AI
AGI Is Fiction; DevOps Is Obligatory
Sam Altman said on stage in Paris in February 2025 that nobody knows what AGI means. Bryson’s definition: AI that is exactly like a person, only somehow immortal or omnipotent. Against that she places DevOps, which she calls obligatory. Someone always decides what counts as data, which algorithm to use, when the tests have passed, when the product is ready for release, and how long to support it after shipping.
“The data programs itself.”
That claim, made to government officials by people from the AI field, is the narrative she has spent a decade correcting.
When Automation Replaces Human Judgment
Eric Schmidt argued publicly that cameras could replace the “valuable eyes” of soldiers. Bryson objected at the time on two grounds: a single point of failure every attacker would target, and the elimination of whistleblowers. Israel’s Lavender program designated 35,000 people as legitimate targets. A second program identified when each target was home with their family before striking. Whether or not the intent was precise targeting, no one could trace who made the decision, how the criteria were coded, or whether the 35,000 figure had any meaningful basis. That is the accountability gap DevOps exists to close.
Inequality, Economic Fear, and Polarization
Polarization and inequality are now at the levels last seen before World War I. Bryson’s agent-based model, built with colleagues at the political economy level, found the driver is economic decline rather than inequality itself. When people face a discontinuity, bankruptcy, foreclosure, losing a home, the rational move shifts from the higher-payoff out-group to the more predictable in-group. Rising inequality produces that same fear for most people even in a growing economy, because the gains concentrate at the top and median incomes fall. Germany and China are outliers because both have, for historic reasons, kept their floor rising.
Accountability Requires Peers Who Can Say No
Responsibility is a property of moral agents who can hold each other to account. That requires a peer relationship. When one actor becomes powerful enough that the rest of society cannot call them to account, the system breaks. Accountability is the extent to which a society can trace and enforce responsibility, and transparency is the mechanism. Bryson draws the line clearly:
“AI is basically a corporate asset.”
Transparency does not mean open source. Microsoft once arrived in court with a printed 2 million line codebase and called that disclosure. What matters is whether the people who need to check a decision can find what they need to find.
EU Regulation as Software Governance Architecture
The EU’s Digital Markets Act targets monopoly power directly, modeled on what the United States stopped pursuing in the 19th century. The AI Act is lighter than it sounds: it requires documentation only for high-risk systems, those deciding loans, school placements, or other life-altering outcomes. The European product liability directive, updated quietly alongside the AI Act, already classifies all software as a product. Google initially threatened to leave the EU over GDPR. Googlers later admitted it became a single API to 28 countries. China and Brazil both adopted text from the EU AI Act before the EU had finished implementing it.
Bias, Auditability, and the Modular Development Answer
Finnish has no gendered pronouns, but Google Translate used to assign “he” to sentences about investing and “she” to sentences about laundry. Bryson’s 2017 paper showed the bias replicates real-world co-occurrence statistics, not a flaw in the algorithm. The algorithm is working correctly. The fix is not to pretend the training data is neutral. Define what fair means for your use case, make it your test, then build a modular pipeline where every stage is auditable and humans set the criteria. The EU AI Act requires operators to verify the product runs within the parameters they were sold. That is a DevOps question, not a philosophy question.
Notable Quotes
The data programs itself. Joanna Bryson · ▶ 5:29
AI is basically a corporate asset. Joanna Bryson · ▶ 19:58
Yeah, nobody knows what it means. Joanna Bryson · ▶ 4:37
Key Takeaways
- AI systems always have a human who chose the data, algorithm, and release criteria — that person is accountable.
- Economic decline, not inequality alone, triggers polarization; software concentration compounds both risks.
- EU product liability law already covers all software; good DevOps practice satisfies the compliance bar.