By Carl Foster, CBO, Standigm.
getty
AI has been a hot topic in the pharmaceutical industry, and the latest ChatGPT hype is fueling even more conversations. Right now, there’s a bifurcation of attitudes, with some scientists excited about the prospects of AI and others generally wary of it—which isn’t surprising. Researchers and chemists are highly educated experts who have been the heroes of making cures in the last several decades with the help of computers.
Responsible Use Of AI Tools
One of the reasons pharmaceutical companies consider using AI is because they’re looking for ways to get to market faster with new discoveries. Still, some can’t trust “black boxes” and there’s a good reason for that.
Black boxes are associated with deep learning, a specific type of AI technique. Some of these systems can produce game-changing outputs, but they’re unable to explain how they arrived at a conclusion in a manner that humans can understand. However, it is possible to develop models that are self-explanatory. This is commonly called “explainable AI.”
Over the last several months, the hype about ChatGPT has even skeptics feeling like perhaps they should understand AI’s potential impact. In fact, some are now even trying adventurous things, such as diagnosing disease in patients with rare disorders, predicting viral mutations for new vaccine creation and replacing animal trials when testing new drugs.
Computer-aided drug discovery dates back to the 1960s. In fact, the pharmaceutical industry was one of the earliest adopters of computer technology, and it has significantly improved productivity.
The biggest hurdle can be found in the differences between wet and dry labs because the speed of data generation differs greatly. Some pharmaceutical companies have already started clinical studies of their AI-designed molecules, which has caused others to feel the urgency for AI adoption.
Today, nearly all large pharmaceutical companies have made significant investments in AI technology, and they have the expertise to apply the technology. Small biotech companies employing only a few chemists do not. They have IT professionals who keep computers running, but they don’t have the resources to apply new applications that cross over into chemistry design.
Also, some individuals fear that the use of AI will harm their careers. I was speaking with someone last week who said his company’s chemists feared for their jobs.
The reality is that, like computers, AI is becoming increasingly pervasive across industries and its impact will be transformative. Roles will change with researchers and scientists achieving more in less time. However, many decision-makers are years out of graduate school and were not taught how to use AI, which leaves them with two choices: upskill or keep doing things the same old way.
The decision-making process changes required for AI technology adoption might be a more profound reason to be skeptical. For example, the interaction of a drug molecule with a biological system has required the pharmaceutical industry to develop sophisticated workflows where many human experts make pivotal decisions based on expensive experimental data. Once AI proves its ability to solve those complex problems, the whole industry can be transformed into a totally different business. And that, for some, is downright scary. For others, it’s nothing short of an exciting adventure.
Using AI is, indeed, a step change because it opens the discovery door to new molecules and compounds that are hidden in a sea of data. While AI can assist greatly in this regard, humans are still necessary to verify the results.
It’s very important to understand how to interpret the results of AI models. Since real-world use cases are complex, researchers and scientists need to get their intuitions from trials. Whether using old target discovery or chemical design methods, or AI-derived results, the key question always asked is whether targets or chemicals have been validated.
Scientists tend to trust algorithms when they can understand them. The best way to trust an algorithm is to see the results validated in the models of choice and, of course, achieve good results consistently. Model explainability also tends to reduce skepticism.
Important Do’s And Don’ts
Technology and business logic must work together. All the hurdles lie between the different disciplines, mostly related to terminology. The best approach is to address minor issues first to build confidence and then tackle more significant problems later. So, do:
• Develop interdisciplinary teams.
• Build a common dictionary of terms.
• Solve minor issues to gain credibility.
Conversely, a separated team with its own key performance indicators (KPIs) will be isolated from other functional groups and it will be difficult for them to break down the walls between them, so don’t:
• Build a dedicated AI team with its own KPIs.
• Implement a small proof of concept project to prove AI technology applicability.
• Forget that in the short term the results will need to be tested and validated in traditional ways.
Learning to use AI is an important skill set in the modern digital world. Though there’s still a healthy level of skepticism, there are also new AI drug discoveries in the works that might not have been discovered for years, decades—or perhaps at all.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
