Medical researchers overwhelmed by the profusion of diagnostic and scientific data may find that artificial intelligence and personalised medicine are a match made in heaven.
By Wilson da Silva
DO YOU REMEMBER IBM’s Watson? It’s the computer cluster with artificial intelligence smarts that appeared on U.S. quiz show Jeopardy! in February 2011, and beat two former winners to claim the $1 million first prize.
How did it do this? For one, it had 200 million pages of stored content that it could access instantaneously, including the text of encyclopaedias, dictionaries, newswire articles and other reference works. It also used huge semantic knowledge and lexical databases to help it understand each question and offer what its internal algorithms predicted would likely be the right answer.
As he conceded victory to Watson in the final bout, Ken Jennings – one of the all-time best Jeopardy!contestants, with 74 straight wins – quipped, “I, for one, welcome our new computer overlords”.
But the truth is, the likes of Watson are not ready to overthrow us. Jennings’s brain (and those of lesser mortals), is still vastly superior to Watson – named after IBM founder Thomas J. Watson. Encased in roughly 1,300 cubic centimetres and powered by the equivalent of a 20-watt light bulb, the average human brain can process an estimated 100¹⁵ flops – that is, 100 quadrillion floating point operations per second.
By contrast, Watson runs on a cluster of 10 refrigerator-sized racks of Power 750 servers with 2,880 processor cores and 16 terabytes of RAM, consumes masses of electricity and does only 80¹² flops (or 80 trillion operations per second).
Nevertheless, there’s more to Watson than public relations buzz. Watson is a computer system that recognises patterns and relationships across vast quantities of data at lightning speed, then extracts information in response to a question. And that question can be asked in ‘natural language’ – the kind we use every day, but which for computers is usually replete with ambiguity.
Combined with its ability to judge a range of possible answers and decide which is most likely the correct one, all in a matter of seconds, Watson represents a new breed dubbed ‘cognitive computers’ – an amalgam of existing equipment cobbled together with some very smart software that can answer questions posed in natural language.
Using advanced algorithms and superfast circuitry, cognitive computers can spot correlations embedded in masses of data, suggest answers and learn from experience. It’s exactly the sort of thing needed in the burgeoning field of personalised cancer treatment.
Diagnosing and treating cancer is incredibly difficult because the disease is so mind-bogglingly complex, and can behave differently in each patient. There are some 200 cancers in humans. Only 5%-10% are known to be hereditary and therefore likely to show up in genetic tests.
So to detect, recognise and devise treatment for any one individual requires a series of examinations, biopsies, screening tests and medical imaging – as well as professional experience, a doctor’s instinct, and more than a little trial and error.
When you consider that, on top of this, there are always new case studies being published, new successes and failures reported, new treatment regimes tested – it’s impossible for any single oncologist to keep up with it all.
That’s where cognitive computers come in. At the renowned Memorial Sloan-Kettering Cancer Centre in New York, researchers have been trialling Watson as a ‘search engine on steroids’: getting it to analyse millions of pages of unstructured text in patient records and scientific literature to generate a long list of hypotheses in response to each oncologist’s question.
Watson spits out answers according to priority and assigns a confidence level for each. It can also supply supporting literature and suggest further lab tests to help refine diagnosis or treatment.
Able to read millions of pages per second of patient histories, treatment notes, lab results and the latest clinical research in medical journals, it can be more thorough than any human when doing the background research. But, in the end, it’s the physician who interprets results and determines the treatment.
Early indications are positive. And just as well – medical information is doubling every five years, so keeping one’s head above the avalanche of published research is now beyond most clinical practitioners.
Rather than becoming our sinister overlords, cognitive computers may end up being our trusted friends.