The Economics of Intelligence
If you’ve interacted with a service bot lately (and I dare you to make an appointment without involving one), then you’ve surely noticed how infuriatingly inefficient AI is at solving any problem worth calling about. Invariably—but only after you’ve wasted five minutes trying to dumb down your query into one of the five problems the bot understands—a human agent must be summoned. Once again, you must state your issue. A call that might have resolved it in two minutes has now robbed you of ten.
And that inefficiency isn’t even the most egregious.
To answer just one query, Gemini, Google’s AI, admits to using between 100 and 1000 watt-hours. In contrast, the human mind processes thousands of queries, performs innumerable calculations, and oversees dozens of complex operations on a piddling 20 watts per day. Our output-for-input ratio is stupendous. The quality, let alone quantity, of our output is hard to beat. But instead of paying humans to help other humans with problems best solved by humans, companies are shelling out billions to replace them with energy-greedy, inept facsimiles. How, by any measure, does this make sense?
AI can and already does benefit humankind in ways that justify its enormous energy requirements. It’s extremely good, for example, at reading medical scans and perceiving what doctors overlook. It can distill billions of datapoints into accurate forecasts, forecasts that can reallocate precious resources and save lives. It can enable robots to perform physical tasks that for humans would be tedious, taxing, or dangerous.
We don’t pause to consider, however, what might be unjustifiable applications of AI, where humans would be indisputably the more efficient and effective agent. I suspect that is because, as the planet’s reigning singularity, we obsess over what might take away our godlike powers, rather than how those powers should be used. To generate bigger profit margins? Make deadlier drones? Ease suffering? Give us more leisure?
By demurring, collectively, on this vital question, we’ve allowed capitalism to decide for us. Companies are rushing to ringfence the world’s silicon and quartz supplies and reshore the facilities that make the advanced chips that AI requires. They’re ramping up investment in energy sources to meet AI’s exponentially-growing energy requirements. They’re using AI to streamline supply chains, speeding the pace at which goods move around the world so that we can consume more of them, faster.
Doubtless this reallocation of resources will juice GDP and grow our economy. But in a world of finite natural resources and ever-frailer human connections, is this an outcome we believe justifies AI’s disruptive impact on society and the planet? Shouldn’t we at least discuss what it is we’re buying, when we buy into AI wholesale, before absorbing its staggering costs?
AI is an amazing tool. Let’s try to get clear, though, on what warrants its application. Because for 20 watts a day, there’s really no problem-solver more brilliant, efficient, or welcome than a human being.
