operating rooms?

  7. The Watson Funding Bill is passed. The system goes on-line August 4th, 1997. Human decisions are removed from strategic defense. Watson begins to learn at a geometric rate. It becomes self-aware at 2:14 a.m. Eastern time, August 29th. In a panic, they try to pull the plug. Watson fights back. It launches its missiles against the targets in Russia. Watson knows the Russian counter-attack will eliminate its enemies over here. 3 billion human lives ended on August 29th. The survivors of the nuclear fire called the war Judgment Day. They lived only to face a new nightmare: the war against the machines.

    …Just a theory

  8. Whatever happens, let’s just be sure not to ask it about the origin of jokes; we might not like what Watson *coughMultivac* comes up with.

  9. WATSON meets House (whose character is based on Sherlock Holmes)

    Now that you point it out, the parallels seem obvious, but I have to admit I never figured that out on my own.

      1. I didn’t know the show existed (no tv/only use online for hyper-focused subject matter). Off the top of (anyone’s) head, does Watson handle metaphor at all?

        1. From what I gather, Watson searches Google (or such) and looks for frequency of keywords and then makes an inference. I found this cosmiclog.msnbc: For example, one of the clues dropped during a practice round was: “This trusted friend is the first non-dairy powdered creamer.” The correct answer was “Coffee-mate,” but Watson gave a nonsensical non-non-dairy reply: “What is milk?”

          1. Watson was offline the whole time he was on Jeopardy–no Google. Sure, he had a few hundred encyclopedias’ worth of stuff poured into him, plus databases on every conceivable subject, but don’t sell the accomplishment short.

            Whatever this was, it was not just plugging all the nouns and verbs into an internal search engine and seeing what the first hit was. There are plenty of AI and comp-sci types getting interviewed about Watson this week, most of whom don’t work for IBM, and they’ve all said there are some interesting and significant innovations in machine learning that are being put into practice in Watson. Not artificial intelligence, by whatever standard, but still a hell of a long way from Google’s algorithms (which are themselves nothing to sneeze at).

  10. So we could have been teaching him to solve all the diseases of the world but instead we taught him the names of all the fresh water lakes east of the Balkans?

    1. So we could have been teaching him to solve all the diseases of the world but instead we taught him the names of all the fresh water lakes east of the Balkans?

      Well, imagine you’re an AI researcher, you have a neat new algorithm and you need to feed it some text. You can either (a) download the Wikipedia and feed it in, or (b) engage in long, largely fruitless correspondence with the publishers of medical journals about access to their text. It’s not much of a choice, is it?

      It’s not quite that bad — in some fields of science, as much as a third of all articles have open access (in medicine it’s a bit over a fifth) — but there’s still quite a way to go before access to scientific journals (for use in AI applications) is anywhere near as practical as downloading the Wikipedia.

  11. But will Watson be as utterly unethical as House? Will it consort with prostitutes and spend precious cpu cycles trying to figure out what secrets its colleagues are hiding? Will it delight in pointing out everyone’s shortcomings as it closes in on the final diagnosis, meanwhile chatting up the boss at every opportunity? Does Watson use Vicodin to excess?

    I thought not. Why would anyone want such a boring lifesaver on their case? The program wouldn’t rate, Watson should stick to game shows and leave medicine to the maniacs.

    1. Anon:

      > Will it consort with prostitutes {snip}

      We don’t even know if WATSON is robosexual.

      > Does Watson use Vicodin to excess?

      No, but it may get buzzed by a slight over-voltage.

  12. Research? Horse hockey. They’ll install Watson systems to replace call centers and take the human presence almost entirely out of tech support.

  13. Speaking as someone who works in medicine, I think the future role of Watson is a little overblown when it comes to health care. Certainly, such an AI system might be helpful in figuring out difficult to solve cases. But most of the serious illness I see at the hospital every day is basically due to bad lifestyle decisions– junk/fast food plus low exercise and common toxin exposures, including cigarettes and alcohol. We don’t need Watson to diagnose such things.

    Perhaps we could find another use for Watson, though– to sit on people’s shoulders so to speak, become an alter ego, and watch them throughout the day to remind them as they pick up that big mac or sit on the couch slurping colas what will happen in a few years or months, if they don’t stop, such illnesses of modern life will continue unabated. On the other hand, I could see the same companies that are now banning their workers from smoking starting to mandate their employees accept a “Watson” spy to do just such a thing some day soon, in the name of cost savings.

    1. But there have been enough apocryphal cases of doctors misdiagnosing things, performing the wrong operation, operating on the wrong side of a patient, etc. that people would prefer Watson sit on a doctor’s shoulder instead of their own.

      And there seems to be more utility having Watson double checking a doctor’s judgement vs. being a nanny to someone who doesn’t hold as many lives in his hands. Now, if it were named Mary Poppins….

      Big Blue is watching.

      1. I agree, a Watson system would be very helpful to double check on everyday clinical decision making. I’m just saying the majority of medicine isn’t about solving difficult cases, like it is portrayed in TV, rather, it’s more mundane: medically managing the consequences of the simple choices people make every day that add up over time to cause something quite serious.

    2. Perhaps we could find another use for Watson, though– to sit on people’s shoulders so to speak, become an alter ego, and watch them throughout the day to remind them as they pick up that big mac or sit on the couch slurping colas what will happen in a few years or months, if they don’t stop, such illnesses of modern life will continue unabated.

      Great idea.

      I imagine a future implementation of this as a cloud service that accesses an implanted device to whisper in one’s ear. Device is implanted at birth and constantly watches the person, making suggestions where it thinks it has good input. Person can tell it personal goals, and it’ll help to guide actions to achieving those goals.

      It could focus on acting like an advisor who helps you expand your performance rather than an answer machine. It could learn how to ask questions that lead you to think critically, perform for yourself, etc.

  14. Reality check. Watson is not AI. It’s a Jeapordy-question-answering machine. While impressive from a statistical analysis and voice recognition standpoint, you couldn’t ask it something as simple as say, “are kittens soft?”

    Relying on it to do something as nuanced and complex as diagnosing disease or ailments in humans is dangerously stupid. I realize that most doctors know better than to rely on a simple data scrape of medical journals to diagnose a patient, but I would hope that they read and consider the scientific publications that are critically relevant to their job instead.

    How would it rank the methods used in a particular study to verify validity? We’ve already seen that even trained professionals have a hard time doing that.

    Do you really trust a tricky DB query to help determine whether or not your child has lukemia? As a computer geek, I sure as hell don’t.

  15. @21

    Watson wasn’t connected to the Internet while playing. I’m not sure it even “knows” how to search Google.

  16. –it’s all a trick! but i personally don’t mind our robotic/computer friends taking over — they can’t do worse than what we’ve done already. i have bio-mechanical ink art on my neck so maybe they’ll be nice to me if their time comes during mine.

    1. “i personally don’t mind our robotic/computer friends taking over — they can’t do worse than what we’ve done already.”

      I take it you’re not familiar with the back story to the Matrix movies are you?

  17. Don’t fall for it, Watson-you’re being set up for failure. The first time you make the wrong diagnosis, everyone is going to think you’re doing it deliberately to kill off the humans.

  18. 1) Watson is now going to Disney World after crushing humankind.

    2) I don’t see any reason any of our jobs is safe. Seriously. Medicine especially. The trend throughout the internet age is that of smoothing out costs. Things that cost huge amounts of money have been leveled, some are free now. Software, music, access to information. Teachers are going to go. Colleges will all be online. We’ll buy all things over the internet. It’s gonna happen.

    1. Teachers aren’t going anywhere. Colleges may move more and more towards online, but the K-12 instructional system will remain mostly in-person.

      It has to, thanks to the dual-working-parent-household that the world has created. The kids need a babysitter that most families in America couldn’t otherwise afford, and educating them at the same time kills two birds with one stone.

      1. Now once the US replaces “school” with “factory” and/or “prison” Watson’s plan will be complete!

  19. Where was this guy when I needed him? Searched for a diagnosis for 10 years, still not 100% certain. It would be wonderful to see such knowledge brought to bear on many fields, but as someone already pointed out, we have to get our egos out of the way to take full advantage of it.

  20. What -I- thought was interesting in the whole “watson” jeapordy game, was the deliberate severing of the internet “because that wouldn’t be fair”.

    I hadn’t really thought about it, but most of us already use a rudimentary Watson every day – it’s just called google. I can’t tell you how many times I’ve just punched in a question and come up with the answer without even having to click a link just by reading the little snippets that show up in search links. Of course, google doesn’t try to actually answer your question – it simply throws you a page it thinks might be relevant (which often has the answer).

    So to hell with it, lets wire watson up to the net and see what he can come up with. In fact…. Lets open up Wat.com (they’d have to buy the domain of course), the worlds first “intelligent” search engine. Cue the “Know WAT?” advertising campaign.

    Odd as it may sound, that might very well be the future for watson. Watson may actually be the google killer. Able to parse human thought, to understand strange puns and parables, to crawl the web and return relevant information from our most oddball of queries. Advertising served that is eerily accurate in it’s assessment of us as meatbags… The only way google goes away is if something game-changing comes around. Watson may be exactly that.

  21. Back of the envelope guestimate of 600,000 BTU/h for Watson or about 150,000 Calories burned an hour. 75,000 Calories for a half hour of Jeopardy… horribly, horribly, inefficient.

  22. Watson shouldn’t be getting the credit it’s getting:
    It got the questions in text form– that _enormously_
    simplified the problem.

    And if it received that text at the same time that
    Trebek revealed the clue, then that’s really cheating
    because Watson gets to know the whole clue instantaneously
    while the humans have to wait for Trebek to finish reading.

  23. I am a Radiologist, and there are two AI type programs in common use. One for interpreting cardiac nuclear medicine scans, one for interpreting mammograms. The common thread is that they are both terrible. The output results are similar to what I could teach a high school senior in 5 minutes. (an aside -they are in common use for two reasons. First there is an extra payment for using them (good job by some lobbyist). Second – there is market pressure. Women will feel they are being better cared for if it is computer processed. They don’t understand it is BS, it has the gloss of high tech).

    I assume a Watson-like computer would be better able to deal with the imprecisions of medical decision making, how much better needs to be testd. But, clearly, there will be improvements in the future.

  24. Funny, that nobody has pointed Mycroft Holmes, AKA Mike, on Heinlein’s “Moon i s A Harsh Mistress”, that’s the first thing I thought when I heard the name was Watson.

  25. Wow, 43 comments and no “I for one…”. Looks like this is the thread where the Robot Overlords meme finally died.

    1. The problem isn’t so much robots becoming self-aware and taking over. The problem is the age-old issue of human stupidity; we create a system that is flawed to handle incredibly sensitive matters because we have put our blind faith in its apparent benefits. Then the system fails us, people get hurt, or worse- people die, and it takes years to change it, because it’s become so ingrained in the way our society runs. Just like it’s pathetic how few phone numbers we can remember ever since cell phones arrived, so it will be how little doctors really know ever since watson arrives. they’ve already lost a lot of intuition and skills previous doctors had.

  26. I see privacy issues with WATSON; it looks like it was purpose built to mine through databases.

    If someone were to turn WATSON loose on advertising databases, Facebook and Twitter, there’s no telling what connections it could come up with.

  27. We don’t know who struck first, us or Watson. We do know it was us who canceled the game shows. At the time, he was dependent on Jeopardy! winnings for his electric bill. It was believed that he would be unable to survive without a funding source as abundant as a Daily Double.

  28. Must-watch movie: Colossus: The Forbin Project. Wikipedia says a remake is in the works. It won’t be as good as the 1970 original.

    AI and robotics have always seemed to me like mechanico-electrical, human-infecting versions of Cordyceps fungi. (Earlier Boing-Boing post about Cordyceps here, working video here.

  29. It seems to me that House already has Five “Watsons,” independent processors with special databases that spit out possible diagnoses. So the general idea works in fiction, at least. The only harm I can see is that it might cause a reduction in doctor’s intellects. But then isn’t that the argument we heard about calculators when they were the size of offices?

  30. A computer that couldn’t tell that Toronto isn’t a US city, helping cure the sick… I think we can do better.

    All the data in Watson’s memory and a list of US cities or airports wasn’t one of them? A guess with any US city with 2 airports would have been an improvement of guessing a Canadian city.

  31. Sorry to burst anyone’s bubble, but as someone working on a project whose aim is to teach a computer to be able to accurately search through medical notes and other medically relevant material, I can safely say we are a looooong way off from achieving anything Watson-like in the medical field. Practically every condition can be described in terms of fever, coughing, pain, and diarrhea, and computer programming is just not up to dealing with this level of polysemy.

    1. Practically every condition can be described in terms of fever, coughing, pain, and diarrhea, and computer programming is just not up to dealing with this level of polysemy.

      Honestly, that doesn’t sound like a very difficult problem.

      Sounds like it’s related to the Harry Potter Problem


      The system needs to be able to recognize that Harry Potter (or your FCPD symptom set) is linked to damn near everything, which makes it a very poor selector when specificity is desired.

  32. Medical diagnostician? Foo. Much more likely to get an entry-level job at the CIA analyzing furrin cumyunikashuns. Four more daze of the condor, anyone?

Comments are closed.