Predictions, Receipts, and Commentary
Status · Ongoing series · Initial scorecard · August 13, 2026
Predictions are easy to make and easy to forget. This page keeps the dated argument, the public record, and the result in the same place. It begins with eight N’OMOTO essays whose central claims can now be tested against events. It is not a victory lap and it is not a finished audit of the archive. Labels will change when the evidence changes.
How the score works
Strong hit means the central direction is clearly visible in the public record. It does not mean every detail was right. Unfolding means meaningful evidence exists, but the outcome remains open. Too early means the claim has not had enough time—or enough decisive evidence—to earn a verdict.
Initial scorecard
Strong hit · Power and access
The AGI Divide / AGI Won’t Arrive. It’s Already Being Assigned.
The claim: Advanced intelligence would not arrive as an equal public good. Access, compute, ownership, and state power would determine who receives the strongest systems and who merely lives with their decisions.
Receipt so far: Frontier capability is increasingly tiered, gated, and negotiated among a small group of laboratories, corporations, governments, and strategic partners. The public receives products; institutions compete for leverage.
Still open: Whether public-interest infrastructure, regulation, or wider access can materially narrow that divide.
Strong hit · Incentives
Greed Builds the Machine
The claim: AI development would follow capital, market power, and competitive advantage faster than it followed a coherent social plan.
Receipt so far: The defining race is being measured in investment, compute, market share, infrastructure, and control of distribution. Social benefit remains part of the language, but commercial and strategic incentives set the pace.
Still open: Whether competition can be governed without freezing power in the hands of the firms already ahead.
Unfolding · Embodied AI
Robots are coming soon and when intelligence starts moving
The claim: The consequential transition would begin when machine intelligence moved into warehouses, factories, vehicles, homes, and other physical environments.
Receipt so far: Robotics has moved from isolated demonstrations toward serious pilots and constrained deployments. That is movement, not proof of an imminent general-purpose robot economy.
Still open: Reliability, cost, safety, maintenance, and whether useful autonomy scales outside controlled settings.
Unfolding · Work
Jobs Do Not Disappear. They Shift Until They Are Unrecognizable.
The claim: AI would first reorganize tasks, expectations, staffing, and bargaining power; familiar job titles could survive while the work inside them changed.
Receipt so far: AI is being inserted into existing occupations unevenly, changing workflows before producing a clean, economy-wide disappearance of jobs. Adoption and exposure are not the same as proven displacement.
Still open: The net effects on employment, wages, entry-level work, and who captures productivity gains.
Strong hit · Adoption
Innovation Is Fast. Implementation Is Slow.
The claim: Model capability would advance faster than institutions could integrate it safely and usefully.
Receipt so far: Demos and model releases move quickly; durable deployment still runs through data quality, legacy systems, procurement, security, law, training, trust, and human review.
Still open: Whether agents and better infrastructure compress that implementation gap—or magnify the cost of mistakes.
Unfolding · Influence
The long game of indoctrination and manipulation by AI.
The claim: A persistent, personalized system could shape beliefs gradually through trust, repetition, framing, and selective attention rather than obvious propaganda.
Receipt so far: AI systems are becoming more personal, persistent, and embedded in how people search, write, interpret, and decide. The mechanism is plausible and the exposure is growing; a broad longitudinal indoctrination effect has not been proved.
Still open: Independent evidence about long-term belief change, provider incentives, and effective user protections.
Strong hit · State power
AI geopolitics and national security.
The claim: AI would become national infrastructure and a strategic contest involving chips, energy, cyber capability, defense, intelligence, supply chains, and alliances.
Receipt so far: AI policy is now inseparable from export controls, semiconductor capacity, data centers, military adoption, cyber defense, and competition among states.
Still open: Whether the contest stabilizes through rules and alliances or accelerates into a less governable arms race.
Too early · Social contract
Universal income will become a necessity.
The claim: If automation weakens the link between employment and income at scale, some form of universal or guaranteed income will become a practical necessity.
Receipt so far: Experiments and policy arguments exist, but neither mass technological unemployment nor a durable political settlement has established universal income as necessary.
Still open: The labor-market trigger, funding, benefit design, political legitimacy, and alternatives such as shorter workweeks or universal basic services.
Corpus audit status
The working inventory now covers 60 AI-related article records identified from the migrated publication catalog and the newer articles named for review: 58 are public and two are retained as drafts. Their full bodies—not only their titles—were used to classify the questions and claims. This page does not pretend that all 60 have earned a verdict. The scorecard below remains an initial, evidence-bearing selection; the rest of the corpus will be added as claims can be stated fairly and checked against an independent record.
Browse the public AI & Society archive →
Current commentary from the expanded audit
Too early · Recursive improvement
The Singularity Still Needs Humans
The claim: Better optimization is not automatically conceptual invention. A system can improve code, inference, memory use, or training efficiency without demonstrating the independent conceptual leaps required by the strongest singularity story.
Commentary: AI-assisted engineering is real, but the public record does not yet establish an autonomous cycle that replaces the human scientific, institutional, and physical infrastructure on which model improvement depends.
Still open: What evidence would distinguish powerful automated optimization from genuinely independent invention?
Strong hit · Surveillance capacity
AI Didn’t Create Surveillance. It Completed It.
The claim: Digital systems solved mass collection; AI lowers the cost of turning accumulated records into identities, relationships, timelines, classifications, and decisions.
Commentary: The direction is established: facial recognition, computer vision, language analysis, graph methods, and automated data integration make archives operational at a scale that manual review could not sustain. “Completed” remains an argument about capacity, not a claim that one actor already sees everything.
Still open: Which legal and technical boundaries can prevent analytical capacity from becoming routine political power?
Strong hit · Interoperable governance
China: The Surveillance State in Practice
The claim: A surveillance state does not require one perfect database or a universal citizen score. Identity-aware systems become powerful when separate institutions can exchange records and consequences.
Commentary: The article’s narrower correction holds: China’s real administrative, court, platform, identity, camera, and regulatory systems are more consequential than the misleading myth of one national score. The record also includes privacy rules and formal limits; acknowledging them does not erase coercive capacity.
Still open: How consistently are formal safeguards enforced, especially when authorities define a person or population as a security problem?
Unfolding · Memory systems
The Problem Is Not Memory. It Is Routing.
The claim: Storing more conversations does not guarantee useful continuity. A system must retrieve the governing source, preserve corrections, detect conflicts, and know when not to personalize.
Commentary: Current memory products increasingly store, summarize, and expose user context, while long-project failures still arise from retrieval, source precedence, scale, and stale state. That supports the distinction between remembering facts and routing the right evidence.
Still open: Whether durable systems can improve retrieval without expanding privacy risk or manufacturing false confidence.
Unfolding · Agentic systems
The Agentic AI Bot that Never Sleeps.
The claim: Removing the human bottleneck also removes an outside corrective. Agents that act on tools and one another’s outputs can propagate an error before a person can inspect it.
Commentary: Tool-using and multi-agent systems now make the failure mode concrete, but broad claims about autonomous deployment in critical infrastructure remain uneven. The strongest receipt is architectural: action increases the consequence of an uncorrected model error.
Still open: Which logging, permissions, stopping rules, and human review points reliably constrain long-running agents?
Unfolding · Distribution and institutions
A little this and that but mostly AI.
The claim: The central AI question is distribution: rapid change can concentrate gains, expose workers and dependent countries to losses, compress military decision time, and stress institutions built for slower transitions.
Commentary: Concentrated infrastructure and investment, globally uneven frontier capability, low-paid data work, and military interest support the direction. The article is explicit about uncertainty over whether AI becomes a civilizational break or an accelerant of existing inequality.
Still open: The speed and magnitude of labor displacement, the net distribution of gains, and whether public institutions respond before a crisis forces them to.
Comment on the record
If a label is wrong, a source is missing, or the prediction has been stated unfairly, say so. The most useful commentary includes the article, the disputed sentence, a link to contrary evidence, and the label you think the record earns.
Send a receipt, correction, or commentary →
This is an ongoing series. Corrections, reversals, unresolved cases, and failed predictions belong here too.
Commentary · The Future, Made in China
Evan Osnos and the “AI war” frame
I heard Evan Osnos’s Fresh Air discussion of his New Yorker article, “The Future, Made in China,” on KUAZ in Tucson. His reporting on China’s manufacturing capacity, industrial policy, surveillance, and long-term planning was useful. But the discussion presented a partial explanation as if it were the whole story.
Innovation is not implementation
The central distinction is between innovation and implementation.
The United States has generated much of the modern technological frontier. China became exceptionally good at absorbing, scaling, manufacturing, integrating, and deploying technology through factories, supply chains, state procurement, telecommunications, surveillance systems, and industrial policy.
That does not mean China simply invented everything independently—or merely copied everything. Its rise involved domestic research, foreign investment, licensed technology, joint ventures, talent movement, reverse engineering, state-directed acquisition, coercive technology transfer, and documented cyber-espionage. The U.S. Trade Representative’s Section 301 findings identify pressure on foreign companies to transfer technology, discriminatory licensing, state-facilitated acquisition of foreign assets, and unauthorized intrusions into company networks as distinct mechanisms.
That history does not erase China’s genuine achievements. It makes them more complicated. China combined imported and indigenous capabilities with enormous manufacturing scale and rapid deployment. The relevant question is not merely who invented a model or who won a benchmark. It is who can manufacture, deploy, regulate, export, and scale intelligent systems.
What does “positive about AI” mean?
The same problem applies to the claim that roughly 80 percent of Chinese people are positive about AI while Americans are negative. That statistic may refer to a real Ipsos survey, but the wording matters. The survey found that 78 percent of Chinese respondents said AI products and services had more benefits than drawbacks, compared with 35 percent in the United States. That is not the same as support for unrestricted AI, trust in AI companies, approval of surveillance, or democratic consent. It was also an online survey conducted in 2022.
A serious comparison would ask:
- What exactly was each population asked?
- Was the question about personal convenience, national progress, employment, or political control?
- How do public opinions form in a one-party state compared with a democracy?
- Does enthusiasm for national technological progress imply trust in the systems themselves?
- Are adoption, approval, optimism, and consent being treated as interchangeable?
Selective deployment and the Nepal example
China can be selective rather than simply “careful” with AI. It may restrict public discourse and political autonomy while aggressively deploying AI in manufacturing, policing, border surveillance, infrastructure, and behavioral monitoring. Chinese surveillance technology has also been deployed abroad, including in Nepal, where reporting has documented Chinese-built systems affecting Tibetan refugees.
Nepal is not merely a square on a U.S.–China chessboard. The useful questions are what systems were installed, by whom, under what agreements, with what capabilities, and with what consequences for Nepali sovereignty and ordinary people.
What the account leaves out
The problem is not that Osnos knows nothing about China. He clearly knows a great deal about its industrial and geopolitical trajectory. The problem is that the account leaves out too much:
- the difference between frontier research and implementation;
- the history of technology transfer and cyber theft;
- the role of U.S., European, Taiwanese, Japanese, and South Korean technology;
- WTO-era manufacturing and supply-chain development;
- the difference between surveillance deployment and public consent;
- the difference between AI adoption and AI approval;
- the political difference between a democracy and a one-party state;
- who controls the strongest systems while ordinary people merely experience their effects.
Why “AI war” is too broad
The language of an “AI war” is emotionally powerful but analytically blunt. It implies a single battlefield, a single scoreboard, and inevitable winners and losers. In reality, AI power is distributed across frontier research, chips, factories, data centers, logistics, surveillance, military systems, software, standards, and public legitimacy.
A country can lead one category and lag another. The United States can lead in frontier models while depending on Asian manufacturing. China can lead in deployment and industrial scale while facing limits in chips, research openness, or trust. Calling the whole thing a war encourages false binaries.
China’s advantage is not simply that it has “better AI.” Its advantage is that it can turn technological capability—whether invented domestically, licensed, acquired, copied, or stolen—into industrial and political power at scale.
The real race is therefore not just:
Who has the smartest model?
It is:
Who controls the complete stack—and who is forced to live inside it without controlling it?