London AI upstart, which counts Novartis as a customer, can teach your computer to read
When Amazon developed a machine-learning tool to make its recruitment process more efficient — the man-made system absorbed the gender-bias of its human makers, and the project was aborted. In the field of biopharmaceuticals, the way researchers train their machine learning algorithms can skew the outcome of predictions. But before those predictions can be made, the engine must learn to read to make sense of explosive volume of knowledge out there.
That is what London-based Causaly has set out to do, and has raised $4.8 million as it works on refining its technology, co-founder Yiannis Kiachopoulos told Endpoints News in an interview.
The key issue is finding evidence in the first place because 90% to 95% of search results are noise, he said. “The problem that we’re solving is finding the evidence in the first place. The reading part and the judgment part is still up to the human.”
Millions of biomedical articles have been published so far, and thousands are added each month. The tiny 11-person startup Causaly, which is working with Novartis and a host of other biopharmaceutical companies, has developed artificial intelligence technology that processes language from this avalanche of published biomedical data — and is designed to extract causal relationships the way humans can — except faster and more efficiently.
The technology is designed to isolate relevant data and make it visually accessible and that’s what scientists want, Kiachopoulos said. “Our users are specifically asking us not to make these judgments for them…and that’s how we see ourselves as augmenting humans and not kind of replacing judgments or anything like that, that gets more problematic.”
Users are given the option of using filters, for instance, they can isolate data that emanates only from randomized clinical trials or from journals with a defined ‘impact factor’ or articles that have been published during a specific time period. The platform is not judging whether that evidence is credible, he said. “We leave the judgment to the human, but we give the human the tools to make the judgment in a better way.”
There is another layer of bias entrenched in data — and that is linguistic bias. Neither Causaly’s technology nor humans can fully eliminate that because cultural contexts play a key role in the way data is processed and articulated, he added.
Causaly raised $1 million in seed funding last year, and this injection of series A capital was led by Pentech and EBRD Venture Capital, with participation from existing investors, including Marathon Venture Capital.
It is hardly the first AI company attempting to assist decision-makers in the fields of pharmaceuticals and health care to aggregate and synthesize information. A growing list of startups including Amplion, Biorelate, Data4Cure, Evid Science, Innoplexus, InveniAI, Linguamatics, Meta, Plex Research, Quertle, Researchably and nference are all working on ways to help scientists and researchers digest and make inferences from the deluge of biological data on offer today.
Meanwhile the traditional, and largely free, search tools haven’t exactly become obsolete. “When I talk to our customers, our biggest competitor is the status quo,” noted Kiachopoulos. “And that is Google Scholar, PubMed.”