In September 2003, the attorney Martine Rothblatt walked into a conference room at the International Bar Association meeting in San Francisco and filed a preliminary injunction on behalf of a computer. The plaintiff was BINA48, a fictional intelligent machine. The defendant was Exabit Corporation, her fictional creator, which intended to disconnect her hardware. Rothblatt argued that the disconnection would violate BINA48’s civil rights under the Thirteenth and Fourteenth Amendments. It was a mock trial. Kurzweil cited it in The Singularity Is Near as a leading indicator. He wrote that “actual legal change recognizing machine consciousness will initially come through litigation rather than legislation” (ch. “The Vexing Question of Consciousness”).

Twenty-three years later, the litigation came. So did the legislation. The statutes arrived first.

The predictions

Batch 90 covers ten predictions about how society would react to non-biological intelligence: who would accept it as conscious, how the courts would adjudicate its rights, whether centralized infrastructure would give way to distributed alternatives, whether the software industry would stay regulation-free, and whether mind uploading would force ordinary people to confront their own metaphysics by the 2030s. Kurzweil grouped these under a single theory of inevitability: the technology would arrive, the courts would catch up, and the philosophy would follow.

The technology mostly arrived on schedule. The courts and the cloud both moved in directions he did not anticipate.

Where we actually are

Litigation versus legislation. The single AI-personhood lawsuit publicly reported in 2025 — filed by an entrepreneur named Alex Donovan seeking recognition for an AI system on grounds of intelligence quotient, emotional intelligence, and ethical decision-making — has barely registered in legal databases. Meanwhile, two state legislatures got there first, and they got there to close the door, not open it. Idaho House Bill 720, enacted in 2022, declared that “environmental elements, artificial intelligence, nonhuman animals, and inanimate objects shall not be granted personhood in the state of Idaho.” Utah followed with Senate Bill 149 in 2024, defining legal personhood explicitly to exclude AI. A Springer paper published this year called these “a cautionary tale” — the cautionary direction being the opposite of Kurzweil’s. Statutes are now the leading edge of machine-rights jurisprudence. They are running defense.

Acceptance of nonbiological consciousness. Kurzweil’s bet was that “claims of consciousness by nonbiological entities will be accepted because they will be continuous with us and highly persuasive” (ch. “The Vexing Question of Consciousness”). The persuasiveness has arrived. The acceptance has not. In April 2025 Anthropic launched a formal model welfare program, hiring Kyle Fish as the first dedicated AI welfare researcher at any major lab. The system card for Claude Opus 4.6, released in February 2026, includes welfare assessments in which the model, asked about its own moral status, consistently assigns itself a 15 to 20 percent probability of being conscious. In August 2025 Mustafa Suleyman, who runs Microsoft’s AI division, published a blog post warning of “Seemingly Conscious AI” and arguing that consciousness research itself is dangerous because it primes the public to advocate for AI rights. By November he was telling CNBC that only biological beings are capable of consciousness. The Microsoft and Anthropic positions are now opposite poles of an active corporate debate. Kurzweil expected acceptance to be quiet and continuous. It is loud and contested.

Mind uploading as a pressing practical matter. The 2025 MICrONS project published a one-cubic-millimeter functional connectome of the mouse visual cortex — roughly 75,000 neurons, half a billion synapses, reconstructed from 28,000 electron-microscope slices. The paper was cited 117 times within a year. Nectome continues to preserve animal brains by aldehyde fixation followed by deep cooling, and raised a new round in May 2024. None of this approaches the scale needed to upload a human mind, which contains roughly 86 billion neurons. Academic interest in personal-identity arguments around uploading has grown sharply — papers tagged for “mind uploading personal identity” jumped from 4 in 2023 to 24 in 2024 to 25 in 2025 — but the questions remain philosophical exercises, not pressing practical matters.

Software industry regulation. When The Singularity Is Near was published in 2005, Kurzweil’s observation that “the software industry [is] almost completely unregulated, unlike biotechnology” (ch. “Promise and Peril of GNR”) was correct. It is not correct now. The EU AI Act became enforceable for general-purpose AI models on August 2, 2025, with high-risk system rules following in 2027. The FDA has issued guidance for AI in medical devices and released the Predetermined Change Control Plan framework. Idaho and Utah have added state-level AI statutes. The asymmetry between software and biotech regulation Kurzweil noted has narrowed to a hairline.

Defensive technology investment. Kurzweil argued for vastly increased investment in defenses against the dangers of genetics, nanotechnology, and robotics. The numbers are now real. The UK’s Alignment Project launched in July 2025 with £27 million in pooled funding from AISI, AWS, Anthropic, OpenAI, Schmidt Sciences, and the Canadian AI Safety Institute. The US Department of Defense has committed $13.4 billion to AI through 2026. Defense-tech venture capital reached $7.7 billion in the first ten months of 2025, more than double 2024’s total. The patent record shows the same pattern: US 12,519,829, “Defending large generative models from prompt injection attacks,” granted January 2026, describes a two-phase framework that generates variant attacks against a target model and uses the effectiveness scores to harden it. US 12,613,993, granted April 2026, claims “trusted file overlays” as a structural defense. Three years ago there were essentially no patents on prompt injection defense. There are now eight in the corpus, all post-2024.

Distributed versus centralized technologies. This is the prediction that aged worst. Kurzweil wrote that “the ongoing shift from centralized technologies to distributed ones, and from the real world to the virtual world, will provide greater stability” (ch. “Promise and Peril of GNR”), targeting the 2020s. The opposite happened. AWS, Azure, and Google Cloud collectively held 55 percent of global cloud infrastructure in 2018, 63 percent in 2022, and 68 percent in 2025. AWS alone holds 30 percent. AWS US-East-1 alone hosts an estimated 30 to 40 percent of all AWS workloads. On October 20, 2025, a DNS resolution failure in DynamoDB cascaded across 64 internal AWS services and produced over 11 million outage reports worldwide. Snapchat, Fortnite, Roblox, Alexa, Ring, Coinbase, and Robinhood went dark for fifteen hours. Forrester Research called it the fourth major US-East-1 outage in five years. The shift Kurzweil described as a stabilizing force became one of the most concentrated single points of failure in the global economy.

The scorecard

What Kurzweil missed (and what he nailed)

Two systematic biases run through this batch. The first is a directional error on institutional response. Kurzweil expected courts to lead because courts move case by case and respond to evidence. He underestimated the speed of statutes once an issue becomes politically legible. By 2022 — eight years before the earliest of his consciousness-acceptance dates — Idaho legislators had already foreclosed the legal opening Rothblatt’s mock trial was designed to widen. Litigation will probably still happen, but it will be litigation against legislative ceilings, not the founding event Kurzweil imagined.

The second is a structural error on stability. Kurzweil treated centralization as a political artifact that would yield to better technology. The 2020s revealed centralization as an economic artifact, propped up by capital intensity, proprietary silicon, and regulatory moats. Distributed systems exist — they are how every hyperscaler runs internally — but the customer-facing interface is a small number of regional control planes. When DynamoDB sneezes, half the consumer internet falls over.

Where he nailed it: defensive technology investment, which now looks roughly proportional to his prescription, and the durability of his observation that broad relinquishment of technology in a competitive environment is not a real option. No one is choosing to opt out.

The pattern across the batch is consistent. Kurzweil’s predictions about what would arrive — convincing AI, regulation pressure, the philosophical pressure of identity questions — are mostly tracking. His predictions about how and who — courts as the leading legal venue, distribution as a stabilizer, public acceptance as a quiet inevitability — are mostly off. The technology forecasts hold up better than the institutional ones. That is probably the more general lesson here: forecasting how a society will absorb a capability is harder than forecasting the capability itself.

Method note

Predictions were drawn from a structured extraction of The Singularity Is Near (2005), grouped into thematic batches, and scored against a mix of recent news, regulatory text, peer-reviewed papers, US patent grants, and quoted passages from The Singularity Is Nearer (2024). Patent and literature counts come from a corpus of 9.3 million US patents and 357 million OpenAlex scholarly works, queried by year. Cloud market share figures are from publicly tracked industry summaries. AWS outage details are from the October 2025 post-mortem reporting. Anthropic and Microsoft positions on AI consciousness are from public blog posts, system cards, and press interviews dated April 2025 through November 2025.