Wat­son’s next feat? Tak­ing on can­cer.

IBM’s com­puter brain is train­ing along­side doc­tors to do what in­di­vid­u­ally they can’t

The Washington Post Sunday - - FRONT PAGE - BY ARIANA EUNJUNG CHA IN HOUS­TON

Can­dida Vi­tale and the other fel­lows at MD An­der­son’s leukemia treat­ment cen­ter had known one another for only a few months, but they al­ready were very tight. The nine of them shared a small of­fice and were al­ways hang­ing out on week­ends.

But she wasn’t quite sure what to make of the new guy.

Ru­mor had it that he had fin­ished med school in two years and had a pho­to­graphic mem­ory of thou­sands of jour­nal ar­ti­cles and rel­e­vant clin­i­cal tri­als. When the fel­lows were asked to sum­ma­rize pa­tients’ records for the se­nior fac­ulty inthe morn­ings, he al­ways seemed to have the best an­swers.

“I was sur­prised,” said Vi­tale, a 31-year-old who re­ceived her MD in Italy. “Even if you work all night, it would be im­pos­si­ble to be able to put this much in­for­ma­tion to­gether like that.”

The new guy’s name was a mouth­ful, so many of his col­leagues sim­ply called him by his nick­name: Wat­son.

Four years af­ter de­stroy­ing hu­man com­peti­tors on “Jeop­ardy!”

to win a sus­pense-filled tour­na­ment watched by mil­lions, the IBM com­puter brain is ev­ery­where. It’s done stints as a call cen­ter op­er­a­tor and ho­tel concierge, and has been spot­ted help­ing peo­ple pick songs. It’s even pub­lished its own cook­book, with 231 pages of what the com­pany calls “recipes for in­no­va­tion.” (The re­views haven’t been flat­ter­ing — one foodie de­clared one of Chef Wat­son’s cre­ations “the worst bur­rito I’ve ever had.”)

But these feats were es­sen­tially gim­micks.

IBM is now train­ing Wat­son to be a can­cer spe­cial­ist. The idea is to use Wat­son’s in­creas­ingly so­phis­ti­cated ar­ti­fi­cial in­tel­li­gence to find per­son­al­ized treat­ments for ev­ery can­cer pa­tient by com­par­ing dis­ease and treat­ment his­to­ries, ge­netic data, scans and symp­toms against the vast uni­verse of med­i­cal knowl­edge.

Such pre­ci­sion tar­get­ing is pos­si­ble to a lim­ited ex­tent, but it can take weeks of ded­i­cated sleuthing by a team of re­searchers. Wat­son would be able to make this type of treat­ment rec­om­men­da­tion in mere min­utes.

The IBM pro­gram is one of sev­eral new ag­gres­sive healthcare projects that aim to sift through the huge pools of data cre­ated by peo­ple’s records and daily rou­tines, and then iden­tify pat­terns and con­nec­tions to pre­dict needs. It is a rev­o­lu­tion­ary ap­proach to medicine and health care that is likely to have sig­nif­i­cant so­cial, eco­nomic and po­lit­i­cal con­se­quences.

Lynda Chin, a physi­cian­sci­en­tist and as­so­ciate vice chan­cel­lor for the Univer­sity of Texas Sys­tem who is over­see­ing the Wat­son pro­ject at MD An­der­son Can­cer Cen­ter, said these types of pro­grams are key to “de­moc­ra­tiz­ing” med­i­cal treat­ment and elim­i­nat­ing the dis­par­ity that ex­ists be­tween those with ac­cess to the best doc­tors and those with­out.

“I see tech­nol­ogy like this as a way to re­ally break free from our cur­rent health-care sys­tem, which is very much lim­ited by the com­mu­nity providers. If you want ex­pert care, you have to go to an ex­pert cen­ter,” she said, “but there are never enough of those to go around.”

In­stead of hav­ing to find spe­cial­ists in a dif­fer­ent city, pho­to­copy and send all the pa­tient’s files to them, and spend count­less hours re­search­ing the med­i­cal literature, a doc­tor could sim­ply con­sult Wat­son, she said.

Jho Low, the 33-year-old bil­lion­aire who is bankrolling MD An­der­son’s $50 mil­lion Wat­son pro­ject, said the ef­fort gre­wout of his grand­fa­ther’s treat­ment for leukemia in Malaysia. Low said that he felt for­tu­nate to be able to con­nect his grand­fa­ther’s doc­tors re­motely with MD An­der­son spe­cial­ists to de­vise the best treat­ment plan. He be­lieves ev­ery­one, rich or poor, should have the same ac­cess to that kind of ex­per­tise.

“This is very per­sonal to my fam­ily. It is re­ally some­thing we have gone through and seen what kind of dif­fer­ence it can make,” said Low, who is a grad­u­ate of the Whar­ton School at the Univer­sity of Penn­syl­va­nia and runs one of Asia’s most suc­cess­ful in­vest­ment firms.

Low is part of an in­flu­en­tial new move­ment in sci­en­tific re­search driven by young phi­lan­thropists and tech ti­tans who have faith that the chips, soft­ware pro­grams, al­go­rithms and big data that pow­ered the in­for­ma­tion revo­lu­tion can also be used to un­der­stand, up­grade and heal the hu­man body.

But the Wat­son pro­ject and sim­i­lar ini­tia­tives also have raised spec­u­la­tion — and alarm — that com­pa­nies are seek­ing to re­place the na­tion’s ap­prox­i­mately 900,000 physi­cians with soft­ware that will have ac­cess to ev­ery­one’s sen­si­tive per­sonal health in­for­ma­tion.

While there’s much de­bate about the ex­tent to which tech­nol­ogy is de­stroy­ing jobs, re­cent re­search has driven con­cern. A 2013 pa­per by econ­o­mists at the Univer­sity of Ox­ford cal­cu­lated the prob­a­bil­ity of 702 oc­cu­pa­tions be­ing au­to­mated or “roboti­cized” out of ex­is­tence and found that a star­tling 47 per­cent of Amer­i­can jobs — from par­ale­gals to taxi driv­ers— could dis­ap­pear in com­ing years. Sim­i­lar re­search by MIT busi­ness pro­fes­sors Erik Bryn­jolf­s­son and An­drew McAfee has shown that this trend may be ac­cel­er­at­ing and thatwe are at the dawn of a “sec­ond ma­chine age.”

Sci­en­tists are al­ready test­ing baker bots that can whip up pas­tries, ma­chines that can teach math in the class­room and ro­bot anes­the­si­ol­o­gists.

Many physi­cians and aca­demics in medicine have come to view Wat­son’s work with reser­va­tion, de­spite re­as­sur­ances from IBM of­fi­cials that they are try­ing not to re­place hu­mans but to help them do their jobs bet­ter.

“I think a lot of folks in medicine, quite frankly, tend to be afraid of tech­nol­ogy like this,” said Ilti­fat Husain, an as­sis­tant pro­fes­sor at the Wake For­est School of Medicine.

Husain, who di­rects the mo­bile app cur­ricu­lum at Wake For­est, agrees that com­puter sys­tems like Wat­son are likely to vastly im­prove pa­tients’ qual­ity of care. But he is em­phatic that com­put­ers will never truly re­place hu­man doc­tors for the sim­ple rea­son that the ma­chines lack in­stinct and em­pa­thy.

“There are a lot of things you can de­duce by what a pa­tient is not telling you, how they in­ter­act with their fam­i­lies, their mood, their man­ner­isms. They don’t look at the pa­tient as a whole,” Husain said. “This is where al­go­rithms fail you.” Wat­son’s evo­lu­tion

Named af­ter Thomas J. Wat­son Sr., IBM’s first chief ex­ec­u­tive, Wat­son was de­signed to be a sub­stan­tial leap for­ward from Deep Blue, the su­per­com­puter that beat chess grand­mas­ter Garry Kas­parov in a marathon three­day man vs. com­puter match in 1997.

Deep Blue’s edge was brute force. It had the abil­ity to cal­cu­late and an­a­lyze up to 200 mil­lion sce­nar­ios per sec­ond— a skill that could be ap­plied to com­plex cal­cu­la­tions as di­verse as mod­el­ing the stock mar­ket and rank­ing the po­ten­tial of small mol­e­cules for new drugs.

But the pro­gram was hand­i­capped by its in­abil­ity to per­form skills that hu­mans master in their first few years of life. It couldn’t make sense of reg­u­lar hu­man speech or any other type of so-called un­struc­tured data or in­for­ma­tion that isn’t or­ga­nized ac­cord­ing to a pre­de­fined for­mula like a chart or ta­ble.

Given that the world is a messy place when it comes to data — from the text of Shake­spearean plays to traf­fic pat­terns in Los An­ge­les — Deep Blue’s abil­i­ties were lim­ited.

Wat­son was imag­ined from the start to be more hu­man.

One of the top pri­or­i­ties for pro­gram­mers was to give Wat­son the power to read and un­der­stand nat­u­ral lan­guage. They also gave it the abil­ity to gen­er­ate hy­pothe­ses and lo­cate and parse ev­i­dence to sup­port or re­fute them.

Much like the hu­man brain, Wat­son has be­come smarter over time by learn­ing from its suc­cesses and fail­ures and from user feed­back. Wat­son is lit­er­ally evolv­ing. In the be­gin­ning, Wat­son’s knowl­edge base was lim­ited to trivia for “Jeop­ardy!” But since its de­but on na­tional tele­vi­sion in Fe­bru­ary 2011, Wat­son has de­voured many thou­sands of literary works, news­pa­per ar­ti­cles and sci­en­tific jour­nal re­ports as well as in­for­ma­tion in­put by hun­dreds of re­searchers and doc­tors na­tion­wide. These ex­perts have helped the ma­chine brain make more rea­son­able in­fer­ences and con­clu­sions by re­view­ing Wat­son’s ideas and telling it whether it is right or wrong and by high­light­ing which sources of in­for­ma­tion are con­sid­ered more re­li­able than oth­ers.

Un­like a hu­man brain that can be dis­tracted, con­fused or inspired by huge vol­umes of in­for­ma­tion, Wat­son is not a cre­ative thinker but a ra­tio­nal one. It looks at known as­so­ci­a­tions among var­i­ous bits of data, cal­cu­lates the prob­a­bil­ity that one pro­vides a bet­ter an­swer to a ques­tion than another and presents the top ideas to the user.

Rob Merkel, who leads IBM Wat­son’s health group, said the com­pany es­ti­mates that a sin­gle per­son will gen­er­ate 1 mil­lion gi­ga­bytes of health-re­lated data across his or her life­time. That’s as much data as in 300 mil­lion books.

“You are deep into a realm where no hu­man be­ing could ever make sense of this in­for­ma­tion,” Merkel said. That’s where Wat­son comes in to cre­ate a “col­lec­tive in­tel­li­gence model be­tween ma­chine and man.”

“We’re not ad­vo­cat­ing that Wat­son re­place physi­cians,” he ex­plained. “We are ad­vo­cat­ing that Wat­son does a lot of read­ing on be­half of physi­cians and pro­vides them with timely in­sights.”

Orig­i­nally made up of a clus­ter of su­per­com­put­ers that took up as much space at IBM as a master bed­room, Wat­son is now trim­mer — the size of three stacked pizza boxes — and ver­sions of it live in the server rooms of IBM’s var­i­ous part­ners.

IBM has so much faith in Wat­son’s in­no­va­tive­ness that in Jan­uary 2014 the com­pany an­nounced that it would in­vest an ad­di­tional $1 bil­lion in the tech­nol­ogy, and it cre­ated a new di­vi­sion to grow the busi­ness. Since then, IBM has high­lighted health care as Wat­son’s pri­or­ity and said it will ded­i­cate at least 2,000 med­i­cal prac­ti­tion­ers, clin­i­cians, de­vel­op­ers and re­searchers to the ef­fort and will part­ner with Ap­ple, John­son & John­son and Medtronic to col­lect pa­tient in­for­ma­tion that con­sumers had con­sented to share.

French bank Credit Agri­cole pre­dicted that as much as 12 per­cent of IBM’s to­tal rev­enue in 2018 could be from Wat­son -re­lated prod­ucts — with a large chunk com­ing from “con­sult­ing” fees that would be billed per use or through a sub­scrip­tion for ac­cess to its ex­per­tise.

It is Wat­son’s work in can­cer that is the most ad­vanced.

Among the most am­bi­tious projects is a part­ner­ship with 14 can­cer cen­ters to use Wat­son to help choose ther­a­pies based on a tu­mor’s ge­netic fin­ger­prints. Doc­tors have known for years that some treat­ments work mirac­u­lously on some pa­tients but not at all on oth­ers be­cause of ge­net­ics. But the ex­pense and com­plex­ity in iden­ti­fy­ing ge­netic mu­ta­tions and match­ing them up with po­ten­tial ther­a­pies has made it dif­fi­cult for more than a hand­ful of pa­tients to ben­e­fit from this new ap­proach. The ser­vice is sched-ba­sic

uled to launch later this year.

Mean­while, Wat­son is con­tin­u­ing its on-the-ground train­ing with can­cer spe­cial­ists.

In 2011, IBM an­nounced that Wat­son had learned as much as a sec­ond-year med­i­cal stu­dent. Since then it’s grad­u­ated and has been do­ing res­i­den­cies at some of the na­tion’s top can­cer cen­ters, in­clud­ing Me­mo­rial Sloan Ket­ter­ing in New York and the Cleve­land Clinic. In late Septem­ber, Wat­son achieved another train­ing mile­stone: It be­gan its first fel­low­ship in a spe­cialty — leukemia— at MD An­der­son.

The revo­lu­tion

The process of cre­at­ing the world’s first ar­ti­fi­cial-in­tel­li­gence ex­pert in can­cer starts with some­thing de­cid­edly low-tech: pa­per. Lots of it.

A team from MD An­der­son and IBM spent months feed­ing the com­puter pro­gram the names, ages and gen­ders, and med­i­ca­tions, lab tests, imag­ing re­sults and notes from each visit for thou­sands of leukemia pa­tients treated there over the past few years.

Leukemia is a can­cer that can be at­tacked in dozens of ways — in­clud­ing high-dose chemo­ther­apy and im­mune-based ther­a­pies such as tar­geted an­ti­bod­ies— and it’s of­ten tricky for physi­cians to de­cide be­tween one or another.

Wat­son — or the On­col­ogy Ex­pert Ad­vi­sor, its of­fi­cial name at MD An­der­son — was tripped up by lit­tle things at first. It some­times had trou­ble telling whether the word “cold” in a doc­tor’s notes re­ferred to the virus or the tem­per­a­ture. Or whether T2 was re­fer­ring to a type of MRI scan or a stage of can­cer. So each pa­tient record had to be val­i­dated by a hu­man.

More­over, Wat­son’s rec­om­men­da­tions were of­ten a lit­tle wacky.

“When we first started, he was like a lit­tle baby,” said Ta­pan M. Ka­dia, an as­sis­tant pro­fes­sor in the leukemia depart­ment. “You would put in a di­ag­no­sis, and he would re­turn a ran­dom treat­ment.”

It turned out that get­ting ma- chines to do sim­ple tasks hu­mans take for granted is hard. In fact, it took Google a year to teach a com­puter to be able to rec­og­nize cats on YouTube.

To teach Wat­son, the doc­tors would have to man­u­ally type in what they be­lieved the “right” course of treat­ments should be and why. They also hand­picked a num­ber of key jour­nal ar­ti­cles from the past for Wat­son to ref­er­ence and started giv­ing it ac­cess to newly pub­lished ma­te­rial.

In Oc­to­ber, the team de­cided Wat­son was ready to start its fel­low­ship.

Koichi Taka­hashi, who was at the top of last year’s class of fel­lows and re­cently ap­pointed an as­sis­tant pro­fes­sor in leukemia, said he’s been im­pressed so far.

Wat­son’s abil­ity to syn­the­size a pa­tient’s history is “amaz­ing,” Taka­hashi said. “He beats me.”

The pro­gram still sur­prises Ka­dia.

“Ev­ery once in a while you’ll see some­thing and think, ‘ This shouldn’t be.’ The other way you’re sur­prised is, ‘ Oh, my God, why didn’t I think of that?’ We don’t like to ad­mit it,” Ka­dia said, “but it does hap­pen.”

Vi­tale, who did her res­i­dency in hema­tol­ogy in Italy, said she thought it was “a lit­tle bit strange” to learn a com­puter pro­gram would be in her class of fel­lows. But now, she said, there’s a good back and forth be­tween her and Wat­son.

She regularly tells Wat­son about jour­nal ar­ti­cles she’s read that might be help­ful, by in­putting a ci­ta­tion and high­light­ing key pas­sages, and Wat­son helps her delve into pa­tient records much faster than she could on her own.

“We are still learn­ing trust,” she said.

One af­ter­noon at MD An­der­son, Vi­tale was sit­ting next to Ka­dia study­ing a pa­tient’s file on the Wat­son pro­gram on the pro­fes­sor’s desk­top.

When the num­bers from the pa­tient’s blood­work came up, Ka­dia frowned.

Shamira Davis, 23, was a pa­tient of Ka­dia’s. They had met two years ago when she was brought into the in­ten­sive care unit, bleed­ing and near death. The stay-athome mother was di­ag­nosed with leukemia, and Ka­dia treated her with chemo­ther­apy and a bone mar­row trans­plant. She had been well since then.

Now it looked as if her can­cer had re­turned.

Vi­tale, who is shad­ow­ing Ka­dia, stud­ied Davis’s back­ground and asked Wat­son what it thought.

A long list of op­tions ap­peared on the screen.

Like med­i­cal doc­tors, Wat­son doesn’t op­er­ate in black and white. In­stead, it of­fers a set of pos­si­bil­i­ties and rates whether it has low, medium or high con­fi­dence. In Davis’s case, Wat­son sug­gested a hand­ful of stan­dard treat­ments as well as ex­per­i­men­tal clin­i­cal tri­als as be­ing of high and medium con­fi­dence. Ka­dia scanned the list, but his in­stincts told him that there was some­thing more promis­ing.

He had re­cently been talk­ing to a col­league about a new clin­i­cal trial for an ag­gres­sive chemo­ther­apy treat­ment, and he thought it was Davis’s best chance.

A few min­utes later, when Davis was told that her doc­tors were con­sult­ing with the “Jeop­ardy!” champ about her case, she was in­trigued. But would she trust a treat­ment rec­om­men­da­tion made by a com­puter or by a hu­man?

Davis didn’t hes­i­tate. “I trust Dr. Ka­dia,” she said.

Guillermo Gar­cia-Manero, a se­nior MD An­der­son leukemia spe­cial­ist who some­times dis­agrees with Wat­son’s rec­om­men­da­tions, said it isn’t so much that Wat­son is wrong but that it’s still learn­ing.

“I’m not say­ing we’re Kas­parovs, but the doc­tors here are the ex­perts, and it’s go­ing to take him a lit­tle time to catch up,” Gar­cia-Manero said. In the fu­ture, Wat­son “will be a fan­tas­tic ad­junct even for a master chess player.”

But even Kas­parov, of course, was beaten by a com­puter in the end.

Com­put­ers have an edge, said Gar­cia-Manero, be­cause they have a pre­dictable view that isn’t im­pacted by any bi­ases about cer­tain types of treat­ments or how tired they are: “Com­put­ers can’t cut corners. Hu­mans cut corners all the time.”

Gar­cia-Manero’s bosses at MD An­der­son and the Univer­sity of Texas have been so pleased with Wat­son’s abil­i­ties in leukemia that they are pre­par­ing to train it in two other spe­cial­ties: lung can­cer and di­a­betes.

“They keep telling me it will not re­place me,” Gar­cia-Manero said. “But I am pretty sure it will re­place me.”

AN­DREW SPEAR FOR THE WASHINGTON POST

Wat­son, the com­puter brain of “Jeop­ardy!” fame, is train­ing to be­come the world’s first ar­ti­fi­cial-in­tel­li­gence ex­pert in can­cer. Above is the pro­gram’s phys­i­cal em­bod­i­ment in the server room at IBMWat­son’s head­quar­ters in New York.

AN­DREW SPEAR FOR THE WASHINGTON POST

IBM VIA AS­SO­CI­ATED PRESS

SETH WENIG/AS­SO­CI­ATED PRESS

CLOCK­WISE FROMTOP: Fredrik Tun­vall, a se­nior client en­gage­ment leader at IBMWat­son in New York, con­ducts a demon­stra­tion in the Im­mer­sion Room. Ken Jen­nings, who won 74 “Jeop­ardy!” games in a row, and another of the show’s past cham­pi­ons, Brad Rut­ter, took onWat­son in 2011 and lost. IBM founder Thomas J. Wat­son Sr. con­grat­u­lates Thomas J. Wat­son Jr. on suc­ceed­ing him as the com­pany’s chief ex­ec­u­tive in 1956.

MICHAEL STRA­VATO FOR THE WASHINGTON POST

Tina Cas­cone demon­stratesWat­son atMDAn­der­son’s leukemia treat­ment cen­ter in­Hous­ton. The IBM pro­gram can syn­the­size a pa­tient’s history, of­fer a set of pos­si­ble treat­ments and rate whether it has low, medium or high con­fi­dence in its rec­om­men­da­tions.

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.