How do we define success?
To answer this question, we must understand how success should be judged. AI isn’t a panacea in drug discovery. The first two end-to-end AI-designed drugs failed in trials to show statistically significant efficacy. Billions have been invested in AI-based drug discovery companies. Many potential discoveries have failed even to reach clinical trials. To date, the financial losses are enormous.
However, failures of individual AI-developed drug projects do not mean that AI cannot have a substantial positive impact on drug discovery. Historically, 90% of new drugs fail at the clinical stage. If AI can reduce this failure rate to 80% or reduce the time to clinical failure by 25%, it would be considered a significant success.
Reducing the clinical failure rate from 90% to 80% would double the number of approved new drugs. This, in turn, would drastically improve the health conditions of the world’s human and animal populations. Health improvement directly translates into better economic conditions for families and governments.
The cost of developing a new drug is often in the $1-$3 billion range. AI can also increase the amount of data new drugs can be rapidly tested against. AI does not have to be responsible for designing drugs from the ground up. It can reduce the time to develop traditionally developed drugs, thereby saving enormous costs, saving lives, and addressing diseases with novel and better solutions. Perhaps it can also figure out other use cases for newly developed drugs.
Beware of the risks!
While the potential positives are significant, there are also plenty of negatives. The ability to design new but dangerous drugs and pathogens cannot be underestimated. AI uses prompts to tell it what to do. If terrorists gain access to an “AI DrugGPT,” they could almost instantly develop substances that are both deadly and without a known antidote. Terrorists, governments at war, or blackmailers could wipe out entire countries. More powerful COVID-19-type pandemics could be around the corner. We may need to bring James Bond back again to protect the world. Remember that AI does not have a conscience.
