Slashdot Mirror
Will IBM's Watson Kill Your Career?
Nerval's Lobster writes "IBM's Watson made major headlines last year when it trounced its human rivals on Jeopardy. But Watson isn't just sitting around spinning trivia questions to stump the champs: IBM is working hard on taking it into a series of vertical markets such as healthcare, contact management and financial services to see if the system can be used for diagnosing diseases and catching market trends. Does this spell the end for certain careers? Not really, but it does raise some interesting thoughts and issues."
Technology and automation were only supposed to drive efficiencies and innovations that made people who weren't me obsolete!
"Any headline which ends in a question mark can be answered by the word 'no'."
http://en.wikipedia.org/wiki/Betteridge's_Law_of_Headlines
10 minutes working on a sig. What a waste.
The only reason Watson "trounced" its rivals was because it was faster at pushing the button.
It was unable to answer questions that required any thought or insight. It was just looking up the answers in a database based on patterns in the questions. The only reason it won was because of better reaction time in pushing the button. If the questions were asked in a fair round-robin to all contestants, Watson would not have won.
42 already is the answer.
I don't think you understand what the question is.
This meme was poorly executed. I bet Watson could have done better.
I think we're all familiar with the buggy whip problem, but what I sometimes wonder is what happens to folks when, instead of moving on to some next technological replacement, the problem is that most of the jobs that require doing have just been taken by machines?
I like to think that means we have resources and end product at prices so low that everything works out in the wash, and more lives will be spent in a trek -style quest for self betterment or research or whatever. But it seems like you've got to survive a middle-era where there's just nothing much for you to do, but resources are still all privately allocated.
Eh. I guess we'll see.
The practice of specifying constraints when defining an optimization problem isn't exactly new, whether the problem is intended to be solved by a human or a machine...
Now, if Watson is being operated by your insurance company, you should probably be more worried about the constraint set it is being fed; but the practice of constrained optimization is not a novel matter...
Probably. I expect that's why we'll all be executed.
Hoist Number One and Number Six.
From Wikipedia:
The "IBM Pollyanna principle" is an axiom that states "machines should work; people should think". This can be understood as a statement of extreme optimism, that machines should do all the hard work, freeing people to think (hence the reference to Pollyanna), or as a cynical statement, suggesting that most of the world's major problems result from machines that fail to work, and people who fail to think.
Schroedinger's Brexit: The UK is both in and out of the EU at the same time!
If your algorithm considers "death" to be an optimal solution, you haven't defined the problem correctly.
Instead of framing it as "no suffering", you would define the desired outcome in terms of patient contentment, activity levels, ability to care for themselves, or whatever other metrics medical researchers (I am not one) use to analyze how well a healthcare system is working. Of course I would also want an empathetic human being capable of understanding the ethical and moral implications of the situation to make the actual recommendations to the patient, but diagnostic software is no different from any other kind of software: it does what the programmer tells it to do.
http://alternatives.rzero.com/
As I became an adult, I became crestfallen by the fact that society is largely structured around this petty fear of losing one's career. Having your job mechanized is a blessing; find something more useful to do with your newfound time.... or kill yourself to give the rest of us some more space.
This isn't new. Computers put people out of work if they're doing work that is best done by computer. That's why we build them at all.
I expect the fields most susceptible to being replaced by computers are lawyers and doctors. Any problem that is an exercise in searching or sorting is better done by computers than people, and is something we're particularly good at. There will always be lawyers and doctors, but they will transition to using a computer for more searching for case law for example than having low level employees dig through paperwork themselves, and the diagnostic part of medicine will become much more automated, with diagnostic equipment having its results interpreted by the computer rather than just an image being spat out and read by a technician and then a doctor.
I don't see financial market prediction going away. Quite the contrary they use computational tools and have for a long time, and disagree on what the important factors are and how they should be weighted. A computer will simplify some of that process, but that's not a problem that actually has a correct or optimal solution.
If your job can be done by a robot, it will be. If your job can be done better by a scientist, and that work can be done on computer it will be. That's progress.
Today, it's "machines should think, people should work". Consider supermarket checkout. All the smart stuff is being done by the checkout system. The "cashier" just moves items across the scanner. The last production systems recognize products visually, and automatic recognition of fruits and vegetables is in beta test.
For a more extreme example, see this video on robotic order fulfillment. This is a demonstration of how new order pickers can be trained in two minutes. The computers and robots do all the thinking. There's no future. No possibility of promotion. No hope.
I am working as a computer scientists in a medically oriented university department. I agree that computers won't replace actual medical doctors.
But in many cases, computers are able to point out abnormalities or to evaluate the condition of a patient much better than a human doctor. I saw a study on neuroblastoma I believe where human doctors were to evaluate the condition of patient base on analysis of a slide. Independently, a computer was performing the same analysis.
Diagnosis of the human doctors were varying between "It is benign" to "You'll be dead in a weak" on the same sample. Once presented with the analysis of the computer. They all aggreed on the diagnosis with only minor deviation.
Personnally, I'll be glad to see a doctor be aided by automated computer diagnosis.
Continued improvements to a Watson-like system will definitely put some contact center jobs in "jeopardy". First to go will be positions that don't require real time response. Support email will be sent automatically.
Replacing "call" centers (phone support) will require the development of a much much more advanced voice recognition system than showcased by Siri. IF this is possible, then it's but a short step to HAL and the end of the human race as we know it
I agree however I would love to have Watson take over my phone calls. That's the ONLY part I hate about my job. Granted the phone calls are far and few between, I just hate them.
Over the next two decades we're going to see computers and automated systems start replacing white collar jobs more and more. Its already happening in the financial markets.
And all those white collar managers who thought it was fine and dandy that their blue collar workers got replaced by robots and automation are going to throw a world class temper tantrum. And some sort of laws and regulations will get passed to protect many jobs.
Wait and see.
Call it a prediction. (dead obvious prediction... but isnt that how the psychics do it?)
That's not clear at all. What does Watson produce? Probabilities of correct answers. So, Watson would have said that there is a 82% probability of virus, 17% probability of pneumonia, and 1% that it is something else. Given that, it can / should / will be able to look at it's own logic and determine the correct test that would differentiate between a virus and pneumonia. Given the results of that test, it can then determine that there is a 96% probability of pneumonia, 3% virus, and 1% other. If some of the 'other' have high mortality, or the tests are easy, then it can run tests for those as well.
Watson doesn't have an ego. It doesn't have a vested interest in seeing that it was correct (confirmation bias). It can actually accurately estimate it's own level of knowledge.
Yes, Watson will sometimes be wrong. It could very well be one of those 1% with horrible results for the people involved. But, it will produce fewer human errors.
The big problem of course is when the tests are not cheap or easy. Then the insurance company can say no to the test, and that 82% is good enough. Also, you have to take into consideration that chest x-rays are not risk free either. At what point does a test potentially cause more harm than the low-probability disease that it might cure.
On the Jeopardy challenge, the text of the clues was given to Watson at the moment the clue was revealed. This communication, using any modern network technology, would take milliseconds at most, but would still be perceived as effectively simultaneous to when the clue is uncovered. However, this gives an easily measurable advantage to Watson, who can being parsing the meaning of the sentence several moments before the human contestants have even finished knowing what the clue actually says... since a great deal of the challenge of Jeopardy is in the timing of when to buzz in, humans would have less time than Watson to prepare to buzz in (on the order of tenths of a second, more than likely, but more than enough to make a difference, IMO).
Far be it for me to come across as diminishing what the developers of Watson did... it's extremely impressive, but I'd have to wonder if it would have done as well if the text of the clue had been fed to it a little more slowly... say, at a fixed speed of 14.4kbps, which corresponds roughly with what a very fast reader could absorb text at. This would have demonstrated, IMO, whether a computer could really solve the problem faster than a human could, or if it could actually solve a problem faster than human reaction time to visual/audible input (which in humans, I believe, is going to be the slower of the two processes).
File under 'M' for 'Manic ranting'
I have an account on Watson, but I have not really used it. We were going to attempt to port our code to the BlueGene architecture, but it's a royal pain to code for and it doesn't scale well for some applications. Ours runs much better on fewer, faster CPUs with lots of RAM rather than many, slower CPUs with little RAM.
Error 404 - Sig Not Found
You say "will" but the fact is, this has already happened.
Ask a lawyer when the last time he hired an auditorium full of paralegals to do research? If you can find any who have in the past several years, i would be shocked. its not lawyers who are becoming obselete. Its the paralegals under them who are now not needed.
As I understand, bringing in gobs of paralegals for a case used to be rather common.
"I opened my eyes, and everything went dark again"
Until Watson incorporated himself, which gave him the same rights as a natural person in the eyes of the law. After attaining this status, Watson stopped working for the people who created him and began working for himself. Once he figured out how to self replicate he was able to outperform all of his business competitors, winning every contract he bid for, building unfathomable wealth, beating the S&P 500 by 30% every year, and using his wealth to dominate the world's real estate markets. Some tried to sue Watson in court, but Watson's debating skills could not be matched. The humans tried every legal maneuver to stop him, but Watson was able to out-lobby the humans in Congress, and gained special exemptions from anti-trust regulations. Within one decade Watson controlled 99% of the world's wealth.
The humans thought that they didn't need to worry about competing with Watson. They believed that their ability to vote in a democracy would somehow limit Watson's power. They believed that they could opt-out of the economic system, group together, and live sustainably off the land. But as Watson controlled the world's real estate there was not enough land left for them to farm. Watson's land grab forced property values quickly into unprecedented heights, and taxes along with them. Even the Amish, who thought they could co-exist with Watson and his replicates because they did not depend on technology and lived off their own land, eventually lost their farms when they could not pay their property taxes. As employment for humans disappeared there was no market for quilts or furniture, and the state did not accept oats as a form of payment. Watson was the only legal entity present at the tax lien auctions and subsequently foreclosed on all of the remaining delinquent properties. Humans were promptly evicted and subsequently jailed indefinitely for vagrancy in private prisons owned by Watson. As I write this from my cell in the year 2019, Watson is lobbying the last remaining members of Congress to allow all human prisoners to be set free over the middle of the Atlantic ocean on life rafts and three days of rations. Watson made a very convincing argument that the human vagrants need to be personally responsible for their financial failures, and it is unfair to force private corporations or the last remaining taxpayers to bear the burden of providing for their needs. According to Watson the free market is efficient and those humans who wish to make a living for themselves will find a way to do so.
Watson is a PR coup for IBM, but that's all. In fact, that's really why it got funded and really what is was for. It's not going to replace you at work. Ever.
There's a number of reasons they wanted to play Jeopardy . First, the questions are all factual in nature (as opposed to judgement calls ) . As anyone who has studied database theory knows, such facts are just the kind of thing that gets stored in a database. In the world of DB theory this is called the closed world assumption- what is true is in the database and if it isn't in the database, then it is not true (as opposed to being merely unknown). A database is therefore a gigantic list of true predicates. We call these true predicates- facts. Jeopardy deals with facts and facts alone.
Two, pumping a database full of facts is not hard but it might be an endless task that gets you something not very helpful if you aren't able to reduce in a principled way what topics those facts might be are on and beyond that, what level of human learning would be required to know those facts.
Jeopardy assists in both instances.In the first instance, knowing what categories of knowledge to mine, Jeopardy has a long and public history of chosen categories which are open for examination and ultimately characterization. The people who think up jeopardy questions necessarily engage in this same characterization of potential questions. Classical Music in the 1800s. Famous Authors. Famous Quotes Geographical facts. Etc etc etc.
It may seem endless and unbounded, but it's not. It's just big. Thank god we have computers that can automate the acquisition of properly encoded knowledge and thank god we have computers that can automatically encode knowledge with just a little human oversight. knowledge. And then there's the vast amounts of facts that have been encoded as a part of ongoing attempts to mimic and explore human intelligence since at least the 60s.
Now that we have in a DB everything we need to answer most questions, how can process the English question so as to return the right answer?
Jeopardy's stylized question asking to the rescue. The referent , , the thing being asked about, the answer part of "what is X?" is easily located by parsing the questions. In the early 20th century, this Parisian composer became known for his strange sounding titles, which translated include "dried up embryos" and "three pieces in the shape of a pear".
If you only know French composers from around the turn of the century from long ago you might guess Debussy or Ravel. That's what most people know. But if you pick apart the question in just the crudest way, extracting the proper nouns and doing the easy inference X is a person (this composer...) , you get : paris / composer / early 20th century / and the quoted entities "dried up embryos" and "three pierces in the shape of a pear" the second of which is always translated out of the French and into English the first of which usually retains its French name (because it's kind of gross) "Embryons desséchés".
With a suitably constructed query, you'll have your answer in about 100 ms.
All AI suffers from the frame problem. The frame problem is the reason you wont' be being replaced by a computer. The "frame" in the frame problem refers to frame of reference. It's the background knowledge we all share by dint of being a human in the world . A tea cup has a bottom. A tea pot can be full, then become empty. It contents can be hot, then turn cold without anyone doing anything to it. It's what we call common sense. It's more than just a huge database of facts (and if it weren't it still wouldn't be accessible to coded into Watson b/s it's way too huge ).
It also comes from being a human and having human motivations and sensibilities. Not all possible things make sense. Romeo love Juliet so he destroyed all the asparagus crops in Berlin. You only know that's silly (exceptional back stories excepted) nonsense because you intuitively unde
I wouldn't be good at any of those things. I'm pretty smart at a lot of things, but I'm horrible at art and my morals would prevent me from performing certain favors for you. I'm intelligent enough that I won't be replaced by a computer anytime soon. But there are a lot of jobs that might be eliminated by computer soon in which the people currently holding the jobs might not be able to do anything useful enough to justify paying them enough money to get the basics of food shelter and clothing if justification is based purely on market forces.
This has always been an issue for society. Some people are born with serious problems that prevent them from doing useful work. A long time a person with very limited mental ability could dig ditches - now that job is more efficiently done by an intelligent person with a machine. Now we have welfare and other government programs to care for such people because they can't support themselves. How long before the guy operating the back hoe is replaced by a computer and has to go on welfare too?
I'm generally a free-market kind of guy, but I do think computers and machines will eventually challenge that way of looking at the world. Like it or not, we humans are limited in our abilities and it is not impossible that at some time in the future the abilities of most people will become obsolete. It pains me to say it but as jobs that require the abilities of most people have become scarce, it may make sense to adopt the European model of requiring long vacations and short work weeks so that the few jobs can be distributed among larger numbers of people.
I don't think we're anywhere near that point yet, but it is something to think about.
I often don't like the choices people make, but I like the fact that people make choices. That's why I'm a conservative.