WHY AI IS THE PRODUCTIVITY BOOST WE’VE BEEN WAITING FOR

Since the unveiling of ChatGPT and GPT-4 there has been no shortage of concerns expressed about artificial intelligence.

AI is going to ruin the education system. AI is going to wipe out a large swath of white-collar jobs. Some highly credentialed people have even suggested that AI may eventually wipe out the human race.

In fact, AI in the style of GPT-4 and its successors is more likely to provide the much-needed productivity boost that advanced economies have been missing for a decade or more. But it will take some clever thinking from business and government to turn this into a reality.

The best way to think of large language models like GPT-4 is as what economists Ajay Agrawal, Joshua Gans and Avi Goldfarb term “prediction machines”. What AI does is reduce the cost of prediction – often dramatically.

Now that may not sound like a big deal at first, but almost every issue that a business or an individual confronts involves making a decision under uncertainty. Think of a business considering whether to launch a new product, enter a new market, make a capital investment, or hire a new employee. Or think of an individual deciding whether to move jobs, invest in their education, or buy a house.

These are all examples of decision-making under uncertainty.

And as everyone who has taken an introductory economics course knows, when the uncertainty is reduced, the decision gets better. In other words, AI is a decision-improving machine.

This is good enough, but the real payoff can be much bigger. As Agrawal, Gans and Goldfarb articulate in their fantastic new book Power and Prediction, the real power of AI is when we move from prediction to transformation. Cheaper and better predictions make it possible for businesses to redesign their entire decision-making processes. And that’s where the big productivity prize lies.

Australia may or may not be at the forefront of AI development – time will tell. But we can lead the race for system redesign. As the authors point out, a good example of this decision-making system redesign comes from yacht racing. It has long been the case that computer simulators have been used to design America’s Cup yachts. But in preparing for the 2021 Cup, Team New Zealand realised that AI could be used to determine the best sailing tactics.

Previously, human sailors would refine tactics in a computer simulator. But Team NZ used AI to run tactics simulations 24 hours a day at high speed. That taught the human sailors who actually had to race the boats a lot. But the big gains came from the fact that boat design could be improved by understanding the feedback loops between design and tactics. The improvements in boat design allowed for better tactics, and so on. This kind of system redesign was only possible with AI.

This harks back to an earlier era when electric power was invented. Electric power was superior to steam power because it loses less energy. So factories gained from removing steam power and plugging in electric power.

But the big gains came from redesigning factories altogether. Factories could be moved away from water sources, and they could be laid out in one big floor rather than an unwieldy multi-story plant attempting to keep production processes close to the single steam power source. All of a sudden this made the modern assembly line and mass production possible.

Much has been made of the open letter signed by Elon Musk, Steve Wozniak, Yuval Harari and others calling for a six-month pause in the development of AI. Their warning was ominous: “AI systems with human-competitive intelligence can pose profound risks to society and humanity.” Well, possibly.

But it’s worth noting that we shouldn’t mistake large language models for artificial general intelligence: recursively self-improving machines with the potential to escape human control and do untold harm (or good) to humanity.

There’s not just a big gap at the moment. It’s completely unclear that gap can be bridged. Moreover, there are good reasons to believe that artificial general intelligences – should they ever arise – may be self-regulating, or regulatable.

It’s also striking that Musk – while calling for everyone else to hit the pause button – has recently hired two leading AI researchers and bought 10,000 GPUs for Twitter’s own AI project. Coincidence or conflict of interest? You be the judge.

Moreover, no lesser expert than Bill Gates came out against the pause, saying “I don’t really understand who they’re saying could stop, and would every country in the world agree to stop, and why to stop” (emphasis added). Artificial intelligence has the prospect to give advanced economies like Australia an incredible productivity boost. But that won’t happen simply from improving decision-making. It will come from using thatimprovement to redesign organisations in a fundamental way. That will take encouragement and support, not pumping the brakes. There’s a clear policy message from this. Australia may or may not be at the forefront of AI development – time will tell. But we can lead the race for system redesign. That will involve collaboration between business and government – finding policies that lower the cost of system redesign, not just taking advantage of the lower costs of prediction which are coming anyway.