Slashdot Mirror
Layoffs at Watson Health Reveal IBM's Problem with AI (ieee.org)
Last month IBM, which has staked much of its future on its flagship AI Watson, announced a major round of layoffs in the division. Now the engineers who had been let go allege that the move shows that difficulties IBM is facing in turning its AI into a profitable business. A report on IEEE Spectrum says: "IBM Watson has great AI," one engineer said, who asked to remain anonymous so he wouldn't lose his severance package. "It's like having great shoes, but not knowing how to walk -- they have to figure out how to use it." The layoffs at the end of May cut a swath through the Watson Health division. According to anonymous accounts submitted to the site Watching IBM, the cuts primarily affecting workers from three acquired companies: Phytel, Explorys, and Truven. These companies, acquired between 2015 and 2016, brought with them hefty troves of healthcare data, proprietary analytics systems to mine the data for insights, as well as their customers. The report adds: Two laid-off engineers from Phytel spoke to IEEE Spectrum in depth. They allege that IBM's leadership mismanaged their company since its acquisition, and say the problems at Phytel are emblematic of IBM's struggles to make Watson profitable. Several other Phytel employees corroborated the basic facts of their accounts. Both engineers worked for Phytel since before its 2015 acquisition, and say they were excited to become part of Big Blue. "Everyone expected that we would join IBM and be propelled by their support, that it would be the beginning of great things," says the first engineer.
>> (company) acquires companies, fires acquired employees
This is news because? This is how the world works.
My irony alarms are going off ... layoffs at the AI division!
All the marketing hype about "AI" didn't translate into sales. Here is how to tell a technology doesn't work: Someone comes out with a technology that does X (like plays Chess) and says yeah, but it could be applied to do useful things Y and Z. The question needs to be asked: then why not demonstrate it doing useful things Y and Z? The reason is because it doesn't work. It is really good at X, but they don't know exactly how it can do Y and Z. Meanwhile, billions have been spent on marketing campaigns. Typically taxpayer money is thrown in to support the farce.
Because in this case, IBM strategy was buying competency in a market they had no footprint in. Not buying a competitor to shutter (which is bad enough).
They proceeded to slap the Watson brand on it, despite no relationship whatsoever to the technology or the people carrying the Watson brand.
This is another chapter in IBM floundering about with the Watson brand, unable to make it profitable after the publicity stunt of the Jeopardy game back in 2011. They've done everything from touting it's ability to generate recipes to medical diagnosis (several times on the medical front, with unrelated technologies at different points in time).
It's indicative of IBM's general annoying tendency to promise that initiative 'X' is going to really turn things around, and then when 'X' is clearly failing to do so, then they rename something completely different 'X' and try again, and rinse and repeat until they (hopefully) can declare 'see, "X" did turn things around!"
XML is like violence. If it doesn't solve the problem, use more.
IBM has a far better record of destroying anything it partners with or acquires.
For the old timers think of Timeplex, Taligent, Pink, OS/2.
About the only things they seem to have going these days are large consulting contracts that are acquired through political connections not technical merit, and legacy Z system support which is still to expensive to migrate.
Because these guys are passionate about what they were doing and they want to share how they feel their hard work is getting screwed up.
XML is like violence. If it doesn't solve the problem, use more.
>> IBM strategy was buying competency in a market they had no footprint in...proceeded to slap (IBM) brand on it
Again, IBM SOP. If IBM (or CA or any other low-contribution, high-licensing-fee behemoth) buys your company, it's usually best to just find a new company and wait for the buyout package.
I've worked with a number of companies acquired by IBM, and what you say is mostly true. This time was unusual as they *really* did seem to try to use that team for a few years before ultimately giving up. They seemed to genuinely think they were unable to win without those people, then the sentiment seemed to become they can't win even with those people, not that they were winning and could cut the team as a result.
XML is like violence. If it doesn't solve the problem, use more.
Back in the old days, the saying used to be "you'll never get fired for buying IBM".
Today? If I was a manager and someone uttered as many as 2 of the 3 letters...I'd can that fscker right on the spot.
Light travels faster than sound. This is why some people appear bright until you hear them speak.........
The problem is that they focused on the wrong industry. Healthcare in America is bloated and inefficient at 18% of the economy, double any other country. But, while many other industries are "bottom heavy" with plenty of powerless assembly line workers or clerks whose jobs can be automated away or shipped overseas, healthcare is "top heavy" with a vast number of professionals represented by powerful organizations.
Decades ago it was obvious that many doctors could be replaced, since a nurse using a paper checklist could diagnose with the same accuracy. This is exactly what was done in many countries, with nurses or PAs handling the routine cases, while referring the difficult cases to MDs. But in America, we instead got an institutional resistance to any reform that could reduce profits. There are no incentives for doctors, or patients, or insurance companies to control costs. It is no surprise that IBM was not able to change this. What is surprising is that they thought they could.
>> many doctors could be replaced, since a nurse using a paper checklist could diagnose with the same accuracy. This is exactly what was done in many countries, with nurses or PAs handling the routine cases, while referring the difficult cases to MDs. But in America, we instead got an institutional resistance to any reform that could reduce profits
Don't know if I agree. Within the past ten years (and corresponding to a massive increase in cost of care) I've seen a lot of "Nurse Practitioners" step in where I used to see doctors with the kids when they were younger.
>> There are no incentives for doctors, or patients, or insurance companies to control costs
Maybe not doctors, but there are now tremendous incentives for companies and employees to control costs. So, we use the $XX company nurse/doctor instead of the $XXX option available under our health care plan. And we try to stay as far away as possible from any Urgent Care or ER services.
What seems to be driving the massive cost increase is that free care is increasing, government payments are capped, and the surviving health care providers are squeezing the last handful of us who can still pay the bills. Meanwhile, no hospital system wants to stop building hotel-quality hospitals with huge atriums ("atrium...get it?") and chasing plastic surgery patients (which requires keeping up appearances), so they keep spending like coke addicts. All we're asking for out here in the middle class is change, which is part of the reason Bernie almost won and then Trump did win. Unfortunately...
"IBM Watson has great AI," one engineer said, who asked to remain anonymous so he wouldn't lose his severance package. "It's like having great shoes, but not knowing how to walk -- they have to figure out how to use it." The layoffs at the end of May cut a swath through the Watson Health division.
The path to money is replacing doctors. Duh. Partner with insurance companies that PAY YOU to make doctor Watson the initial contact for their customers.
You call in, or log in and give it your symptoms, answering it's questions (Maybe sending it photos). It gives you a diagnosis, advice, a prescription, and/or refers you to a specialist or a bloodwork lab.
That means giving Watson the authority of a real doctor, that can actually DO things like diagnos and write prescriptions. And that means making it liable for fucking up. If Watson is good, then that should be viable.
This makes money for Watson and that dev team, they get paid by the insurance companies. This makes money for the insurance companies as the rate for one computer on the Internet is hella cheaper than all those general practioners. This is a general improvement for the customers as their doctor is on tap 24/7 and doesn't have a hideous co-pay.
This screws over doctors who will undoubtedly become neoluddites or further specialize.
This screws over Insurance companies that don't actually want you to make use of them. They make money off of healthy people that carry the policy just out of fear. If you actually USE the insurance, the insurance company loses money. To that extent, health insurance companies anything that makes the health-care system simpiler and easier to use. And thus we see why IBM can't appear to sell Watson.
"Decades ago it was obvious that many doctors could be replaced, since a nurse using a paper checklist could diagnose with the same accuracy"
I'm not aware of any studies of substance that substantiate such a bold claim. I am aware of some weak studies looking at NPs managing chronic diseases that had already been diagnosed. Perhaps that is what you are referring to?
"This is exactly what was done in many countries, with nurses or PAs handling the routine cases"
Really? Most health care providers in Europe and Australia have never heard of or worked with a PA or NP. Further, I've never heard of their nurses making diagnosis or managing diseases.
I work for an international health care company so I have some idea of the lay of the land here. If you have some sources to back this up, I'm interested. Otherwise, I'm fairly sure this is just BS.
... and chasing plastic surgery patients ...
It is ironic that you mention cosmetic medicine as a reason for runaway costs. Actually, the opposite is true. Cosmetic medicine is one area that has NOT seen costs skyrocket, and for many treatments prices have declined. The reason is that cosmetic treatments are generally not covered by insurance so patients are paying out of their own pocket. Prices are listed upfront, and the prices are often openly published in advertisements, something that is not done in almost any other area of medicine. It is one of the few areas of healthcare with a functioning free and competitive market.
Another area that has seen dramatic declines is corrective vision treatments. I paid $3000 for LASIK back in 1999. Today, the same clinic charges $299 per eye. This is another area generally not covered by insurance, and with upfront pricing.
It's called revenge. If you were doublecrossed by an organization, then it's not unlikely that you would want to dish dirt on them *and* take their money.
What possible benefit could there be to posting on Slashdot? Yet you are doing it.
You can't come up with a reason for posting - so there's none?
I post on Slashdot specifically to practice writing and debating skills. It gives immediate feedback, so meshes well with Gladwell's "20,000 hours of practice" theory.
I also post to help innoculate myself against insults and reduce my dependence on "what other people think", a character flaw of mine that many people have. (Note: I don't consider your response matching one of those.)
I can understand the need to talk, and for social empathy and all that.
It's just that to make a promise and enter into a binding agreement (for money) and then *break* that promise...
Is the need for social empathy so great that people will take a chance on ruining their lives for it?
It doesn't *seem* like a good trade-off, but then I'm less on the emotional side of things than most people.
There is the old saying that if you think you know the solution, you do not understand the problem. That seems to apply to a lot of technology acquisition in the health fields.
Here is the problem. In medicine, there are some disorders and diagnoses that are very well understood in terms of physiology and pathogenesis, well characterized as to making a correct diagnosis that correlates with an effective response to prescribed treatment, and easy to teach to young physicians who can in turn provide accurate diagnoses and good care to people with those problems.
However, there are also problems that are oddball or non-obvious diagnoses, problems due to occult or infrequent disorders, problems with atypical symptoms or atypical responses to intervention, or else problems that reflect altered or atypical pathogenesis or else disorders of complexity or dysdynamia in complex multi-control systems in which one person's illness has different signs and symptoms than the next person with the same illness. If compared to algorithmic processes such as computer programming, some of these patients and problems would be seen as illogical, out-of-sequence, or data corrupted, yet a valid diagnosis can be made by someone who understands these deeper levels of seeming illogic. Whether or not contemporary doctors and medical education still rise to the challenge is another issue altogether, but when medicine is done right by smart properly educated physicians, correct diagnoses and treatments can be made for very non-obvious problems. This is because genuine intelligence is better than artificial intelligence at doing these non-obvious complex tasks.
But, you say, therein is the value of AI, indeed the whole premise of AI, that it can perceive patterns and associations in data that even smart people will not necessarily see. That may be true, but AI can be no better than the set of data it is trained on. The article states "these companies . . . brought with them hefty troves of healthcare data, proprietary analytics systems to mine the data for insights". Big databases from corporate healthcare enterprises do indeed have lots of data , but it is not necessarily quality or robust or relevant data. It is the kind of perfunctory or bulk data that is filled out into forms, or is coded in ICD and CPT numbers (industry standard diagnosis and procedure codes). The data is curated or trivialized to what can be entered by overworked professionals in order to generate bills, or else by low level billing and data clerks. It is often data that is not relevant to the technical medical issues, and even when it is, it is not the detailed or nuanced data that allows for the oddball, atypical, and one-off diagnoses.
Suppose there are 7 different disorders that can affect the pinky toe, 3 of which are common and readily recognized by medical students and physician extenders, 2 more of which are recognized by the average properly educated physician, and 2 of which are odd and likely to be recognized only by experienced experts. The difference in diagnosis for these latter two might be because the toe points at an angle five degrees different than normal. If the data keeping records have only 4 approved codes to recognize 7 diagnoses, and if there is no place to record the angle of the toe, then the AI training set will not be able to understand the oddball diagnoses. Note as well that the data entry front ends that are often in modern medical records depend on the easy info. The problem is that the easy diagnoses can already be made by people with a baseline education. The one-off diagnoses depend on levels experience of multi-factorial observation and pattern recognition that real experts have but which the medical record is often scant on.
Technology has become a self-indulgent plaything for companies and venture capitalists. "Let's automate or computerize this or that . . . because we can. Isn't this fun?" If you develop enough infrastructure or visibility or limited success to beguile the next dog up th