Worker dignity is important; but measuring it solely by income share pushes culture, security, merit, and local belonging into the background.
Acemoglu’s own observation (“working classes have grown cold toward social engineering”) confirms this: people want not only transfers, but not to be looked down upon. Jacobin’s demand for more class organization and a “war” accounting can fall into the same trap: organization, identity, and a foreign-policy line can quickly evolve into majority veto, party militia, or cultural imposition.
“Institutions should steer technology” is not an innocent sentence. Who will steer it? Which commission, ministry, or party majority? Acemoglu’s own theory says extractive institutions are designed by elites. Therefore the claim of steering AI “on behalf of society” can produce a new extractive layer: procurement, data, computing power, and standard-setting pile up in a single center. The China example already challenges this formula; relatively extractive political institutions made high growth possible for a long time.“Right institutions = the right direction of technology” is not a universal laboratory result.
“The market chooses wrongly, therefore let’s regulate” is incomplete. The attention economy and profit priority of big tech platforms are real.But “let’s regulate AI and social media to protect democratic discourse” in practice turns into the authority to decide who gets to speak. If the algorithm distorts for profit, the state/party algorithm distorts for votes. The suspicion applied to the company is not applied to the public monopoly. This asymmetry contradicts the limitation of power at the core of liberalism.
Jacobin’s call for more worker organizations and class-based steering transfers the same asymmetry to the left elite and the union apparatus. The definition of “worker-friendly AI” is vague and disproportionate. Is a model that speeds up an accountant’s work worker-friendly, or anti-worker if it shrinks the staff? The same software produces two results in two firms.
In Acemoglu’s own macro studies, AI’s 10-year total factor productivity effect is modest. Against this picture, “AI will destroy liberalism, therefore let’s politically steer technology from now on” is exaggerated. Slowing frontier research as “less profitable but worker-friendly” also carries the risk of shrinking the pie to be shared. The community emphasis also easily goes astray.
Correcting liberalism is not done by sacrificing individual rights to society; it is done by properly producing public goods (security, schools, infrastructure, transparent budgets). The risk of inequality and power concentration is real. What is wrong is automatically posing the solution as “more steering + more regulation + a political definition of the right technology.”
If Acemoglu’s institutional theory is applied consistently, the real thing to fear is not only private companies’ AI, but the apparatus that manages AI and public data from a single hand in the name of “shared prosperity.” Jacobin’s class-war-organization prescription increases the chance of legitimizing this apparatus; it does not reduce it.
In other words, both thinkers fail to solve the problem of limiting power; they only change whose hands it will be concentrated in.
That is why a cyber-secure, closed-circuit, domestically infrastructured public AI layer is necessary. This layer does not profile the citizen; it traces money and authority (budget deviations, procurement anomalies, incentives, staff inflation, SOE losses). The model does not say “who is right”; it flags anomalies.
The decision remains political, but its raw material cannot be hidden. Government, opposition, and all political groups look at the same raw data, the same query language, and the same model version.
@vonderleyen
@BarackObama
