OpenAI's Future Society: What's Wrong with It

OpenAI has released a document Industrial Policy for the Intelligence Age. It is useful to read it as a political statement.

OpenAI offers a lens through which the state and business will discuss AI in the coming years: labor, taxes, energy, social policy, control over systems, international coordination. An obvious claim to participate in designing institutions.

If we compress the text into large blocks, OpenAI proposes the following.

Labor and Access.
Workers are offered a formal voice in AI implementation. Schools, libraries, small businesses, and less affluent regions are to be given expanded access to basic AI capabilities. This block also includes payments during spikes in the labor market crisis, projects for a 32-hour workweek, and the idea of converting efficiency gains into money for employees.

Redistribution.
The document proposes restructuring the tax base when income shifts from wages to capital, corporate profits, and automation. It separately discusses a Public Wealth Fund through which citizens can receive a share of the benefits from AI growth. Universal basic income, in short.

Infrastructure.
OpenAI directly states that data centers should pay for their appetite without hidden subsidies from households. At the same time, it proposes accelerating the expansion of networks and transmission infrastructure.

Control and Risk Management.
For powerful models, the document proposes provenance verification, audit logs, special inspection regimes, threat localization instructions, restrictions on state use of AI, and mechanisms for public participation in setting ethical frameworks.

Substantively, this document has strong points. OpenAI acknowledges that productivity growth alone does not guarantee increased prosperity for the majority. The company allows for a scenario where AI benefits are concentrated within a narrow circle of firms, while workers face greater turbulence without symmetric profit.

But do not be mistaken, this is not a utopian text with Altman's puppy-dog eyes. All this work serves OpenAI's own interests well (otherwise it wouldn't exist):

1. Acceleration is considered the baseline scenario. The document discusses managing consequences, mitigating costs, redistribution, and safety. The question of whether the race for ever more powerful models is justified is barely touched. For OpenAI and co., this is a comfortable starting position.

2. The problem of concentration is described in the logic of subsequent redistribution. First, a few companies get computing power, top researchers, access to government contracts, and control over infrastructure. Then the state is asked to devise taxes, funds, and compensation to return some benefits to society. This framework hardly disrupts the market structure on which the AI labs themselves grew.

3. The idea of special rules for the most powerful systems seems reasonable, but at the same time creates an institutional barrier around a few players. The mass layer of applications and weaker models remains relatively untouched. The upper tier of the market becomes encrusted with audits, trusted institutions, complex procedures, and special status. This architecture reduces risks. At the same time, it reinforces hierarchy.

4. OpenAI clearly wants to participate in creating the legislative environment. At the end of the document, the company offers grants and up to $1 million in API credits for those who will further develop these ideas. This is the same familiar influence on the circle of experts, the discussion agenda, and the list of issues that will be considered central.

The document should be read carefully and with a cool head. It contains a lot of specifics, sometimes more than you would expect from a corporate text. At the same time, it is an attempt to pre-describe the future order so that the leading role of the largest companies appears natural, useful, and without alternative. Bees are not against honey.


❗️❗️❗️❗️❗️❗️❗️❗️ / Not banned in Russia