The current state of the art in artificial intelligence (AI) lags far behind the standard tropes of science fiction. Most science fictional representation of AI is based on ‘general’ AI, producing a convincing emulation of human personality, insight and intelligence with a propensity to then extrapolate that to a mad tech-based demagoguery. Fortunately reality is still a little different with most AI being ‘narrow’, restricted to making our lives easier via the application of machine learning and Natural Language Processing (NLP) to interpret and generate value from big data sets.
IBM’s Watson and Google’s DeepMind are the most well-known embodiments of this vanguard defeating human chess grand masters and proponents of ‘Go’ and ‘Jeopardy’. Watson has been applied to a number of healthcare related problems including diagnosis of human cancer patients but to date has not been greatly successful with MD Anderson terminating their collaboration with IBM when Watson failed to achieve a 90 per cent or better diagnosis rate in cases of suspected Leukaemia. The whole sorry saga has cost MD Anderson US $62 million to date.
Experts have suggested that IBM’s Watson has fallen short on two counts: the quality of the data going into it (rubbish in, rubbish out); and its ability to work with unstructured data via NLP. Unstructured data is just what the name suggests, text based speech, such as this article, commonly found in physicians notes, research papers, patents and commentaries.
Other companies are notably achieving early commercial success by focussing on improving this state of affairs in one or more ways. Linguamatics in Cambridge, UK have enjoyed strong success in both pharma discovery and healthcare markets by delivering a text/data mining NLP and machine learning derived platform. Babylon Health, a UK business, is focussed on ‘teaching’ its AI how to accurately diagnose patients for GPs by recording thousands of man hours of GP diagnoses across thousands of GP/patient interactions. Thus taking the approach of quality data inputs and sheer scale enabling their AI to be successful in reducing GP workload by screening the more routine cases out of their caseload.
FlatIron, based in the eponymous building in New York and funded by Roche Pharmaceuticals among others, is focussed on gaining insights in oncology patients from using AI to trawl through patient electronic health records (EHRs) to gain insights from the sort of unstructured data that resides in doctor’s notes as well as the more structured data from lab reports. Trinetix, based out of Boston, is applying a similar approach but going broader trying to annex data from as many EHRs as it can gain agreement from separate health groups to access. It is then allowing those groups who collaborate with them free access to the resultant data whilst selling access to the wider data to the pharmaceutical industry as a patient recruitment tool for clinical trials.
Benevolent AI, based in London, is focussed on drug discovery, using its own version of AI to access patent data and research publications from public and private sources to find new disease associations for known drug targets in order to re-purpose established drugs into new indications, or discovering novel drugs for established drug targets. This area is viewed as potentially very lucrative with investors keen to commit significant funds to support it. Both Benevolent and FlatIron are ‘unicorns’ with valuations exceeding US $1 billion despite few if any revenues.
All are using AI to churn through terabytes of unstructured data which would previously have taken many human lifetimes to achieve in order to provide data associations and insights that would have been the sole preserve of human minds. Only time will tell how successful they will be, however, there are no shortage of investors betting on them being so.