With the first AI-aided drug entering clinical trials soon, the tech world should take stock of what’s going on. As an economic benefit, these AI-developed drugs can speed up the process of developing and getting these drugs to market. On average, it could take about one-third of the time it usually takes to get a drug through FDA trials and available for consumers. However, even with the benefits, it can provide to society, there are a few drawbacks that might affect how we see AI-derived drugs in the future. Sure, we can do it, but should we?
Concerns About Bias and Generalizability
Artificial intelligence utilizes a data set to derive results. In other applications of AI, the datasets available are limited to a particular cross-section of society. Drug firms use collected information from hospitals and, in some cases, genetic libraries to help them develop drug solutions. Machine learning algorithms are likely to create new drugs based on what they have in their datasets. These new medicines may apply to a particular area of the world or genetic makeup based on the available information. As more data is collected, the creation of drugs will become more usable across more regions. As of now, we’re not sure if AI-developed medicine will be able to be used by a wide enough cross-section of the populace to make it worthwhile.
Mitigating Risk in Biopharma
Medicine can be a dangerous field, and each procedure that a patient undertakes could have “death” as a side-effect. It’s not uncommon to hear a list of side effects that are possible when taking a particular medicine, some of which are unnerving to the patient. In the case of drug companies, these side effects are an acceptable risk. However, by employing AI to develop their drugs, biopharma companies reduce their risk.
The goal of biopharma companies is to create new drugs faster, safer, and cheaper than their competitors. In a recent conference at Harvard Medical School, the topic of AI in biopharma was given some consideration. The most significant concern raised by the attendees was how AI operated. In general, data scientists don’t actually witness AI making its decisions. This “black box” methodology makes it difficult to discern AI bias. While we can see drugs developed faster and potentially cheaper, how safe they are is debatable.
A Long Road to Consistent AI Drug Development
AI, like humans, learns from its mistakes. However, for it to learn, it has to make those mistakes first of all. There isn’t likely to be any single massive event that the AI-drug industry can point to as something that will change the face of the industry. Instead, it will be a series of small victories and a series of setbacks that the biopharma industry should learn from. Curating and cultivating the data that makes up the datasets for these AI drug development engines is one of the ways to manage the inherent problems with AI bias. Unless they deal with issues like these soon, AI might add to the risk that drug companies have to shoulder instead of mitigating it.