In a recent commentary on regulating AI, Roger McShane, deputy U.S. editor of The Economist, wrote: “We all know that innovation and regulation are inherently incompatible. Once they put speed limits on roads, manufacturers stopped producing new cars. When air-traffic control was created, airlines stopped expanding. And when it was decided that new drugs should be tested to see if they actually work, pharmaceutical companies halted all research.”

While these lines were obviously written sarcastically, there is actually a strong case to be made that regulations can indeed often stifle innovation. For instance, pharmaceutical regulations in Canada are a real quagmire, governed by various entangled regulatory bodies, both provincial and federal. Partly because of this, new innovative medicines take substantially more time to reach Canadian patients than they should. In fact, Canada is last among the G7 and 19th out of 20 OECD countries in terms of approval delays for innovative drugs. People suffer and even die as a result.

Similarly, innovation in the car industry is infinitely slower than in the AI sector. Several years ago, I was an investor in a small startup company based in a suburb north of Montreal that was aiming to introduce a new (and tangibly longer-lasting) generation of brakes. It was ultimately killed, in part, because of the regulatory process. My point here is not that there shouldn’t be any regulations on new generations of brakes, because it does indeed come in handy to be able to actually stop your vehicle when you need to. Rather, I am saying that it is in fact quite difficult to calibrate regulations in an optimal manner and that regulators, when faced with the choice between perceived security and innovation, will almost always choose the former.

Furthermore, we shouldn’t lose sight of the fact that the makers of new medicines, or of brakes, or of planes, or whatever else, have a distinct and inherent built-in incentive of their own to avoid killing their clients, regardless of any government dictates. Such is the power of reputation (and fear of liability) in a competitive market setting.

But this bias in favour of security (which again is deeply embedded in the very incentive structure provided to regulators) also involves risks. However, our ability to foresee these risks linked to inertia and delays is intrinsically weak. So, too, is our ability to see the knock-on effect of regulations, including those related to price controls. On this point, see Hayek’s “The Use of Knowledge in Society” and Bastiat’s “That Which is Seen, and That Which is Not Seen.”

On the flip side of that, a government is not some sort of benevolent and all-knowing abstract creature. Rather, it is simply a collection of bureaucrats and politicians. Several of them may be well-meaning, though I see no reason to believe this would be truer for them than for any other subset of humanity. In any case, an important question to ask is: Do these bureaucrats and politicians have the level of knowledge required to properly regulate an ever-changing frontier technology like AI? To ask the question is to answer it.

Moreover, there is the issue of what economists call “regulatory capture.” As J.D. Tuccille wrote recently in the National Post, huge industry-leading corporations can better absorb regulatory costs than their upstart competitors. It is, therefore, in the narrowly defined self-interest of incumbent firms to lobby for regulation, both to have a say in how it is shaped and to protect their existing dominant market position. As Tuccille concludes, despite the scare stories they’re peddling, AI executives are “almost certainly less concerned about dangers to humanity than they are about threats to their market share.”

In summary, it is true that not regulating AI, or regulating it very lightly, does involve potential risks and costs. But so, too, does overregulating or badly regulating it.

In that context, the self-regulating framework for AI put in place by the Trump administration, combined with the set of voluntary safety standards just signed by major AI executives, seems like a sophisticated and balanced approach to addressing these issues. It finds a wise middle ground on the matter. And this assessment comes from someone who otherwise believes that Donald Trump is generally a calamity for the world order as well as the future of classical liberalism and the rule of law. But if, say, a really bad person loves dogs, that’s certainly not a valid reason for hating them.

Yes, some interpretation of the verdict of the recent American midterm elections on these issues is that U.S. voters are skeptical and even angry at AI bros and feel angst against data centres. However, this could very well ultimately create an opportunity for Canadian governments to stay clear of knee-jerk reactions and increase our competitive edge rather than trying to always emulate (and outgun) the worst impulses of American politics.

This article was originally published at The Hub. Republished with permission.

← Back to Perspectives