IT IS increasingly obvious that we are on the cusp of a revolution in artificial intelligence that will be no less profound than the arrival of the printing press or the internet, as we explore in this special issue. Nobody can say for sure exactly what this future will be, but optimists 鈥� including many of those working in the companies behind the technologies 鈥� foresee one in which AI will allow us to live our best lives (see 鈥淗ow this moment for AI will change society forever鈥�). Not everyone shares that rosy outlook, however, and whether it is existential risk of an AI coup d鈥櫭﹖at (see 鈥淲hy do some AI researchers dismiss the potential risks to humanity?鈥�), continued sexist and racist bias or swingeing job losses caused by automation, the need for legislation has become the issue of the day.
AI presents a larger than usual challenge on this front. The pharmaceutical industry can offer inspiration, but AI is both more complex and more opaque. It is arguably the first of humanity鈥檚 creations that nobody fully understands. Is conclusive proof of safety even possible?
Advertisement
It is true that pharmaceutical regulators don鈥檛 need to know how a drug works to ensure it is safe to prescribe. But clinical trials are easier than checking that an AI capable of everything from diagnosing cancer to autonomously driving a bus is safe in every possible scenario. And while it takes years to trial a new drug, new AI models are appearing almost weekly.
Tech firms must accept responsibility. The code and data inside today鈥檚 AI models 鈥� emerging almost exclusively from companies rather than academic institutions 鈥� often aren鈥檛 being released. This lack of transparency isn鈥檛 good enough given the stakes. We need a robust framework to vet AI models and penalise those that fall short of strict ethical rules, but for that to work, tech firms should lead from the front by sharing the information needed to assess the risks. That may not suit their shareholders, but when the risks are so large, money shouldn鈥檛 come first.
These systems may be the most complex we have had to regulate, yet as we hear in our feature 鈥淗ow smart is ChatGPT really 鈥� and how do we judge intelligence in AIs?鈥�, none of it is magic. Those creating AIs need to stop acting like it is.
