Thoughts from an AI skeptic

Bubble? Did I say bubble?

There are several aspects of the AI revolution about which I am skeptical. I think it’s more than the reflexive skepticism of an old man confronting a new technology he only partially understands. But you be the judge.

Let’s begin with data centers.

Data centers have become a constraint on the ability of AI companies to move as fast as they would like. The centers take up acres of real estate and consume large volumes of water and electricity.

While AI might be a big boost to overall productivity and economic growth, data centers really aren’t for the communities in which they are located. They initially produce a large number of construction jobs, but are not job intensive enterprises once up and operating. After construction, they are economically stagnant. They don’t produce knock-on economic activity or opportunities.

There is no reason for a state to subsidize them, as Arizona does. Gov. Hobbs is right to call for the repeal of our data center tax incentive.

Data centers can, if not subsidized, produce a large amount of property tax revenue for the host city or county. And there are not a lot of externalities once built. They just sort of sit there, not generating much traffic or activity.

Approval of them should rest at the municipal and county level, with local officials weighing the tradeoffs between alternative land uses, potential property taxes, and water and electricity consumption.

This doesn’t relieve the constraint data center approvals are creating for the pace of AI buildout enthusiasts insist is necessary. And that leads to the premise about which I am most skeptical.

AI companies are vacuuming up unfathomable sums of capital based upon the belief AI applications will generate equally unfathomable revenues. And that those revenues will flow overwhelmingly to the AI system that becomes the best the quickest and stays there.

AI companies and capital markets see this, in large part, as a winner-takes-all competition. So, spend big now. And override those pesky local approval processes holding things up.

I’m willing to believe that there will be broad adoption of AI applications, and even that it will occur sooner rather than later. However, this will, in very large measure, be to improve general business processes – coding, accounting, inventory control, supply-chain management, correspondence, trouble-shooting customer service problems.

For the overwhelming majority of these tasks, the best, cutting-edge AI system won’t be necessary. The second, third, fourth, or fifth best will be able to do the job well enough.

My guess is that to a very large degree, broad adoption of AI applications will be a price-sensitive value proposition with many suppliers, not the winner-take-all competition the capital markets are funding. Being the best the quickest and staying there won’t guarantee a commanding market share.

There are three areas where being the best the quickest and staying there would seem vital enough to merit a premium price: military applications, medical diagnoses, and cybersecurity. If there does end up being a stratified market with many suppliers, a lot of current investors will end up crying in their beer.

Now, a bit about the public policy implications of all this.

Equity markets are being bolstered by the race to be the best the quickest and staying there. If a lot of current investors end up crying in their beer, there will be a shake out, and perhaps a painful one.

The Fed keeps an eye on equity markets. However, the Fed hasn’t demonstrated the ability to recognize a bubble as it is forming, or to effectively ameliorate one or clean up the aftermath. If there is a shake out, those being shook out will argue that national security and economic competitiveness require a bailout. Except, perhaps, in the three areas where being cutting edge is truly vital, the answer should be no.

OpenAI has already floated the idea of the U.S. government taking an equity position in AI companies. The statists in the Trump administration are interested. Bernie Sanders thinks it’s a swell idea. This would move companies potentially riding a bubble into the “too-big-to-fail” category, making a bailout much more likely. Very bad idea.

I’m not a Luddite. I doubt that the broad adoption of AI will result in large-scale, irreplaceable job losses. Capital will continue to be creative in taking advantage of, and fully employing, both technology and human labor. As it always has.

However, if this time is different, and the productivity gains of AI don’t also drive labor redeployment and wage gains, there would be a role for government to offer income support to workers. But I don’t think this needs to be set up in advance, as some are advocating. I don’t think AI adoption will move too fast for public policy to catch up with its consequences as they become known, rather than being speculative or feared.

That also applies to a regulatory framework. There should be a national one, rather than a state-by-state approach. But it also can be worked out as the technology develops, adoption progresses, and issues emerge and mature. Some, such as deep fakes and cybersecurity threats, already have. However, getting the regulatory balance between safety and innovation right in the abstract — as opposed to striking it for specific, concrete issues — is highly unlikely.

I suppose I should confess that I don’t use AI. So, if you want to dismiss all this as ill-informed ramblings of an intellectual fossil, you have a point.

And, if you insist on knowing, I’ve not taken a ride in a Waymo either.

Reach Robb at robtrobb@gmail.com.