Naomi Visanji and Jonathan Rezek are sitting in the lobby restaurant of Ocho, a boutique hotel on Spadina Ave., sipping tea and talking honestly about their 6-year-old.
He’s a genius, by all accounts, and has progressed from upstaging adults with his quick wit, to speed-reading and solving complex problems.
But he’s young and untested, and academics and scientists have yet to understand his full potential.
His name is Watson.
Visanji has left him under the watchful eye of her colleagues at Toronto Western Hospital, where she logged him onto her computer before leaving to meet Rezek.
“I’ve actually left Watson now running five separate searches of the entire medical literature on over 15,000 drugs,” says Visanji.
She hopes he hasn’t crashed.
Of course this is the Watson that was built by IBM to understand answers on Jeopardy and come up with the right questions.
Since his appearance on the game show in 2011, IBM has expanded Watson’s talents, building on the algorithms that allow him to read and derive meaning from natural language. The computer system can pore through documents millions of times faster than any human.
And among other functions, IBM adapted Watson for use in medicine.
Toronto Western, part of the University Health Network, is the first hospital in Canada to use Watson for research in Parkinson’s, a neurological disorder.
The centre has a track record of running clinical trials for off-label drug use, which means taking a drug approved for treatment of one condition and repurposing it for another. Researchers here believe Watson can help them speed up this process to find a cure for Parkinson’s.
Visanji, 39, is a scientist at the hospital’s Morton and Gloria Shulman Movement Disorders Centre, the country’s biggest Parkinson’s clinic.
She and her colleagues have been working with Rezek, ever since the 56-year-old IBM executive — who was diagnosed with Parkinson’s at age 50 — pitched the idea last year during an examination with his doctor Connie Marras, a neurologist and epidemiologist at the centre.
“I’m there in my mindset of [his] clinical visit,” recalls Marras, sitting in her 9th floor office at Toronto Western, “and he then derails me and says ‘I’ve got this thing called Watson that I want to talk to you about.
“And the first reaction is, man, I’ve got my own agenda here. Don’t derail me,” she says, laughing.
“You don’t expect your patients to come up with their own novel ways of tackling the problem.”
Jeopardy superstar Ken Jennings tells a funny story in the video of his TED talk in 2013, two years after he lost to Watson on the popular quiz show.
The 74-game record holder had been contacted in 2009 by Jeopardy and asked if he would compete against a supercomputer that IBM was building. A programmer himself, Jennings had taken courses in artificial intelligence and was fairly confident he could win.
But as the computer was being developed and Jennings began receiving updates, doubt started to set in.
At one point, he saw a chart in a journal article that showed Watson’s question-answer progress as a fever line with a cloud of dots above it, each plotted to correspond to the performance of a former Jeopardy champion.
The line was starting to get awfully close to the cloud.
“I sort of saw this line coming for me, and I realized this is it, this is what it looks like when the future comes for you,” he told the TED talk audience. “It’s not the Terminator’s gun site,” he added, referring to the famous Arnold Schwarzenegger movie where machines take over the planet.
“A lot of us will have our skills made obsolete by automation at some point,” says Jennings in an interview with the Star, “but you so rarely get to see it in graph form.”
Like Deep Blue, the supercomputer built by IBM to play chess, Watson was created as one of the company’s Grand Challenges, a public relations exercise to build a computer that could read questions and answer them.
IBM wanted to “leverage knowledge the way that it’s naturally communicated by humans, which is text,” says Eric Brown, who worked on Watson during the Jeopardy days and is now head of innovation for the healthcare and life sciences division of IBM Watson.
The system was named after Thomas J. Watson, who rose from salesman to IBM chairman and then passed the reins over to his son Thomas J. Watson Jr. in 1956.
Research began in 2007, and for the first year, progress was slow, as scientists and engineers figured out how to build Watson, including what software they’d need to process information and how it would flow through him. They harvested systems and technologies already built by staff.
About 12 employees who had been working on natural language processing for a decade contributed separate components — algorithms they created that needed to be migrated over for use in the Watson system.
One algorithm might be a parser, responsible for breaking down a sentence to identify the nouns, verbs or prepositions, the subject and object. Another algorithm might look for proper names of people, places and organizations. Maybe another looked for relationships between them.
During the course of development, says Brown, they ran about 8,000 experiments on Watson to test how he would perform, running through previously played Jeopardy questions to find the chinks in his armour.
It was “wash, rinse, repeat,” says Brown.
When something went wrong, an algorithm would be removed and made more sophisticated so that Watson could understand, for example, that when he read “George Washington, elected as first president of America,” it meant the same as “George Washington was the first president of the United States.” In some cases, researchers had to provide him with more factual information so he could answer the question.
IBM harnessed machine learning so that when Watson’s “expert algorithms” provided him with the best possible answers, he was trained to choose the most probable. And they built him a mechanical finger so that he could hit the buzzer.
When he was fast enough to perform in real time, IBM ran sparring drills with past Jeopardychampions, about 155 matches in all.
Jennings says he knew the computer “had game” as soon as he watched the tapes Jeopardy sent over of Watson in practice rounds.
But it wasn’t until he was sitting in a press conference with some IBM brass that he realized the system’s true potential.
“They were talking about how you could use the same algorithm to do business analytics and tech support and whatnot,” he says “Then the penny dropped. This wasn’t principally about the PR value of the Jeopardy match, though that was the kind of publicity any company would kill for. They wanted an algorithm that could answer questions and this was a way to get people to care about that arena.”
The company’s executives and their clients were in the audience on the day of the matches in January 2011. Brown was there, too, seated with the other members of the technical team in the auditorium of the T.J. Watson Research Centre in Yorktown Heights, about 55 kilometres north of Manhattan, where the shows were being taped. He says there was a lot of nervous excitement in the crowd.
At that point, Watson was off-screen, literally a room full of 10 fridge-sized racks of 90 IBM servers, which could perform 80 trillion operations per second, cooled by fans that whirred noisily. The system had no Internet connection.
On screen, an avatar of Watson glowed behind the middle podium.
The avatar, based on the company’s Smarter Planet Logo, was a globe surrounded by rings of particles that sped up when Watson was puzzling over an answer, which he received in text form.
The particles swirled to the top of the globe when he was sure he had the right response. Or dropped to the bottom when he was feeling less confident.
No matter how emotional he appeared, he voiced his responses in the same flat tone.
(In recent years you may have heard that same voice in conversation with Bob Dylan or filmmaker Ridley Scott in commercials for IBM, but Watson doesn’t typically speak. Business Insider reported last year that IBM resurrected the voice in its advertising to fight “misconceptions about artificial intelligence turning evil and taking over the world.”)
On Watson’s right was Jennings. To his left was Brad Rutter, who held the record for highest Jeopardy winnings, a total of $3,385,702 (U.S.).
By the end of the two matches, the computer racked up $77,147 (U.S.) to Rutter’s and Jennings’s $21,600 and $24,000 respectively. Jennings wrote below his Final Jeopardy answer “I for one welcome our new computer overlords,” a humorous take on meme made popular on The Simpsons.
Jennings says Watson was able to win because of his ability to buzz in quickly with his mechanical thumb as soon as a Jeopardy staffer triggered the “buzzer enable” light on stage.
Most human contestants trying to get to the buzzer first, succeed only half the time, he says.
“Nearly 100 per cent of the time Watson wanted to buzz in, it got in ahead of me and Brad,” says Jennings.
Regardless, he says, the “lesson of the Watson match is just how far and fast AI can advance when there’s a moonshot-like deadline and someone to sign the cheques.
“Question-answering algorithms were a joke for decades,” he says, “but you throw millions of dollars and your best engineers at the problem and it turns out barriers can fall quickly.”
After Jeopardy, the medical community recognized the computer system’s capabilities and came calling right away.
Watson had the potential to manage the exploding increase in digital information, including electronic patient records and the thousands of scientific studies published every day. In 2014, the company announced it would invest $1 billion to establish the IBM Watson Group, which would offer cloud-based cognitive computing.
A year later, 14 hospitals in North America were using Watson toprocess the DNA sequencing of cancer patients to find the mutations responsible for the disease so they could identify treatments, if available. It’s a process that takes the computer system only minutes, but would take a clinician weeks to do manually, according to the Modern Healthcare website.
In 2016, Watson was put to the test by the University of North Carolina. They had him review past case histories of 1,000 late-stage cancer patients — for whom there is often no standard of care. His treatment recommendations lined up with what the doctors had advised. And in a third of those cases, Watson came up with a treatment option no one had thought of.
The university is now using Watson regularly to assist its Molecular Tumor Board, the team of experts who meet weekly to decide if there are treatment options for patients.
His predictive abilities have even extended to guide dogs.
In an IBM-sponsored podcast called Wild Ducks, geneticist Jane Russenberger, of the non-profit Guiding Eyes for the Blind, talks about how Watson was used to review 25 years of historic data to see if he could accurately predict which dogs had graduated from their guide-dog program, and which trainers had been successful.
Watson considered behavioural data, family history and health records, as well as other information, and his predictions were 100 per cent accurate.
“That blew my mind, said Russenberger in the podcast, noting that it takes $48,000 to train one dog and that a third of the dogs don’t graduate. “I never expected that would happen.”
Watson has gone on to aid in drug discovery for pharmaceutical companies and solve tax problems for H&R Block customers. He’s discovered five new genes linked to Amyotrophic Lateral Sclerosis, and in his spare time, he has written a cookbook and created a recipe for a new rum.
But it’s his appearance on Jeopardy that really gave him his star turn.
“I’ll run into people . . . who, of course, saw Watson on Jeopardy and I don’t have to explain what the problem is that we were trying to solve,” says IBM’s Eric Brown. “A system that can answer Jeopardy questions at the level of a human champion … people get that.”
IBM executive Jonathan Rezek was teaching tech startups how to use Watson when he decided to raise the idea of using the technology for Parkinson’s research with his own doctor just over a year ago.
“It took me a bit of time to build up my nerve,” says Rezek. “I didn’t want to do anything to jeopardize the relationship I had with [Dr. Marras]. I didn’t want to seem untoward or forward.”
But he thought it was worth the risk.
By the time you start showing the motor impairment that comes with Parkinson’s, 80 per cent of your dopamine-producing neurons that control movement are gone, says Rezek. “Parkinson’s is a really slow moving disease. It’s hard to do research on it. So anything you can do to make research go faster is a positive.”
Soon after Rezek’s appointment with Dr. Marras, IBM organized a presentation for Toronto Western researchers who quickly realized Watson’s potential.
But it took months to find funding and finalize contracts with IBM so they could use Watson — which has grown from one computer system to multiple data centres around the world — for use in Parkinson’s research.
The project officially launched in September with $300,000 in funding from the Ontario Brain Institute (OBI) to pay for licensing fees for a year.
In December, the system identified 52 approved drugs that could potentially be used to treat Parkinson’s. An off-label clinical trial could happen in months.
“All of the slow parts were humans,” says Visanji. “The Watson part was very fast.”
However, the system doesn’t know what to look for on its own. Doctors and scientists have to “train” Watson.
In this instance, he was tasked with reading more than 20 million summaries of scientific studies that were available free online. (Accessing the entire study typically requires a payment to the journal.)
Researchers trained him to look for any mention of alpha-synuclein, a common brain protein that clumps together in Parkinson’s patients, an action that scientists think causes the disease.
Watson then looked in the same text for a mention of an approved drug in Ontario.
“It really is the most simplistic strategy,” says Tom Mikkelsen, president and scientific director of the OBI. “What it is looking for is the statistical nearness of the words.”
Watson ranked the list of 52 drugs from best to worst.
Visanji says 21 of the drugs are worthy of further study, and of those, 16 had never been linked to Parkinson’s before.
“I was asked the question one time, how would you approach this same problem that you’ve posed to Watson without Watson,” says Visanji. “And the answer is we wouldn’t have. We couldn’t have physically done it.”
It’s a shortcut that could shave years off approval of a new drug to treat the disease, a process that typically takes at least a decade and costs millions of dollars. Typically, one in 10,000 drugs studied in the lab will make it to a clinical trial, a drop-off that scientists call the Valley of Death.
“If there’s something out there that’s already gone through its toxicology testing; that’s gone into humans; that’s completely safe … then bingo,” says Visanji. “Why would you look for something else?”
In Parkinson’s patients, the dopamine producing neurons tied to movement gradually die until there are none left. Currently, patients are treated with levodopa, a drug in use since the 1960s that metabolizes into dopamine in the brain. However, levodopa becomes less effective over time because the neurodegenerative process continues until eventually there aren’t enough healthy neurons to take up the dopamine from the drug and function in response to it.
Patients lose mobility but they can also become apathetic and depressed, suffer pain and even lose their sense of smell. Dyskinesia, involuntary movement exhibited by actor Michael J. Fox, is a side effect of the medication.
The disease typically begins about 20 years before there is a clinical diagnosis, which is made only when symptoms begin occurring. There are no clinically approved biological markers for Parkinson’s.
Rezek was diagnosed after he fell and broke his right arm while biking around Taylor Creek Park in Toronto. The break healed but his arm began to shake violently during rehab. After an MRI to rule out other diseases, he was told by a neurologist that he had mild idiopathic Parkinson’s, the most common type of the disease.
He says his voice has faded and his right foot is not as limber as it was. His right hand shakes so he has learned to write with his left.
“The weirdest thing about this project for me is every time we do a presentation I get to listen to the list of symptoms — you can get this, you can get that, you can have this happen to you,” he says. “I’d prefer to live in denial about that for a while, but you can’t.”
He has continued to work because doctors have told him the best two things he can do are to keep his brain active and exercise, which he does everyday.
Visanji too, has a personal connection to the disease.
Her mother, 74, who lives in the U.K., was diagnosed with Parkinson’s four or five years ago, and Visanji has watched it rob her of the ability to do many of the things she loved, including garden.
“She is sitting there in her retirement surrounded by things that she can’t enjoy anymore,” says Visanji. “It’s really, really tragic.”
That, together with the unusual experience of working on the project with Rezek, someone who has the disease, is a motivating factor.
“I love the personal story that all of this came from Jonathan,” says Visanji. “Thank God for patients like him, that kind of come up with these ideas and speak out. We’re really grateful to him.”
The next step for the team at Toronto Western, if they can find another $300,000 in funding, is to take Watson’s list of drugs and narrow the search even further.
Scientists can look at patient data in Ontario to see if the incidence of Parkinson’s is lower in people on any of the 52 drugs, which could mean the drug is beneficial in some way.
They’ll also test the best 20 or so candidates in the lab to see what effects they have on alpha-synuclein aggregation.
The best candidate will be chosen for the off-label clinical trial.
“You could get a clinical trial up and running in six months after we’ve picked the best candidate,” says Visanji. “I see this as something that could happen. It’s a reality. I really hope it does come from this.
“And if it does, the short cut is insane.”
If Watson is successful, Visanji predicts there will be wide acceptance of artificial intelligence for use in scientific research.
“I think it’s fair to say that, at the moment, artificial intelligence, or cognitive computing, is viewed by a lot of the scientific community with skepticism,” says Visanji.
“But I think that will change over time as people start to publish work and show that sort of critical proof of the things that Watson predicts.”
Montreal has already become a hub for AI institutes, and, in March, it was announced thatToronto will become home to an artificial intelligence institute affiliated with U of T and backed by more than $150 million in public and corporate funding.
As Visanji talks in the hotel restaurant, she opens her laptop and logs on to the IBM interface to see if Watson is still running the massive searches she set up before leaving her office an hour ago.
He is. In fact he has completed one and has displayed the results on-screen.
It seems Watson is still hard at work, even though his head is in the clouds.