Chris Williamson argues the AI regulation conversation is one-sided — focused entirely on preventing harms while neglecting a "theory of AI goods." He contends that for AI to serve the public, problems need funding and legible data, which governments often fail to provide. His central example: the IRS could build an LLM to automate tax filing, since it already has income data and owns the tax code, yet government only thinks about what to stop AI from doing.
Watch on YouTube ›Get the market thesis, key claims, assets, contradictions, and follow-up questions from any financial video — then unlock a version personalized to your portfolio, watchlist, and favorite speakers.
Chris Williamson opens with a critique: the AI policy conversation is overwhelmingly about what we don't want from AI, with almost no discussion of what we do want. He argues this asymmetry needs correcting — we need a "public agenda for AI" and, in his framing, a "theory of AI goods." He lays out two prerequisites for AI to solve a public problem. First, money must be behind the problem, because meaningful compute remains costly. Second, the problem must be "legible to the system," meaning the government must create structured data that machine learning can ingest — and he notes this is something government routinely fails to do. His concrete use case is the IRS automating tax filing via an LLM. The IRS already holds income data (ground truth) and writes or shapes the tax code. Williamson argues there is no reason most people should need to pay an accountant. …
No market-relevant macro bias is expressed in this transcript. Williamson's argument is about AI governance and public-sector technology policy, not financial markets, asset prices, or economic conditions. The near-term setup he describes is a policy landscape dominated by precautionary AI regulation with no affirmative public-sector deployment agenda.
Over the coming months, the AI policy conversation may begin to incorporate a "positive agenda" framing if advocates like Williamson gain influence — but this depends on political will and is not a market-moving dynamic. No financial market macro bias is expressed.
The structural question is whether governments build the data infrastructure and procurement frameworks to deploy AI for public goods. This is a multi-decade institutional-capability question, not a near-term market call. No financial market macro bias is expressed in the transcript.
The public sector is only thinking about what it wants to prevent AI from doing, neglecting a necessary theory of AI goods.
The speaker notes that government focuses on AI harms (which the speaker affirms as important) but lacks an equally developed positive agenda for what AI should achieve in the public interest.
The IRS could deploy an LLM that does citizens' taxes with them, eliminating the need for most people to pay an accountant.
The IRS already knows income (ground truth), knows the tax code, and helps write it, so the data and rules are available for a conversational tax-filing system.
An AI concierge could navigate any government service for a citizen because it would know the citizen, their situation, and what government programs can do.
The government is currently very hard to navigate; an AI that knows the person and the available programs could bridge that gap.
Unlock the full claims, asset map, scores, related transcripts, follow-up questions, and AI chat — shaped around your portfolio, watchlist, favorite speakers, and risks.
