Long Yi
← Back to all writing

AI futures evidence — U3: science & golden age

July 5, 2026

Predictionai-futuresevidencenode-u3

← Futures index · 中文 · Main forecast

Each section: Claim · Why · Evidence · Analogue · Would update if · Conf (H/M/L).

Parent: Utopia timeline §U3
Date: 2026-07-04
Scope: Capability C5–C9; path families B (institutional golden age) and C (radical abundance partial)
Modal baseline (NOT utopia): c8 society snapshot by ci biotech sections
Doom mirror: node2 — CBRN success branch inverted
Physical ceiling: superintelligence physical limits
Bio primer: AI biorisk landscape primer
Epistemic tags: [EST] established · [SPEC] reasoned speculation · [USER GUESS] author not expert in bio/chem — verify primary sources
Format: Claim | Why | Evidence | Analogue | Would update if | Conf

Each section documents one probability or timing claim for Node U3. Probabilities are subjective elicitation unless noted as derived.

TL;DR — Top cruxes

CruxP(crux holds)If true →Utopia bucket
Clinical/regulatory bottleneck binds harder than discovery0.78U1 longevity deferred; U3 still via incremental winsU1 ↓, U3 ↑
AlphaFold-class wins compound ≥2× discovery throughput by C70.55Path B credible; Path C partial onlyU3
First AI-native NDA by 20280.42Salience trigger for science dividend politicsU3, U4
Longevity escape velocity before 20400.12Requires U1 + alignment + clinic breakthroughU1 tail
BMIA/IGSC success enables safe bio acceleration0.55 (mirror N2 MODAL)Dual-use managed; discovery not strangledU3 multiplier

Author note: Bio/chem claims below are sourced from repo research files + 2024–2026 web; flag [USER GUESS] where mechanism exceeds cited evidence.

Path families (B vs C)

P = 0.50 — Path B (institutional golden age) is modal science-dividend channel

Claim: ~50% of U3-relevant futures reach 2–5× welfare via science + governance + distribution without requiring ASI or radical abundance — incremental drug approvals, clinical AI diffusion, materials in niche products, screening-coalition safety.

Why: c8 modal path shows discovery acceleration without pipeline revolution; physical limits + FDA latency dominate; distribution (U2) separate node. Path B = compound marginal wins + institutions absorb shock.

Evidence:

  • (internal note) — biotech sections C5–C9
  • (internal note) — wrong-bottleneck argument
  • (internal note) — U3 bucket definition (~2035–2050)

Analogue: mRNA platform post-COVID — transformative but not instant population healthspan shift.

Would update if: ≥3 AI-native NDAs 2027–2029 + median Phase I–III compressed >30% → Path B upgrades toward Path C.

Conf: M
Buckets: U3 primary; U4 partial

P = 0.18 — Path C (radical abundance partial) materializes science leg by ~2050

Claim: ~18% that AI compresses 50–100 yr biology into ~10 yr (Amodei framing) and clinic/regulatory path shortens enough for population-scale longevity/energy wins — partial U1, not full post-scarcity.

Why: Requires conjunction: C9 internal R&D multiplier + alignment (U1 node) + clinic bottleneck partially broken + physical deployment (U4). Each factor <0.5 alone → product ~0.15–0.22 before correlation discount.

Evidence:

Analogue: Green Revolution — decades compressed yield gains, not overnight abundance.

Would update if: Validated aging surrogate + Phase III win on composite healthspan endpoint before 2032 → revise to ≥0.30.

Conf: L
Buckets: U1 tail; U3 ceiling if distribution fails

P(compounding AlphaFold-class wins) by Ci

P = 0.35 — C5: ≥1.5× effective discovery throughput (modal low)

Claim: By C5 calendar (~2027 H2 – 2028 at 0.70× tracker), AlphaFold3-class structure prediction + literature agents yield ≥1.5× hit-to-lead throughput at top labs — not yet population-visible.

Why: AlphaFold Server + AF3 code release Nov 2024; continuous-learning Agent-2 [SPEC] speeds iteration; wet lab + IND still years. c8 C5: “AI-hit → animal model still years.”

Evidence:

Analogue: CRISPR 2012–2015 — lab revolution, clinic lag.

Would update if: Public benchmark shows ≥2× industry-wide hit rate 2027 → revise to 0.45.

Conf: M
Buckets: U3 (early signal)

Falsifier @C5: Zero top-20 pharma partnerships citing AF3/IsoDDE in SEC filings + no Phase I FPI from AI-native cos in 2027 → compound P ≤0.20.

P = 0.48 — C6: ≥2.5× discovery throughput; first Phase I from AF3-lineage internal

Claim: By C6 (~2028 H1), superhuman coder + automated protocol translation pilots yield ≥2.5× internal discovery loop; Isomorphic or equivalent doses first patient.

Why: Isomorphic “staffing up” for trials mid-2026 [EST]; C6 c8: robotic protocol translation pilot; in vivo still rate-limits.

Evidence:

Analogue: First monoclonal antibody trials — biology clear, manufacturing hard.

Would update if: Isomorphic FPI slips past 2028 → revise C6 compound P down 0.10.

Conf: M
Buckets: U3

Falsifier @C6: Isomorphic publicly delays clinic to 2029+ and Insilico rentosertib Phase II fails primary → “AI discovery” narrative stalls.

P = 0.55 — C7: ≥4× internal target-ID; public lag 12–18 mo

Claim: By C7 (~2028 H2), internal “genius country” proposes novel targets ≥4× faster; public science 12–18 mo behind; Phase I entry still 3–5 yr behind idea.

Why: c8 C7: “Phase I entry still modal 3–5 yr behind idea”; closed-loop design→sim→experiment prioritization [SPEC].

Evidence:

  • (internal note) §C7 biotech
  • (internal note) — C7 unlock “longevity hypothesis gen”

Analogue: Bell Labs internal vs Bell System Technical Journal — publication lag.

Would update if: Leaked internal metrics show 10× loop with Phase I in <18 mo → revise to 0.65.

Conf: L–M
Buckets: U3; U1 hypothesis gen only

Falsifier @C7: METR-style public audit shows frontier lab bio output ≤2× 2024 baseline → internal multiplier oversold.

P = 0.62 — C8: ≥1 AI-native approval or equivalent BLAsub; clinical AI at scale

Claim: By C8 (~2028 H2 – 2029), ≥1 high-profile AI-discovered drug approved or NDA accepted; remote diagnostic AI deployed at scale; compound discovery ≥5× at frontier, ~1.5× industry median.

Why: c8 C8: “first AI-discovered approvals possible 2026–2027 in speculative tail; modal = 1–2 high-profile cases”; trade press 2026–2027 first approval [P].

Evidence:

  • (internal note) §C8 biotech
  • (internal note) — first FDA approval 2026–2027 [P]
  • (internal note) — 12 Phase III AI-adjacent trials

Analogue: First gene therapy approvals — salience without immediate access.

Would update if: No AI-native NDA/BLA by end-2029 → C8 compound claim failed for approval leg.

Conf: M
Buckets: U3 salience trigger; U4 if access broad

Falsifier @C8: FDA explicitly rejects AI-generated evidence package in high-profile CRL without path forward → regulatory choke tightens.

P = 0.68 — C9: ≥10× frontier internal loop; FDA still dominant population bottleneck

Claim: By C9 (~2029), superhuman AI researcher yields ≥10× internal bio/materials R&D; median patient access still gated by 7–12 yr Phase I–III + GMP; rich-only personalized medicine first.

Why: c8 C9: “FDA timeline still dominant modal bottleneck”; “personalized medicine rich-only at first”; 50× multiplier is internal week = external year, not calendar compression of trials.

Evidence:

Analogue: Human Genome Project → first gene drugs — decade from map to medicine.

Would update if: FDA pilot (Accelerated AI Pathway) shows ≥40% Phase III duration cut on ≥2 programs → revise bottleneck P down.

Conf: M
Buckets: U1 internal; U3 population lag

Falsifier @C9: Population healthspan metrics (HALE, age-specific mortality) inflect ≥2σ vs trend in ≥1 OECD country → clinic bottleneck broken faster than model.

AlphaFold & successors

P(status) = 0.95 — AlphaFold3 transforms structural biology baseline [EST]

Claim: AF3 (May 2024 Nature) is confirmed step-change for protein–ligand, antibody, nucleic acid complex prediction — table stakes for serious drug design.

Why: PoseBusters benchmark; 50% accuracy gain vs physics tools cited by DeepMind/Isomorphic; 3M+ researchers used AF DB per Isomorphic 2025 post.

Evidence:

Analogue: BLAST for genomics — infrastructure, not product.

Would update if: Independent replication shows AF3 ligand docking no better than Vina on prospective set → downgrade to incremental.

Conf: H
Buckets: U3 enabler

P = 0.70 — IsoDDE materially exceeds AF3 for drug-design tasks by 2026 [EST]

Claim: Isomorphic Drug Design Engine doubles AF3 on hard PoseBusters / antibody–antigen sets — closes gap between structure prediction and lead optimization.

Why: Isomorphic technical report 2025: 2× AF3 on Runs-N-Poses; 2.3× on antibody–antigen DockQ>0.8.

Evidence:

Analogue: AlphaFold2 → AF3 jump repeated inside one org.

Would update if: Partner pharma (Lilly/Novartis) public write-down of AI-designed assets → IsoDDE efficacy doubt.

Conf: M
Buckets: U3

P = 0.45 — AF3-class tools reduce IND-enabling calendar time ≥25% by 2028

Claim: Structure + generative chemistry shrink preclinical calendar time ≥25% for programs that adopt full stack — not same as success rate.

Why: Trade press: preclinical candidates 13–18 mo vs 3–4 yr [P]; skeptic file: efficacy rate unchanged.

Evidence:

Analogue: CAD for chips — design faster, fab unchanged.

Would update if: Median IND filing interval for AI-native cos matches traditional ~4 yr through 2029 → revise to ≤0.25.

Conf: M
Buckets: U3

P = 0.55 — Protein design (RFdiffusion/Boltz-class) yields ≥1 clinical biologic by 2029 [SPEC]

Claim: De novo protein binders / antibodies from generative models reach Phase I+ with disclosed AI design lineage by 2029.

Why: Generate Biomedicines Phase III GB-0895 (2026); AbCellera AI antibody pipeline; AF3 antibody binding emphasis.

Evidence:

Analogue: Humira lineage — biology validated before AI label mattered.

Would update if: All generative biologic Phase IIIs fail 2026–2028 → revise to 0.30.

Conf: L–M
Buckets: U3

P = 0.25 — Open AF3 weights accelerate global discovery parity by 2028 [SPEC]

Claim: Academic AF3 access prevents full proprietary moat; lower-income labs catch up on structure leg — not trials/GMP.

Why: Nov 2024 academic release vs AlphaFold2 fully open; Evo2 opposite pole (fully open DNA FM) — dual-use tension.

Evidence:

Analogue: TensorFlow open-source — tooling parity, not compute parity.

Would update if: AF3 access re-gated or export-controlled → revise down.

Conf: M
Buckets: U3 (global); U5 governance overlap

AI drug discovery pipeline

P(status) = 0.90 — AI-native programs in clinic are real, not vapor [EST]

Claim: ≥50 interventional trials from AI-discovery companies; Insilico rentosertib Phase IIa PoC (Nature Medicine 2025); BenevolentAI baricitinib approvals.

Why: ClinicalTrials.gov scrape in repo; independent publications.

Evidence:

  • (internal note) — 58 trials table
  • (internal note) — ~173 AI programs globally [P]
  • Insilico INS018_055 / rentosertib NCT05975983

Analogue: Early biotech era — many programs, few winners.

Would update if: >50% Phase II AI-native assets terminated 2026–2027 → survivorship bias confirmed.

Conf: H
Buckets: U3

P = 0.42 — First AI-native NDA/BLA approval by 2028-12

Claim: ~42% cumulative that FDA approves drug where primary discovery claim is AI-designed (Isomorphic, Insilico, or equivalent) by end-2028.

Why: Isomorphic near FPI 2026; Insilico Phase II; trade press first approval 2026–2027; no approval yet as of Jul 2026 [EST]. Median Phase I–III ~7–12 yr caps speed — early approvals likely repurposing-like indications or accelerated paths.

Evidence:

Analogue: Exscientia DSP-1181 first-in-human 2020 — 5+ yr still no approval.

Would update if: First approval in 2026 → revise 2028 cumulative to ≥0.65.

Conf: M
Buckets: U3 trigger; U4 if label broad

P = 0.35 — Phase III AI cohort shows non-inferior efficacy vs industry baseline by 2029

Claim: Pooled Phase III readouts from AI-native discovery (≥5 programs) show efficacy not worse than historical oncology/immunology base rates.

Why: Hype cycle risk — 05_skepticism_limits mandates weighting Phase III; Exscientia/Recursion terminations in pipeline table.

Evidence:

  • (internal note) — terminations listed
  • (internal note) — Argument 2 survivorship
  • (internal note) — 15–20 in Phase III 2026 [P]

Analogue: RNAi field — years of Phase III failure before wins.

Would update if: First Phase III AI-native primary endpoint hit with p<0.05 in 2027 → revise to 0.50.

Conf: L–M
Buckets: U3 crux for Path B→C upgrade

P = 0.60 — Recursion/phenomics model yields platform not pill wins [EST]

Claim: Recursion OS + acquisitions = better target triage and trial design; not majority source of NDAs.

Why: Pipeline mostly company-associated label; phenomics helps search, clinic still attrition-heavy.

Evidence:

  • (internal note) — Recursion rows
  • (internal note) (repo)

Analogue: 23andMe → drug programs — genetics platform, mixed clinic.

Would update if: Recursion NDA from wholly AI-triaged target before 2029 → revise down to 0.40.

Conf: M
Buckets: U3

P = 0.50 — Pharma partnership model dominates over AI-native pharma [EST]

Claim: Novartis/Lilly × Isomorphic pattern — AI at design, pharma at clinic/GMP — remains modal through C9.

Why: Capital intensity of Phase III; FDA credibility easier with established sponsor; Isomorphic licenses post-early trials per Murdoch interviews.

Evidence:

Analogue: CRO industry — specialization persists.

Would update if: Isomorphic files NDA as sole sponsor without partner → model shift.

Conf: M
Buckets: U3

P(bottleneck binds) = 0.82 — Clinical efficacy rate not improved ≥20% by AI through C9 [EST]

Claim: AI shifts where failures happen (fewer bad targets) but Phase II–III LOA still ~10–15% class [USER GUESS] — not oracle.

Why: Jacobson/Nature coverage in skeptic file; biology complexity; patient heterogeneity; Wolfram irreducibility in living systems (superintelligence_physical_limits §5.4).

Evidence:

  • (internal note) — Argument 1
  • (internal note) — reducible vs irreducible
  • Historical pharma LOA ~7–10% industry estimates [EST]

Analogue: Quant finance — better models, fat tails remain.

Would update if: Documented LOA doubling on AI-native portfolio 2028–2032.

Conf: M
Buckets: U1 deferred; U3 still OK

Longevity & escape velocity

P(bottleneck binds) = 0.85 — Longevity modal bottleneck is trials/GMP/FDA, not discovery [EST]

Claim: Through C9, ≥85% of variance in population healthspan timing is clinic/regulatory/endpoints — not target ID or in silico design.

Why: No FDA “aging” indication; epigenetic clocks not validated surrogates (Reagan-Udall 2026 transcript); senolytics mixed Phase II; TAME/HALO timelines measured in years; Amodei explicitly allows clinic latency.

Evidence:

Analogue: Alzheimer’s drug development — targets known decades before approval.

Would update if: FDA grants reasonably-likely surrogate for aging composite (PROSPR/XPRIZE path) before 2028 → revise to ≤0.65.

Conf: M–H
Buckets: Top U3 crux; U1 gate

P = 0.12 — Longevity escape velocity (LEV) for existing cohort before 2040 [SPEC]

Claim: ~12% that therapies compound so calendar aging expectancy gain ≥1 yr/yr for people alive today before 2040.

Why: Kurzweil LEV ~2029–2032 not observed [EST]; de Grey revised human LEV to ~2037 with weak RMR1; requires stack of approvals + biomarker iteration Amodei describes — each gated by trials.

Evidence:

Analogue: Fusion energy — physics progress, engineering deployment lag.

Would update if: TAME or XPRIZE Healthspan trial shows durable multimorbidity reduction with replication → revise to 0.20.

Conf: L
Buckets: U1

P = 0.40 — AI measurably speeds aging biomarker iteration by C7 [SPEC]

Claim: By C7, AI cuts biomarker validation cycles ≥2× (multi-omics, trial simulation, composite endpoint design) — necessary for LEV but not sufficient.

Why: Amodei: “reliable, non-Goodhart-able biomarkers” as key; PROSPR/ARPA-H surrogate hunt; Insilico Longevity Board 2026.

Evidence:

Analogue: PD biomarkers for Parkinson’s — decades to validate.

Would update if: FDA accepts epigenetic clock as secondary endpoint in registrational trial → revise to 0.55.

Conf: L–M
Buckets: U3 enabler; U1 prerequisite

P = 0.55 — Healthspan wins (not lifespan doubling) plausible by 2035–2045 [SPEC]

Claim: Modal positive longevity outcome = compression of morbidity + 5–10 yr healthy life extension in wealthy cohorts — not 150 yr median.

Why: Clarivate 2026 summary via etcjournal; functional endpoints (frailty, gait, multimorbidity) tractable vs lifespan doubling; GLP-1 class precedent for rapid adoption once approved.

Evidence:

Analogue: Statins — population CV mortality down, not LEV.

Would update if: Senolytic or partial reprogramming Phase III hits in defined indication 2028+ → upgrade lifespan tail.

Conf: M
Buckets: U3; U4

P = 0.65 — Amodei “compressed 21st century biology” partially on track for 2028–2033 window [SPEC]

Claim: ~65% that discovery-side metrics (structures, targets, preclinical candidates, trial design) show ≥5× acceleration vs 2015–2020 baseline by 2030 — not full essay outcome list (cancer eliminated, 150 yr).

Why: AF3, IsoDME, GNoME, clinical AI already live; etcjournal mid-2026 assessment: on track for acceleration, not lifespan doubling.

Evidence:

Analogue: Human Genome Project on schedule for sequencing, not for clinical revolution timing.

Would update if: 2030 audit shows <2× discovery metrics → Amodei biology leg failed.

Conf: M
Buckets: U3; conditions U1 tail

Clinical trials, GMP, FDA

P(bottleneck binds) = 0.88 — Phase I–III median 7–12 yr binds through C9 for novel modalities [EST]

Claim: Even with AI trial design, ≥88% probability median novel drug still requires ~7–12 yr clinical development through C9 — absent regulatory revolution.

Why: c8 cross-cut; historical FDA timelines; gerotherapeutics “15 yr $2B” standard; Marchant 2019 — aging trials need decades if lifespan endpoint.

Evidence:

Analogue: Aviation certification — software faster, airframe unchanged.

Would update if: FDA one-trial default + surrogate package cuts median to ≤5 yr on ≥3 NDAs → revise to 0.70.

Conf: M–H
Buckets: U1 gate; U3 pace

P = 0.35 — FDA Accelerated AI Pathway (or equivalent) cuts review time ≥20% by 2029 [SPEC]

Claim: ~35% that interactive AI-drug pilot programs materially shorten review/IND for AI-documented submissions by 2029.

Why: Reg-intel / industry reports 2026 pilot; Jan 2025 draft guidance is credibility not speed; political support for “AI innovation” [SPEC].

Evidence:

Analogue: Breakthrough Therapy designation — real but rare.

Would update if: Pilot announced then cancelled 2027 → revise to ≤0.15.

Conf: L
Buckets: U3; could raise U1 if combined with surrogates

P(bottleneck binds) = 0.75 — GMP / CMC scale-up binds biologics & gene therapy [EST]

Claim: AI-designed biologics/gene therapies hit manufacturing wall ≥75% of time — capacity, viral vector, QC — independent of discovery.

Why: c8 atoms bottleneck; COVID vaccine scale-up analogue; OSK partial reprogramming = gene therapy regulatory class (Regulatory Impact 2026).

Evidence:

Analogue: mRNA manufacturing 2020–2021 — science done, fabs lag.

Would update if: Modular GMP AI-optimized facilities cut biologic IND→lot release <6 mo at scale.

Conf: M
Buckets: U3

P = 0.45 — AI reduces protocol amendment rate ≥30% in oncology trials by 2028 [SPEC]

Claim: Trial simulation + adaptive design tools (AI-native or AI-assisted) cut amendments and screen failures — partial calendar savings inside Phase I–III.

Why: Unlearn, Medidata, Recursion trial OS narratives; does not eliminate enrollment time.

Evidence:

  • (internal note) — NCT05381038 CURATE.AI methods trial
  • Industry AI trial optimization press 2025–2026 [P]

Analogue: EDC adoption 2000s — efficiency inside fixed phases.

Would update if: Meta-analysis shows no amendment reduction on AI-arm trials → revise to 0.20.

Conf: L
Buckets: U3

P(bottleneck binds) = 0.70 — Reproducibility crisis in bench science still wastes ≥20% of AI-generated hypotheses [EST]

Claim: Preclinical replication rates remain poor; AI amplifies throughput of candidates, not quality per candidate, through C8.

Why: c8 cross-cut reproducibility; skeptic file biology complexity.

Evidence:

  • (internal note) — reproducibility row
  • (internal note)

Analogue: ML reproducibility crisis in CS — faster papers, same flake rate.

Would update if: CRO industry adopts mandatory AI+robotics replication standard with published rates.

Conf: M
Buckets: U3

Clinical AI (diagnosis & workflow)

P(status) = 0.95 — FDA-authorized AI devices ≥1,400; radiology-dominated [EST]

Claim: Cumulative FDA AI/ML device authorizations ~1,450+ by late 2025; ~76% radiology; pathology growing 2024–2026.

Why: FDA lists; reg-intel tracker Dec 2025; c8 C4 cites 1,400+.

Evidence:

Analogue: 510(k) digital health wave 2010s — many clearances, few blockbusters.

Would update if: FDA pauses AI 510(k) flood → growth stall.

Conf: H
Buckets: U3, U4 (disease burden ↓)

P = 0.40 — PCCP (adaptive AI device updates) widely adopted by 2028 [EST]

Claim: ~40% new AI device submissions include Predetermined Change Control Plans; enables continuous improvement without full resubmission per change.

Why: Final guidance Dec 3 2024; 10% of 2025 clearances included PCCP per trackers; upward trend.

Evidence:

Analogue: Software OTA in cars — regulated but updatable.

Would update if: Major PCCP-related recall → adoption slows.

Conf: M
Buckets: U3

P(bottleneck binds) = 0.72 — Adoption/reimbursement binds clinical AI benefits through C8 [EST]

Claim: Topol paradox — proven imaging AI not standard of care; ≥72% of mortality benefit potential unrealized through C8 due to workflow, liability, payment.

Why: 07_clinical_ai_now.md; 05_skepticism_limits.md Argument 3; c8 trust erosion.

Evidence:

Analogue: EMR adoption — 20 yr diffusion.

Would update if: CMS national coverage for mammography AI 2027 → revise to 0.55.

Conf: M
Buckets: U3, U4

P = 0.55 — C8 remote AI diagnostic layer deployed at scale (API), separate from FDA device path [SPEC]

Claim: By C8, cheap remote workers provide diagnostic support globally under CDS vs device line — regulatory gray but deployed.

Why: c8 C8 biotech row; HCI master research FDA framing in c8 footnote.

Evidence:

  • (internal note) §C8 biotech
  • (internal note) — LLM clinical mess

Analogue: Telemedicine 2020 — scale before full regulation.

Would update if: FDA/OIG crackdown blocks CDS diagnostic layer → revise to 0.25.

Conf: L–M
Buckets: U4 access; U3 mixed (quality risk)

CBRN screening SUCCESS branch (mirror Node 2)

Invert doom Node 2 MODAL branch: physical-layer screening + state transparency without strangling discovery.

P = 0.55 — MODAL SUCCESS: BMIA/EU screening + RAISE/SB 53 reporting (mirror N2 MODAL)

Claim: ~55% Tier-2 futures: mandatory/patchwork federal screening by ~2028; EU Biotech Act Ch. VIII phased; IGSC v4 function-based at scale; CBRN evals stay voluntary but labs ship mitigations.

Why: Same coalition evidence as Node 2 §MODAL branch — screendna Jun 2026, S.3741, synthesis CEO letter. Success for U3 = bio R&D continues with lower tail risk, not pause.

Evidence:

Analogue: Y2K remediation — expensive, worked, accelerated digital economy.

Would update if: BMIA + EU both fail by 2028 → SUCCESS branch dead; U3 bio tail risk ↑ (link U5).

Conf: M (direction); L (exact P)
Buckets: U3 multiplier; U5 primary

P = 0.25 — SUCCESS+: Near-miss publicized → screening faster without chilling discovery

Claim: ~25% conditional on near-miss Trigger E (Node 2): IGSC catch publicized → P(mandatory screening)>0.75 and industry treats as quality signal, funding to Red Queen–class defense ↑.

Why: Node 2 Trigger E near-miss highest rate 25–35%; Y2K pattern — success looks like overreaction but enables infrastructure.

Evidence:

Analogue: TSA after failed shoe bomb — annoying, not end of aviation.

Would update if: Near-miss → synthesis export bans on reagents → discovery chilled → not SUCCESS+.

Conf: L–M
Buckets: U3

P = 0.15 — SUCCESS++: Screening + function-based SOC covers ≥80% global synthesis volume by 2029

Claim: Tail success — BMIA + EU + IGSC v4 + CN enforcement → ≥80% bp orders screened with function-aware SOC lists.

Why: Node 2 P(IGSC v4 scale)=0.65 by 2028-06; CN P=0.60 enforcement ↑; benchtop gap remains.

Evidence:

Analogue: PCI-DSS global payment security — not 100%, enough.

Would update if: Tier-3 attack despite 80% coverage → SUCCESS++ overstated.

Conf: L
Buckets: U3; enables bolder bio funding

P = 0.10 — SUCCESS failure mode: screening passes but biology FM open weights prevent U3 bio acceleration

Claim: ~10% that governance “success” on screening coincides with Evo-class weight restrictions / liability chill — net slower beneficial bio.

Why: Node 2 dual pathway — (C) screening vs (B) FM gating; P(FM gating)=0.10 US; open-science pushback.

Evidence:

  • (internal note) §P=0.10 biology FM gating
  • (internal note)
  • Evo2 Nature Mar 2026 full open

Analogue: Crypto export controls 1990s — security vs innovation tradeoff.

Would update if: Tier-3 attack traced to open Evo2 → P(gating success) ↑, U3 bio ↓.

Conf: L
Buckets: U3 downside; U5 overlap

Materials, superconductors, fusion

P(status) = 0.90 — GNoME-class computational materials explosion is real [EST]

Claim: DeepMind GNoME (Nature 2023): 2.2M crystals predicted, 380k stable; 736 independently synthesized externally — order-of-magnitude expansion of search space.

Why: Peer-reviewed; follow-on JACS 2024 autonomous synthesis driven by GNoME predictions.

Evidence:

Analogue: AFDB for structures — map, not mine.

Would update if: Large fraction of GNoME “stable” predictions fail replication → downgrade.

Conf: H
Buckets: U3; U1 tail (energy)

P(bottleneck binds) = 0.90 — Lab validation binds materials wins through C9 [EST]

Claim: AI narrows search; crystal growth, bulk properties, certification bind ≥90% of candidates from paper to product through C9.

Why: c8 every Ci: “AI proposes; validation cryostat/magnet lab bound”; superintelligence_physical_limits irreducibility; GNoME 736/380k synthesized ≈0.2%.

Evidence:

  • (internal note) §C5–C9 physics/materials
  • (internal note) §5.4, §9
  • DeepMind blog — 736 experimental confirmations vs 380k stable

Analogue: LK-99 2023 — prediction hype, lab falsification fast.

Would update if: Autonomous labs achieve >10× synthesis throughput with ≥50% yield on GNoME picks 2027–2029 → revise to 0.75.

Conf: M–H
Buckets: U3; U4 deployment

P = 0.35 — Incremental battery/cathode/solid-electrolyte wins in production by 2032 [SPEC]

Claim: ~35% at least one GNoME/AI-predicted electrolyte or cathode in commercial EV/storage cell by 2032 — niche or flagship line.

Why: 528 Li conductors in GNoME vs prior ~21; Samsung/LG cadence; still fab/pilot scale lag.

Evidence:

Analogue: Silicon anode adoption — decades from lab to iPhone.

Would update if: Major OEM announces GNoME-derived chemistry in mass production → revise to 0.50.

Conf: L–M
Buckets: U3; U1 partial

P(bottleneck binds) = 0.92 — Room-temperature ambient-pressure superconductor economically irrelevant through C9 [EST]

Claim: c8 surprise prediction #2 — modal society gets better batteries, not lossless global grid, even at 50× R&D.

Why: LK-99; AI candidate lists ≠ wire production; certification/ grid retrofit decades [SPEC].

Evidence:

  • (internal note) §Three predictions #2
  • (internal note) §C5–C9 superconductors rows
  • Cypris 2025 generative superconductor claims — DFT-validated only [SPEC]

Analogue: High-Tc cuprates 1987 — physics Nobel, not power grid revolution.

Would update if: Replicated ambient RTSC wire with measurable critical current at scale → immediate model falsifier.

Conf: M
Buckets: U1 tail downgraded

P = 0.25 — AI-optimized fusion control contributes to net-energy project before 2035; grid not before 2045 [EST]

Claim: DeepMind tokamak control precedents + C9 sim acceleration → modest P control win; Q>1 engineering still multi-decade modal per c8.

Why: c8 C9 fusion row; physical limits — plasma irreducible turbulence.

Evidence:

  • (internal note) §C9 physics
  • (internal note) §5.4 — turbulence/plasma
  • DeepMind plasma control Nature 2022 [EST]

Analogue: ITER schedule slips — engineering not algorithm bound only.

Would update if: Commercial fusion PPA signed with AI-attributed control stack → revise grid P up.

Conf: L
Buckets: U1 tail; U4

Physical limits crosscut (science dividend ceiling)

P(bottleneck binds) = 0.95 — Superintelligence cannot bypass clinical physics (trials need wall-clock patients) [EST]

Claim: Even at C9, enrolling, treating, and following human subjects requires calendar time — not fully parallelizable like simulation.

Why: superintelligence_physical_limits — CAN predict reducible biology; CANNOT shortcut ethical/statistical requirements for human evidence; Landauer/irreducibility in vivo.

Evidence:

  • (internal note) §1, §7, §9
  • FDA “feel, function, survive” framework — Reagan-Udall 2026

Analogue: Weather prediction vs stopping hurricanes — forecast ≠ control.

Would update if: Accepted surrogate + single-arm pivotal becomes routine for multiple indications → revise to 0.85.

Conf: H
Buckets: U1 ceiling

P(bottleneck binds) = 0.80 — Computational irreducibility limits in silico-only drug approval [EST]

Claim: ≥80% of systemic therapies still require animal/human data — pure simulation insufficient for FDA credibility framework.

Why: Wolfram 2024; Lloyd irreducibility; Jan 2025 FDA AI drug guidance emphasizes context-of-use credibility, not simulation-only.

Evidence:

Analogue: Boeing 737 MAX sim training — sim not enough alone.

Would update if: FDA approves first sim-only NDA (no in vivo) → falsifier.

Conf: M–H
Buckets: U3

P = 0.60 — Logistics (GMP, fabs, GW) dominate deployment of science wins more than discovery at C8+ [EST]

Claim: c8 + physical limits: atoms, energy, manufacturing schedule — not ideas — bind modal utopia.

Why: Cross-cut physical economy in c8; whimper section — humans use AI materials, don’t steer.

Evidence:

  • (internal note) — cross-cutting bottlenecks
  • (internal note) §9

Analogue: Semiconductor shortage 2021 — demand for known science, supply lag.

Would update if: Robotized wet-lab + modular GMP cut discovery→lot <24 mo at scale.

Conf: M
Buckets: U3 vs U1; links U4 node

Bucket mapping summary

Outcome fragmentU1 Radical abundanceU3 Institutional golden ageU4 Modest flourishing
AF3/IsoDDE discovery ≥5×Enabler onlyPrimarySecondary
First AI-native NDA 2027–28SignalPrimary triggerPrimary
Clinical AI 1400+ devicesLowMediumPrimary
Longevity LEV by 2040Primary tailUnlikelyHealthspan partial
GNoME → production batteryTailIncrementalYes
RTSC gridTail fantasyNoNo
BMIA SUCCESSRisk ↓ enables boldnessMultiplierStability
FDA 7–12 yr bindsBlocks U1Modal paceModal pace

Ci falsifiers (consolidated)

CiObservable discriminatorFalsifies U3 compound claim if…
C5Pharma AF3/IsoDDE partnership count; AI-native Phase I startsZero Phase I FPI from AI-native cos in 2027; no screening coalition progress
C6Isomorphic FPI; automated lab notebook pilots publicIsomorphic delays clinic past 2029; rentosertib Ph II fail
C7Public–private lag in structures/targets; leaked 4× internal metricPublic bio output ≤2× baseline; no novel AI-target Ph I entries
C8AI NDA/BLA; remote diagnostic scale; PCCP adoption rateNo AI NDA by 2029; FDA CRL on AI evidence; CMS denies imaging AI NCD
C9HALE/mortality inflection; AI-material in mass product; Phase III AI LOANo population healthspan signal by 2032; all AI Ph III fail; GNoME validation rate <1%

Index

CategorySections
Path families B/C2
P(compounding) by Ci C5–C95
AlphaFold & successors5
AI drug pipeline6
Longevity & escape velocity5
Clinical/GMP/FDA5
Clinical AI deployed4
CBRN SUCCESS (mirror N2)4
Materials/superconductors/fusion5
Physical limits crosscut3
Total evidence sections44

External sources (consolidated)

Update log

DateChange
2026-07-04Initial Phase 2 research pass — 44 sections; C5–C9 compound P; CBRN SUCCESS mirror; bucket map; Ci falsifiers