Podcast version:
On Anthropic and the Challenge of AI Integration
Freeform thoughts on Anthropic’s latest papers which highlight the challenges non-frontier companies face to integrate AI and help the AI economy reach attain its desired payoff.
Anthropic recently released “How employment is changing at Anthropic.” This report seeks to model for other companies just how far LLMs can be implemented to drive productivity and improve profit margins. While many have balked at the predictions of Dario Amodei (CEO of Anthropic)—such as him saying software engineering will be obsolete by 2027—as they fail to see the results, the truth is nuanced.
It reminds me of tensions between Boomers and Millennials. While Boomers were better savers than Millennials, the Millennials would not have been close in accruing as much wealth as the Boomers if they had the latter’s thrift. The gripe that Millennials should be saving more, and the gripe that Boomers miss the need for updated life models, therefore, both contain truth.
Similarly, Dario Amodei’s predictions, I believe, are true. Software engineering could go obsolete by 2027, but only if they integrate AI with the same enthusiasm as Amodei’s Anthropic. Both the prediction and the subsequent griping are true in a sense. Earlier this summer, Google put their finger on a significant gap between AI ubiquity and AI integration. Despite 90% of the workforce now being exposed to AI, the technology has not been integrated anywhere near its potential. And that is why Anthropic’s latest report aims to spark better integration.
Below, I surface some key findings that nod to the sources of tension that threaten not only Amodei’s predictions, but the ability for the AI economy to unfold before public opinion effectively blocks it.
“The post-World War II union-management settlement forged in the 1950s rested on exactly this kind of bargain: labor unions like the United Auto Workers and the United Steel Workers allowed management prerogative over work reorganization (and even intensification) in return for pay increases tied to productivity increases and to a rich array of benefits.”
Comparing the Industrial Revolution to the AI Revolution is not always a true analogy. It is analogous that both are creating a new market by tapping into resources in a technologically novel way: the Industrial Revolution harnessed natural resources to create new products and markets; the AI Revolution harnesses the textual artifacts of human intelligence to create a superintelligence that may generate new products and markets. They are disanalogous in terms of the material that they harness, as just noted.
Because of this, I do think that history will repeat itself with respect to how AI will drive an ideology of world peace. Amodei’s prophecy of global, democratic peace echoes the Manchester School that pushed an ideology of free-trade that promised world peace, eradication of hunger and poverty, etc. in the Industrial Revolution. The Manchester School was a key influence on Marx.
These prophecies did not come to pass in the Industrial Revolution, so there is reason to be skeptical of AI’s promises. One must argue that the prophecies failed because of technological or geopolitical constraints that are no longer in the way as AI enters the scene of world history.
In any event, the appeal to labor unions strikes me as disanalogous. The report itself states that workers are positive about AI if the increasing demand of productivity translates to increasing compensation.
Yet, the report also states:
“First, widespread stock ownership gives employees a strong interest in the company’s success. Of course, this is an easy posture for a company that has seen substantial increases in valuation. For many firms, shared ownership may be insufficient or otherwise undesirable.”
Anthropic has had to increase compensation through ownership share to complete AI integration. As the report itself mentions, AI integration is half-hearted without additional financial incentives. This is rather common-sensical, for workers aren’t organically driven to fully implement a new technology that makes them more disinterested, replaceable, and yet accountable for radically higher levels of productivity.
However, as the quote above rightly identifies, only companies with a substantial increase in valuation can pull this lever. In effect, only the leaders of the AI frontier—like Anthropic—can accelerate the AI integration to have a chance at securing the desired payoff.
Moreover, even if AI integration is accelerated, the payoff is only secured if new products and markets emerge. Otherwise, the integrated AI technology will accelerate the replacement of human workers instead of extending economic opportunities:
“But prioritizing reallocation forces AI-adopting companies to consider ways that AI can create new tasks for new products and new markets, rather than funneling AI applications into replacing human workers to save on labor costs. This could help employers resist the temptation of automation in which AI or other technologies just barely outcompete human workers: employers replace human workers, imposing substantial social costs in lost income for displaced workers, while delivering minimal productivity benefits.”
We have become accustomed to think of the AI frontier as a race between nations, but there is also the at-home race for AI integration to reach its payoff before displaced workers sour public opinion, crumbling the political will to continue the AI economy.
Anthropic is enthusiastic about their successes in integrating AI and making their operations more lean and scalable. The bearish case does not lie in this potential, for Anthropic’s report proves it—just as the technological progress hitherto has been pointing to. I’m quite bullish on the ability for AI integration to go even further than Anthropic outlines.
Rather, the bearish case lies in the (tacit or explicit) lack of will to integrate AI in non-frontier organizations/companies, and the inability for those organizations to implement financial incentives to provide a counterforce—as was needed in the post-World War II era via unions.
If the lack of will is not addressed, then the payoff of an AI economy will not be reached before public resistance crumbles the whole experiment. That is most immediate, concrete race in the American economy.
What I’ve outlined above is the bearish interpretation of Anthropic’s own report. There may still be a bullish case to be made. Perhaps there are solutions to fill the holes that the report surfaces. For example, the increasing valuation of Anthropic may give way to the the opportunity for wealth to be redistributed to provide the funds to catalyze AI integration in non-frontier companies. Or, perhaps some other lever exists to drive AI integration other than shared ownership of a company. Or, perhaps no lever is needed as another catalyst to AI integration in non-frontier companies may emerge organically—there may just be a non-ideological lag in integration.
Either way, the report is right to provide a sense of urgency around AI integration.