AI in Medicine

Coming up with new medicines is a time-consuming and costly affair. It was inevitable somebody would throw AI at the problem, writes Rahul Matthan. A pharma company named Collaborations Pharmaceuticals tried using an AI named MegaSyn:

“The company was convinced that if machine-learning algorithms trained on chemistry and molecular engineering were used, it would be able identify new, never-before-seen compounds that had a high probability of curing diseases.”

 

The AI threw up a huge number of possible chemicals for various diseases. Sensibly, the company then decided to add a filter: reject any chemicals that could have serious side-effects.

 

You must have guessed what happened next:

“The trouble is that once a feature like this has been designed, it is very easy to flip the switch—to use the algorithm to design toxic chemicals instead of just filtering them out.”

In fact, some researchers at the company secretly tried just that – and yes, it did come up with at least one chemical that was already known to be so deadly that it had been banned by the UN.

 

We like to think of tech as good or bad. Whereas, tech is amoral, says Matthan. After all, MegaSyn started off as a “good tech”. But it was then flipped to become “bad tech”. The knee-jerk response to such things isn’t necessarily the right one:

“Our instinctive reaction to the duality inherent in a powerful technology is to shut it down, believing that it would be far better for us to forgo the many benefits that it offers than risk the harms that could befall us. If this becomes our knee-jerk response to every new risk that technology poses, we will end up mindlessly stifling all innovation simply because of the harms that it might end up causing.”

 

The AI genie is well and truly out of the bottle.