Kurzweil Scorecard: The Medicine That Was “Already Happening” in 2005

Most Kurzweil predictions point to a future date. This batch is different. Across Peering into the Brain, Gene Chips, Somatic Gene Therapy, and Reversing Degenerative Disease, Kurzweil listed things he claimed were already true in 2005 — specific techniques, specific labs, specific companies, one personal health experiment. These are the easiest predictions to score, because the clock ran out the day the book went to press.

Twenty-one years later, some look prescient. A few look like the author reading a press release and mistaking it for a platform shift. And one — involving a company on whose board he sits — has had a harder road than the book implied.

The predictions

Kurzweil wrote that “transcranial magnetic stimulation can stimulate or virtually lesion small brain regions, diminishing or enhancing skills and even inducing mystical experiences” (ch. “Peering into the Brain”). He pointed to the Fast ForWord program as proof that plasticity could be harnessed in software, citing John Gabrieli’s fMRI work showing the intervention “enlarged and activated the left prefrontal language region” (ch. “Brain Plasticity”). He described misfolded-protein accumulation as the root of Alzheimer’s (ch. “Life’s Computer”). He reported that gene-expression profiling could classify leukemias and had already corrected a misdiagnosis; that gene chips were replacing animal toxicity tests; that liposomes were working as a “molecular Trojan horse” for brain-cell gene delivery; that Andrew H. Baker’s Glasgow lab had solved targeted adenoviral gene therapy for endothelium (ch. “Gene Chips” and “Somatic Gene Therapy”); and that United Therapeutics had begun 2005 human trials of autologous engineered stem cells for pulmonary hypertension. He also claimed he had reversed his own type 2 diabetes and that degenerative disease accounted for “about 90 percent of deaths in society” (ch. “Reversing Degenerative Disease”).

Each is a testable statement. Let’s test them.

Where we actually are

TMS for depression went from laboratory to FDA-cleared standard of care. Magnus Medical’s SAINT protocol, FDA-cleared in 2022, uses an 8-minute resting-state fMRI to locate an individualized dorsolateral prefrontal target, then delivers fifty sessions of intermittent theta-burst stimulation over five days. In the pivotal study, 19 of 21 treatment-resistant depression patients (90.5%) met remission criteria immediately after the course. US 12,115,383, granted October 2024, describes a closed-loop TMS system that measures evoked responses and re-tunes the protocol per patient — exactly the individualized “virtual lesioning” Kurzweil described. The clinical-trials registry carries 641 completed TMS studies and 281 actively recruiting.

The mystical experiences sub-claim is a different story. Kurzweil was referring to Michael Persinger’s “God helmet” experiments. A 2005 Swedish replication (Granqvist et al.) blinded the helmet and found personality traits, not the magnetic fields, explained the reports. Two decades of sham-controlled work have not rescued it. TMS can change mood, cravings, and motor cortex excitability. It cannot reliably produce religious states.

Fast ForWord did not turn out to be what Kurzweil thought it was. The auditory-timing theory of dyslexia that underpinned the program lost its scientific argument to the phonological-deficit model. Strong, Torgerson and colleagues published a systematic meta-analysis in Journal of Child Psychology and Psychiatry in 2011 that pooled randomized data and concluded there was no significant effect of Fast ForWord on any reading or language outcome versus active or untreated controls. A 2013 critical review reached the same conclusion. Gabrieli’s fMRI signal was real; the causal arrow from that one software product to reading gains was not.

The amyloid hypothesis finally produced a drug — and it targets exactly the species Kurzweil named. Lecanemab (Leqembi, BAN2401), which received full FDA approval in July 2023, is a monoclonal antibody that selectively binds soluble amyloid-beta protofibrils — the intermediate-aggregate species Kurzweil singled out in 2005. Donanemab followed in July 2024. Both slow cognitive decline modestly (roughly 27% and 35% at 18 months) and both carry real amyloid-related imaging abnormality (ARIA) risks. Kurzweil was right about the target. He was wrong about the path — the therapy that worked was an extracellular antibody clearing aggregates, not a fix to the intracellular ubiquitin-proteasome disposal system he emphasized.

Gene-expression profiling is now how leukemia gets classified. The 2022 WHO classification of myeloid neoplasms and the 2022 International Consensus Classification both define AML subtypes primarily by molecular and expression signatures rather than morphology. US 11,111,542, granted September 2021, claims a method of prognosing AML using expression levels of three genes drawn from a panel including DNMT3B, ZBTB46, NYNRIN, LAPTM4B, CD34, and GPR56, compared against reference cohorts. This is the exact thing Kurzweil said had “already demonstrated feasibility” — except in 2005 it was a research demonstration, and now it is a prognostic product with granted claims.

Toxicogenomics is real, but it’s sharing the stage, not owning it. US 10,665,323, granted May 2020, describes a “Predictive Toxicogenomics Space” score — a compound-induced transcriptional fingerprint used to predict dose-dependent toxicity of pharmaceuticals, cosmetics, and agrochemicals in vitro. FDA Modernization Act 2.0, signed December 2022, explicitly authorized non-animal methods (organ-on-chip, in silico simulation, cell-based assays) for IND-enabling toxicology. Animal testing has not been abolished; it has been reduced. The winning alternative looks more like a combination of organoids and AI simulation than a 2005-era expression microarray.

Liposomes-for-brain-delivery happened — but AAV won the gene-therapy race. US 12,178,885, granted December 2024, describes apolipoprotein-decorated liposomes that form a plasma protein corona, dock on vascular endothelial cells, and ferry payloads across the blood-brain barrier — exactly the Trojan-horse architecture Kurzweil described. In practice, the approved CNS gene therapies (Luxturna, Zolgensma, the newer Rett programs) use adeno-associated viral vectors, not liposomes. And the biggest liposomal-delivery story of the past five years was mRNA vaccines, not brain gene therapy.

Targeted adenovirus was a promising platform that clinical reality rerouted. Andrew Baker’s group (now at Edinburgh) produced foundational work on capsid-modified adenoviruses for endothelial targeting. But systemic adenoviral gene therapy was sidelined after the 1999 Jesse Gelsinger death and the field’s pivot to AAV. Adenovirus’s biggest modern win was as a vaccine chassis — the Oxford-AstraZeneca ChAdOx vector for COVID-19 — not the targeted organ-specific gene-therapy platform Kurzweil anticipated.

United Therapeutics’ stem-cell PAH trial took a rough path. The Phase 1 safety study (NCT00469027) of autologous progenitor cells transfected with human endothelial nitric oxide synthase completed. The follow-up Phase 2/3 SAPPHIRE trial (NCT03001414) was terminated. UT’s commercial PAH franchise rests on inhaled treprostinil (Tyvaso) and tadalafil, not the engineered-stem-cell approach. In The Singularity Is Nearer (2024), Kurzweil shifts his UT example to 3D-printed organ scaffolds seeded with reprogrammed stem cells: “United Therapeutics (a company for which I am a board member) is applying this approach … to someday grow entire lungs, kidneys, and hearts.” A different technology on a different timeline.

Kurzweil’s personal diabetes claim is individually unverifiable and collectively validated. He reported reversing his own type 2 diabetes via a self-designed protocol. We can’t audit one person’s HbA1c history. What we can say: the DiRECT trial, in its 5-year extension published in Lancet Diabetes & Endocrinology in 2024, showed that among original intervention participants in remission at year 2, 26% remained in remission at five years. The principle he operationalized in the 1990s is now randomized evidence.

“90 percent of deaths” is too high. CDC 2024 provisional mortality: heart disease (683k), cancer (620k), stroke (167k), chronic lower respiratory (146k), Alzheimer’s (116k), diabetes (94k). The ten leading causes together accounted for roughly 71% of 2023 deaths. Chronic degenerative disease dominates mortality — direction right — but “90%” overstates the share unless the category is stretched to cover almost every non-accident death.

The scorecard

What Kurzweil missed (and what he nailed)

The pattern here is consistent and useful beyond this batch: Kurzweil was unusually good at identifying the targets that would matter, and unusually willing to over-index on the platform in front of him at the moment of writing.

He named amyloid protofibrils as the thing causing Alzheimer’s eighteen years before the FDA approved a drug that binds them. He named expression profiling as the future of leukemia diagnosis two decades before it became the axis of the WHO classification. He named TMS as an individualized neuromodulation tool a decade before Stanford published SAINT. Those are hits.

Where he misses, the failure mode is nearly always the same: he picks the specific company, product, or vector that was in the news and treats it as the winning implementation. Fast ForWord, liposomes for brain gene therapy, Baker-style adenoviral targeting, engineered endothelial progenitor cells for PAH — all real science, all superseded by a different implementation of the same idea. The gene-to-CNS concept won; liposomes lost to AAV. The TMS-for-brain-modulation concept won; the God-helmet sub-story did not.

Read for direction, the 2005 medicine chapter holds up better than critics admit. Read for specific companies, it is nearly the opposite. Twenty-one years is long enough for the right mechanism to arrive wearing a different name.

Method note

We scored each claim by pulling a landscape from 9.3 million US patents and a ~357-million-work OpenAlex scholarly snapshot, filtered to high-citation material in the relevant timeframe. For each prediction we also read the full claims and abstracts of the most relevant granted patents, pulled matching clinical-trial records, and cross-checked against FDA, CDC, and peer-reviewed outcome data retrieved during this session. Patent numbers, study identifiers, and outcome statistics come from primary sources or their direct summaries.