The Acemoglu Wars Are All About AI

Regarding that Economist article ...

When I wrote about how terribly wrong Noah Smith was about Daron Acemoglu and Simon Johnson, I didn’t know that I was volunteering for service in the Acemoglu Wars. The Economist had not yet published its piece,“The world’s most influential economist is oddly unconvincing. Daron Acemoglu is revered. Why?” This rather remarkable article cites Noah’s criticisms in support of its main thesis, as well as comments by Tyler Cowen,¹ and has very much become its own Thing.

I have other stuff that I would have much more fun writing about (specifically: the interesting connections between the movie Backrooms and Susanna Clarke’s novel, Piranesi). Still, there is something useful to be extracted from the more recent imbroglio, which I wanted to write about in my original post, but didn’t have space. So then, once more into the breach …

I’m not going to talk about the various ways in which the anonymous Economist writer’s piece on Acemoglu mistook debates among economists. Others have done that already, and spats over econometrics do not play to my strengths. I do want to ask why the Economist’s writer chose to wade into this particular morass, since it is clear that spats over econometrics did not play to their strengths either.

The article was written, to question, as the anonymous writer puts it, “whether [Acemoglu’s] reputation at the very apex of the economics profession—and the intellectual influence that goes with it—is justified by his scholarship.” As the writer themselves suggests, this is relevant because of what Acemoglu is saying about AI.

These academic disputes could end up mattering a great deal. Mr Acemoglu may not think AI is about to turbocharge productivity, but he does worry about its effects on democracy and jobs. His new book focuses on its potential social harm rather than its potential benefits. His proposal to limit that harm, “pro-worker AI” that would augment the value of human labour rather than replacing it, sounds terrific—but also obvious.

Mr Acemoglu’s influence in the AI debate is clear in a recent statement, signed by dozens of prominent economists, which argues that “we must act now” to “steer AI in a direction that complements humans and benefits society”. But who are “we”? And who is to decide what sort of AI does or does not complement humans? Even economists who signed the petition say they are not entirely sure. So great is Mr Acemoglu’s stardom that it can sometimes blind the critical faculties.

The writer (who I imagine does not represent the views of all Economist writers) prefers a different approach to Acemoglu: Let Technological Progress Rip.

The book plays down a more compelling claim: that technological progress in itself has made the average person many times richer than in pre-industrial times.

This ties together with what Noah has said already, and with a particular ideology that is propagated by a highly influential subset of Silicon Valley entrepreneurs and economic commentators. Marc Andreessen lays out the case for this ideology, in strikingly lively terms, in his “Techno-Optimist Manifesto”:

We are being lied to. We are told that technology takes our jobs, reduces our wages, increases inequality, threatens our health, ruins the environment, degrades our society, corrupts our children, impairs our humanity, threatens our future, and is ever on the verge of ruining everything.

Truth. Our civilization was built on technology. … Technology is the glory of human ambition and achievement, the spearhead of progress, and the realization of our potential. For hundreds of years, we properly glorified this – until recently.

Andreessen’s manifesto was its own declaration of war:

We believe any deceleration of AI will cost lives. Deaths that were preventable by the AI that was prevented from existing is a form of murder.

and the effort to discredit Acemoglu seems to me to be one offensive in that broader conflict. On the one side is a gallant alliance of CEOs, venture capitalists, econbloggers and journalists-without-bylines,² fighting the good fight for AI and technological progress. On the other is a dismal conglomeration of oddly unconvincing economists, shabby intellectuals, smelly left-activists and other such human detritus, purporting to “steer” technology and send us all stumbling down the road to serfdom.

Or so it seems. The problem is that “technological progress” is not the undifferentiated force that Andreessen, the Economist writer and others take it to be. The question is not whether or not it ought to be steered, but which course will be taken and who ought be entrusted with doing the steering.

******

Why was this unimpressive article written? Put more spicily, how exactly does it contribute to the broader war effort? The answer lies in the disconnect between the mood-music of the economics profession and what its theories about technological innovation can actually tell us. Economists, by and large, are in favor of market freedoms as a path towards greater human prosperity and happiness. After all, well established economic theories say that one can’t improve on markets as a way of connecting means to wants under reasonable seeming technical conditions.

Unfortunately, however, technological innovation doesn’t fit readily into the standard framework. Assumptions about e.g market actors having complete information about the world sit awkwardly with the fact that innovation necessarily involves discovering things that were not previously known.

This hasn’t stopped economists from trying to understand how key aspects of innovation can be reconciled with standard economic theories, and they have done a lot of very interesting thinking. I’m not going to summarize the debates leading from Robert Solow’s work to Paul Romer’s ideas about endogenous growth and beyond. Acemoglu’s own work on directed technical change approach explains how different paths of economic development may be biased in the direction of different “factors of production”: most importantly, labor or capital. But I think that it is fair to say that while substantial progress has been made discussing the consequences of technological change, explaining the trajectories that are discovered is still a hard problem, and liable to remain a hard problem, both because it involves processes of discovery that are inherently social and unpredictable in ways that fit badly with the useful simplifications that economists prefer. The result is that folk theories flood in to fill the unfortunate intellectual gap between what many economists want to be true and what they can show to be plausible.

Noah, in his response to Acemoglu, Johnson and myself, seems to believe that technological innovation is what happens when a brilliant billionaire and an undifferentiated mass of consumers love each other very, very much. That is why we need to keep brain-geniuses like Elon Musk as closely attached to the side of the angels as we can manage.

General Motors and Ford have struggled and failed for decades to make commercially viable electric cars, but only Elon Musk was able to do it. NASA struggled with spiraling costs and declining launch volumes for decades, but Elon Musk was able to create a company that out-launched China and NASA put together. Now at a time when America and its old-line companies are struggling to rebuild the semiconductor industry, the only man who seems able to build the requisite fabs is…Elon Musk, once again.

Elon Musk, I believe, is not a good man. He is a cruel and selfish man. I don’t believe he murdered millions of people when he dismantled USAID, but I believe it was still a terrible and immoral thing to do. And he has supported some of the worst politicians in America and elsewhere. And yet, he is an effective man — almost preternaturally so. In many domains, his individual entrepreneurial and organizational abilities are the only thing that has kept China from steamrolling the U.S. in key technological sectors.⁷

What do you do with a man like that? The only thing that makes sense is to utilize his talents as best you can, and do your best to prevent him from subverting the national interest along the way. The same goes for the rest of the techbro class. Only in a few eras do we decide to worship titans of industry like Steve Jobs. But in every era, we need them.

Shorter Noah: love ‘em or hate ‘em, without techbros, we would be unable to innovate or get things done and China would devour us.

This theory of the world is unsurprisingly popular with the “techbro class.” And it probably has some explanatory power. As Quinn Slobodian and Ben Tarnoff say in their otherwise unsympathetic book, Musk did in fact get important things done at SpaceX. But Great Man Theory is, in the end, notably difficult to test. If we were able to run the last twenty years of tech development 100,000 times, randomly substituting different qualified people into key roles, perhaps we would indeed discover that Elon Musk and his brethren made an unique and indispensable causal contribution to the progress we have made. But perhaps we would not. What we do know is that the actual Elon Musk that we have to live with on this here timeline was flagrantly incompetent in government, contrary to Noah’s previous predictions.

Neither economic theory or econometrics tells us then, whether it is best to have technological development directed by a self-selecting elite of bold striving entrepreneurs, or whether some other social and economic arrangement might work better. Claims about the indispensability of Elon Musk et al. stem more from the sensibility and conventional wisdom of some economists (by no means all) than well-established theory, let alone testable empirics.

What we do know from economics, is that trajectories of technological development are often associated with increasing returns. In other words, when you choose a path, it may become self-reinforcing.

That has implications for economic efficiency. What if the trajectory that you choose turns out to be less efficient than some other possible trajectory? Perhaps we will be stuck. You can claim (and some libertarian economists have claimed), that far seeing capital markets will be able to guide us from the wrong trajectory to the right one, but that may be problematic given even moderate uncertainty. What if the trajectory systematically favors one class of actors (say: those whose advantages stem from their possession of capital) over another set of actors (say: those whose advantages stem from their labour power)? One group may end up on the wrong side of the economic bargaining relationship for a very long period of time. Finally, innovation does not happen in a vacuum, but in the context of institutions, which can systematically affect winners and losers. Those institutions are themselves path dependent. And that brings us back to the anonymous scribe for the Economist, who writes:

Nations prosper when institutions are good, and stagnate when they are bad. True. But what, exactly, is an institution? Rules, norms, enforcement, culture—everything, really. Where do institutions come from? “Critical junctures” and “institutional drift”

I don’t know whether the sarcasm in the last sentence stems from sincere ignorance or deliberate obtuseness. But “critical junctures” play a crucial role not just in pathways of institutional development but in technological development too. A critical juncture is the short period when one path is selected rather than another, leading to one set of self-reinforcing increasing returns phenomena being chosen, and another being left to wither amidst the tares.

Will the path that is chosen be the one that maximizes overall economic welfare, however understood? Will this path turn out to systematically advantage one group of actors at the expense of others? If you think that technological innovation is an increasing returns phenomenon (my sense is that this is not a hugely controversial claim among economists, though you absolutely could challenge it from other perspectives), then you absolutely want to pay attention to critical junctures. These are the moments at which the one trajectory rather than the other gets picked. Dismissing such conjunctures in a sneering aside is to ignore the argument that the other side is making rather than to address and refute it.

And that, in turn, brings us back to the politics. The Economist writer says that Acemoglu is wrong to propose that AI be “steered,” since “technological progress in itself has made the average person many times richer than in pre-industrial times,” helping themselves to two invisible assumptions: that AI development will not be steered if it is left unregulated, and that unregulated technological progress trumps regulated technological progress.

The first assumption is visibly untrue. AI’s development is being steered, right now, by a small self-selecting group of business leaders and investors. Do their interests reflect the general good? Perhaps you can make a case for that - but you need to make that case.

Has the technological progress that has historically made people richer than in pre-industrial times been untrammeled by institutions and regulation? Obviously it has not. You can certainly argue that it would have been better, or faster, had it not been hindered by bureaucrats well-meaning or malign, and that claims about e.g. the vital role of the entrepreneurial state are stuff and nonsense, but you would have to make that case, rather than just filling in the argumentative gap with grandiloquent self-confidence.

******

As the title to this post suggests, I believe that the sudden outburst of hostility to Acemoglu has everything to do with the historical moment that we are now in. The article’s own conclusions suggest that it might not have been written if AI regulation was not a live political issue, if we were not, in Acemoglu’s terms, at a critical juncture where we will make one choice, rather than another. AI is surging, and there is ferocious debate about its consequences. On the one side, those like Andreessen, who argue that regulation and such are the “enemies” (his word) of technological progress, and need to be defeated. On the other, a variety of people who believe that AI ought be regulated (even if they disagree about how).

I don’t have any profound objection to people claiming that unregulated markets are the necessary best path towards AI that will benefit everyone. I think that they are wrong, on the basis of past historical experience, but I want them to make the best case that they can, since there’s obvious value to market-led entrepreneurialism. But, and I stress this, I want them to make the best case that they can. Simply assuming they are right, and pretending that the counterarguments do not exist, is both intellectually shoddy and unhelpful. It is also personally vexing. I really would prefer to be writing about topics where I don’t yet know what I think, rather than being stuck doing garbage pick-up yet again.

The actual choice is not between some undifferentiated path to “technological progress” and its wicked, envious or misguided opponents. The choice is between different paths to progress, and the live question is which groups or structures are best suited to identify better paths, and avoid worse ones, and (much more ambitious; not the topic of this post) to figure out mechanisms of self-correction when things go screwy. It is neither brilliant nor original to say that any such groups, structures and mechanisms will involve some mix of markets, bureaucracy, democracy, and perhaps other social technologies of discovery, coordination and self-correction too, rather than the sole predominance of the one or the other. But figuring out better mixes and avoiding worse ones is really fucking hard. It would be nice to see more of the pro-market people acknowledging this and engaging accordingly.

1. Tyler coincidentally wrote that Noah was right in his fight with me over Acemoglu and Johnson, without explaining further. I wonder whether he still maintains this, given subsequent developments, but don’t particularly expect to find out.

2. I’d make a sarcastic reference here to “Justice is like the hawk. Sometimes it must go hooded,” but in fairness, Economist journalists are supposed to be anonymous, even if they divulge their identity more often these days than they used to.