Below is a set of non‑consensus hypotheses I’d bet on more than the median expert, but less than certainty. Think of them as decision-relevant guesses—useful if you’re allocating time, money, or attention. Each has: a one‑line thesis, why it might be true, what would change my mind, a confidence band, and what to watch next.
Quick ground rules: These are not medical or investment advice. They’re forward‑looking hypotheses; some will be wrong. I’m aiming for ideas with big consequences if right.
1) Aging is a programmable, multi‑system state, not just wear‑and‑tear.
- Why it might be true: Converging signals from epigenetic “age” clocks, transient cellular reprogramming in mammals, immune/endocrine resets after major physiological events, and parabiosis‑like studies suggest coordinated levers, not just damage accumulation.
- Would change my mind: Human trials that move clocks without functional gains—or gains that always come with high cancer rates.
- Confidence: 0.6
- Watch: Trials showing durable improvements in function (frailty, VO₂max, immune repertoire) alongside age‑clock shifts without elevated oncogenic risk.
2) A large slice of mood and cognitive disorders are immune‑metabolic phenotypes.
- Why: Reproducible links between systemic inflammation and depression, “sickness behavior,” and metabolic comorbidities hint at subtypes poorly served by one‑size‑fits‑all psych labels.
- Would change my mind: Strong negative RCTs where inflammatory biomarkers don’t stratify treatment response.
- Confidence: 0.55
- Watch: Clinical pathways that co‑manage psychiatry, sleep, metabolic health, and gut issues under one roof.
3) The binding constraint on growth is coordination cost, and AI is a coordination technology.
- Why: Most knowledge work is meetings, status checks, synthesis, and compliance—glue code among people. LLM/agent systems are already strong at summarizing, routing, and enforcing process.
- Would change my mind: If agentic workflows plateau at “clever autocomplete” and can’t own entire processes with accountability.
- Confidence: 0.6
- Watch: Orgs that ship more with smaller teams because internal coordination gets automated end‑to‑end.
4) Education gains come from feedback bandwidth, not content—and AI tutoring will scale it.
- Why: The Bloom “2‑sigma” effect (human tutoring) is feedback‑driven. Early AI tutors approximate this at near‑zero marginal cost.
- Would change my mind: If long‑run outcomes (retention, transfer, life earnings) don’t move despite short‑term test boosts.
- Confidence: 0.65
- Watch: District‑scale deployments with randomized rollouts and multi‑year follow‑ups.
5) Personalized nutrition will eclipse general guidelines once the microbiome is properly measured.
- Why: Huge inter‑individual variance in glycemic response, drug side effects, and weight outcomes tracks with microbiome and genetics.
- Would change my mind: If large n‑of‑1 trials show small, unstable effects after personalization.
- Confidence: 0.65
- Watch: Reimbursement for microbiome tests and diet protocols tied to objective endpoints (A1c, weight‑maintenance, symptom remission).
6) Cislunar space beats Mars for the next two decades of real economics.
- Why: Communications, Earth observation, manufacturing precursors, and logistics all pay near Earth; Mars is inspirational but capital‑intensive and low‑ROI near‑term.
- Would change my mind: A clear commodity with Mars‑only economics, or a rapid, cheap transport breakthrough that collapses cycle times.
- Confidence: 0.75
- Watch: Refueling infrastructure in LEO/lunar orbit, high‑cadence cargo, and government procurement for cislunar situational awareness.
7) The biggest near‑term climate win is overbuilding clean power—abundance enables mitigation and adaptation.
- Why: Cheap electrons accelerate electrification (heat pumps, EVs), enable DAC/CCUS where needed, and unlock industrial shifts (green ammonia, steel).
- Would change my mind: Persistent bottlenecks in siting, transmission, or storage that keep wholesale prices high and intermittent.
- Confidence: 0.55
- Watch: Sub‑$20/MWh renewables + storage at scale, heat pump share > gas globally, pragmatic build‑out of long‑distance transmission.
8) Health is logistics: environment and routine, not new molecules, will drive most chronic‑disease gains.
- Why: Indoor air, sleep, movement, light exposure, and diet account for a large risk share, but we manage them poorly. Instrumented buildings and ambient nudges are finally feasible.
- Would change my mind: If tightly managed environments don’t reduce disease burden beyond current care.
- Confidence: 0.6
- Watch: Building codes and employer plans that treat air quality, light, and sleep like safety standards.
9) Agentic software unbundles the firm.
- Why: When process execution, compliance, and integration are automated, companies become orchestration layers over networks of small teams and vendors.
- Would change my mind: If robust agent ecosystems fail to emerge due to trust, liability, or interface brittleness.
- Confidence: 0.6
- Watch: Auditable “process wallets” where agents prove what they did, and insurers underwriting agent‑run workflows.
10) Metascience will shift from publish‑first to replicate‑first.
- Why: Preregistration, code availability, synthetic data checks, and AI reviewers reduce false positives; many marginal literatures will shrink.
- Would change my mind: If incentives don’t flip—tenure and grants still reward novelty over reliability.
- Confidence: 0.55
- Watch: Funders and journals paying for registered reports, replication awards, and negative‑result repositories.
11) GDP understates the AI/software welfare boom; new statistics will track time saved and risk reduced.
- Why: Consumer surplus, latency cuts, and error reduction don’t show up cleanly in output measures. We need household‑level “time accounts” and quality‑adjusted service indexes.
- Would change my mind: If measured productivity soars in lockstep with adoption (i.e., current national accounts already capture it).
- Confidence: 0.6
- Watch: Official experiments with time‑use satellites, hedonic adjustments for digital services, and “risk‑adjusted productivity” metrics.
12) Law becomes machine‑readable: policy as an API.
- Why: Compliance is code; countries will compete on developer‑friendliness (clear schemas, sandboxed trials, auto‑audits). Micro‑juries and continuous update mechanisms will appear first in regulated niches.
- Would change my mind: If legal systems reject machine‑readable norms or liability makes automation uninsurable.
- Confidence: 0.5
- Watch: Governments publishing canonical schemas for taxation, benefits, and reporting; “compliance compilers” as a product category.
13) Math discovery shifts to machine‑generated → human‑interpreted, formally verified pipelines.
- Why: Progress in auto‑formalization and proof search suggests a future where most new results are computer‑found and mechanically checked; humans curate concepts and narratives.
- Would change my mind: Plateaued performance on non‑trivial conjectures or a backlash against black‑box proofs without insight.
- Confidence: 0.6
- Watch: Major theorems first proved by machines then human‑simplified; journals requiring formal proofs for some classes of results.
14) Information hygiene will be regulated like food labels.
- Why: Chronic exposure to adversarial, slot‑machine feeds plausibly carries population‑level cognitive and mental‑health costs.
- Would change my mind: Strong evidence that exposure patterns don’t predict outcomes once confounders are handled.
- Confidence: 0.55
- Watch: Jurisdictions mandating disclosure of recommender objectives, “nutrition labels” for feeds, and user‑controllable ranking.
15) AI safety will look like aviation safety—systems engineering and incentives over grand theory.
- Why: The biggest near‑term risk surface is misuse and integration failures, not sci‑fi autonomy. Tooling (evals, monitoring, containment), liability, and standards reduce incident rates.
- Would change my mind: Emergence of qualitatively new autonomous failure modes that elude systems engineering.
- Confidence: 0.6
- Watch: Industry‑wide incident databases, standardized red‑team regimes, and insurer‑driven safety baselines.
16) Demography is about time, not money: fertility responds most to time cost and coordination friction.
- Why: Delayed pairing, long commutes, low schedule control, and fragmented community matter more than per‑capita GDP once basic needs are met.
- Would change my mind: Policies that increase time sovereignty but fail to move fertility.
- Confidence: 0.55
- Watch: Effects of flexible schedules, dense childcare networks, and pro‑community housing on birth timing rather than lifetime totals.
17) Open‑spec supply chains will eat large chunks of hardware.
- Why: RISC‑V, contract manufacturing, and global QA make it viable to publish designs and compete on integration, service, and brand.
- Would change my mind: If IP enforcement + geopolitics fragment manufacturing beyond interoperability.
- Confidence: 0.5
- Watch: Tier‑1 consumer devices launching on open cores with credible security and support.
18) The first credible signs of non‑human intelligence will come from ecology, not radio astronomy.
- Why: Dense environmental sensing and animal‑communication models may surface algorithmic regularities that don’t fit known biological processes.
- Would change my mind: Years of null results despite massive sensor coverage and better models.
- Confidence: 0.3
- Watch: Compressibility anomalies in bioacoustics or sensor nets that survive adversarial checks.
Two meta‑claims that explain the rest
- S‑curve blindness: We overfit linear trends and underweight phase changes when a coordination or measurement bottleneck breaks. Many “stagnation” narratives are artifacts of the wrong ruler.
- Cultural priors are stronger than we admit: A lot of what looks like “innate” common sense is cultural firmware transmitted by language. As models learn from culture at scale, they’ll mirror (and sometimes correct) those priors.
How to use this (practically)
- Pick 2–3 to pressure‑test. I can help design falsifiable indicators and a lightweight “scorecard” you can track quarterly.
- Decide what you’d do if one is true. Where would you invest, hire, move, or learn differently?
- Ask for the deep dive. If you choose any item, I’ll map the evidence, the best counter‑arguments, and a 12‑month watchlist. If you want, tell me which domains you care about—health, education, energy, governance, space, AI—and I’ll expand those sections into concrete, near‑term bets.