The danger of ‘going too far’ with AI

Andrew Garrett, EVP of scientific operations for the clinical research organization (CRO), ICON, spoke on a panel at the Financial Times Pharmaceutical & Biotechnology Conference that took place in London last week, about the integration of artificial intelligence (AI)-enabled services in R&D.

Outsourcing-Pharma had the chance to discuss with Garrett on the industry’s increasing adoption of AI, as well as outlining what the potential risks and challenges are of the new technology’s potential for disruption.

“On the one hand, we have to be ‘evangelists’ to the opportunities opened, which we have never had before, and, on the other hand, we should be ‘healthy cynics’, watching out for the dangers of going too far with AI,”​ he said.

Asked about how it is possible to ‘go too far’ with AI, Garrett said that there is a danger of being ‘seduced’ by the technology and think that it can solve any problem, if the user believes that complexity is beneficial.

“Building more complex services doesn’t mean that these will be more robust or any more capable of predicting outcomes; sometimes it’s better to keep it simple,”​ the executive explained, citing as well the importance of the quality of the data.

More specifically, Garrett noted that data generating processes need to be very carefully executed. He provided the example of a randomized clinical trial, where the data is being correctly generated and collected to produce evidence, “whereas, if you just collect data from social media or health records, you cannot be sure on whether these are representative or not.”​

According to Garrett, industry experts should avoid biased data, in order to make sure that they will not end up with a biased analysis, which can be achieved by investing time and effort into understanding their data.

The executive expressed the opinion that at the moment, the industry is “putting too much focus on the output,” ​and added: “People are thinking about the algorithms and the magic of it, and forget about how the data is generated in the first place.”​

Asked about how he sees the industry being affected by adoption of AI services in the foreseeable future, Garrett, said that there will be a takeover of virtual and hybrid clinical trials and ‘telemedicine’.

However, over the next few years, the biggest change that the industry will see, due to the prevalence of AI, according to the executive, will be “the development of new, more refined and relevant endpoints, based on digital technology, particularly in the area of central nervous system disorders.”​