What AI Has in Common With Pharmaceuticals
Over the last year, I’ve noticed two very different conversations taking place around artificial intelligence.
The first is the conversation most people are familiar with. New models are released, benchmark scores improve, businesses experiment with adoption, and headlines focus on capabilities. Depending on who you ask, AI is either going to dramatically increase productivity, eliminate entire categories of jobs, transform every industry, or some combination of all three. Most of the discussion focuses on what these systems can do and how quickly they’re improving.
The second conversation is quieter, but I suspect it may ultimately prove more important. That conversation is about governance. Should governments regulate AI? If so, when? What risks justify intervention? What obligations should organizations developing frontier models have to the public? How much oversight is appropriate, and who should be responsible for providing it?
What I find interesting is that these questions are often discussed as though society has never faced a similar situation before. The technology is certainly new, but the underlying challenge feels familiar.
As somebody who has spent most of his career working in enterprise technology, I’ve seen countless examples of organizations struggling to determine when governance becomes necessary. Nobody wants unnecessary bureaucracy. Nobody wants processes that exist solely for the sake of having processes. At the same time, organizations rarely introduce governance frameworks when everything is going well. Governance usually appears when the consequences of getting something wrong become significant enough that leaving decisions entirely in the hands of the people closest to the problem is no longer viewed as sufficient.
When I think about AI, I keep coming back to a pattern that appears repeatedly throughout history. Technologies that create substantial economic value, have meaningful national security implications, and possess the potential for widespread harm rarely remain entirely self-governed indefinitely. The details vary from one industry to another, but the general trajectory tends to be remarkably consistent.
AI appears to be moving into that category.
The economic implications are already becoming difficult to ignore. Whether one believes AI will create more jobs than it eliminates, or whether one believes it will primarily change the nature of existing jobs, there is little evidence to suggest it will leave the economy untouched. Software development, customer service, marketing, sales, legal research, content creation, and countless other disciplines are already changing. Businesses are restructuring teams, revisiting operating models, and reevaluating assumptions that have remained relatively stable for years. Technologies that reshape economic activity tend to attract attention from policymakers, regulators, and governments because the effects are no longer limited to the organizations building the technology.
The national security dimension is equally important. Governments around the world are investing heavily in AI research and deployment. Some of those investments are focused on economic competitiveness. Others are focused on cybersecurity, intelligence gathering, logistics, military planning, and autonomous systems. Historically, once governments begin viewing a technology as strategically important to national interests, the regulatory conversation changes. The technology is no longer evaluated solely through a commercial lens. Questions about access, control, security, and long-term strategic advantage begin entering the discussion as well.
The final category is probably the most controversial because discussions about harm often become exaggerated very quickly. Mention AI risk and many people immediately jump to science fiction scenarios. That isn’t really what I’m talking about. Most transformative technologies have the potential to create enormous value while simultaneously introducing risks that require active management. Automobiles transformed transportation. Pharmaceuticals transformed medicine. Aviation transformed travel. None of those industries became regulated because society concluded the technologies were bad. They became regulated because society concluded the consequences of failure extended beyond the organizations producing them.
The pharmaceutical industry is the example I find myself returning to most often.
Imagine a company developing a breakthrough treatment for a serious illness. The company’s researchers are confident in their findings. Investors are enthusiastic. Patients want access as quickly as possible. The potential benefits are enormous. Under those circumstances, very few people would argue that the appropriate response is to bypass independent review entirely and allow the company to determine on its own whether the treatment is safe for widespread use.
That isn’t because society opposes innovation. Quite the opposite. It’s because the potential consequences of getting the decision wrong are significant enough that independent evaluation becomes part of the process.
I am not suggesting AI should be regulated exactly like pharmaceuticals. Every technology is different. The risks are different, the timelines are different, and the mechanisms of oversight would almost certainly be different as well. What I am suggesting is that the forces that created oversight in other transformative industries appear to be emerging around AI for many of the same reasons.
The argument I hear most frequently against regulation is that governments move too slowly and understand technology poorly. There is certainly truth in that concern. Poorly designed regulation can absolutely create more problems than it solves. History provides plenty of examples of that as well.
What I struggle with is the assumption that the alternative is no regulation at all.
As far as I can tell, societies generally do not allow technologies with major economic impact, national security implications, and significant public risk considerations to remain entirely self-governed forever. Eventually, whether through legislation, oversight bodies, standards organizations, industry frameworks, or some combination of all four, governance emerges.
The real question may not be whether AI will eventually be regulated.
The more important question is whether we can build thoughtful, technically informed frameworks before a major incident creates pressure to build them in a hurry.
If history is any guide, the quality of the answer to that question will matter far more than the existence of regulation itself.