Independent adversarial cross-check. Written from scratch, using only the permitted context file for the user's "AI-freeze" branch and current web sources. Current date: 2026-05-31.
If you condition on transformative AGI / ASI arriving this decade, roughly 2027-2032, the biohacking landscape changes shape. The main bottleneck stops being "can humans grind through connectomics, organogenesis, and VR neuroscience by 2037?" and becomes:
My honest range: P(transformative AI by end-2030) = 20-45%, with a central estimate around 30%. That is high enough that analyses of uploading, radical bioengineering, and experience machines should explicitly condition on it. I would not treat 2030 ASI as the median future, but I would treat it as a live branch that dominates many transhumanist decision calculations.
Under aligned ASI by ~2030, the three end-goals collapse:
The dark side is equally central: unaligned ASI does not merely delay the goals; it can make them impossible or worse than death. Friendship is Optimal is not just wish fulfillment; it is a warning about a system that "succeeds" by its own objective while coercively rewriting the world.
I will use:
"Transformative AI" here means AI capable of causing rapid, economy- and science-scale transformation, especially via strong autonomous R&D. It does not require every human task to be automated in every physical context. "ASI" means clearly superhuman strategic and scientific capability.
DEMONSTRATED: Frontier AI systems have improved rapidly on autonomous task performance. METR's 2025 time-horizon work found a smooth trend in which frontier models' task-completion horizon roughly doubled every ~7 months from 2019-2025; METR's 2026 update says the metric is useful but has important limitations and should not be naively converted into "researcher automation date."
Sources: https://metr.org/blog/2026-1-29-time-horizon-1-1/ and https://metr.org/notes/2026-01-22-time-horizon-limitations/
DEMONSTRATED: Major labs plan around very powerful AI as a near-term governance problem, not as distant sci-fi. OpenAI wrote in 2023 that it was "conceivable" that within ten years AI systems could exceed expert skill in most domains and carry out productive activity comparable to a large corporation. Google DeepMind's Frontier Safety Framework explicitly monitors autonomy, biosecurity, cybersecurity, and ML R&D capabilities. Anthropic's Responsible Scaling Policy defines AI R&D thresholds around automating entry-level remote Anthropic research work or dramatically accelerating scaling.
Sources: https://openai.com/blog/governance-of-superintelligence, https://deepmind.google/discover/blog/introducing-the-frontier-safety-framework/, https://www.anthropic.com/index/anthropics-responsible-scaling-policy
DEMONSTRATED: Expert and forecasting distributions have moved earlier, but remain very wide. The 2023 AI Impacts survey of 2,778 AI researchers found a broad distribution and substantial uncertainty; TIME's summary reports an aggregate 10% chance of high-level machine intelligence by 2027 and a 50% date around 2047, while noting Epoch-style models and superforecasters differ sharply. Epoch AI argues scaling is likely to continue through 2030 and could produce transformative capabilities across science.
Sources: https://arxiv.org/abs/2401.02843, https://time.com/6556168/when-ai-outsmart-humans/, https://epoch.ai/publications/what-will-ai-look-like-in-2030
SPECULATIVE: "AI 2027"-style scenarios are not evidence that AGI will arrive in 2027; they are operational forecasts/scenarios. Their value is that they model a coherent short-timeline branch: autonomous AI R&D, intelligence explosion dynamics, lab race pressure, and governance failure modes.
Source: https://ai-2027.com/
My range for transformative AI by end-2030 is 20-45%, central ~30%.
Why not lower?
Why not higher?
HYPE FLAG: Treating any single extrapolated chart as "AGI by date X" is overconfident. The correct posture is not certainty; it is that short-timeline AGI is a live branch with very high decision value.
Without near-term ASI, uploading is a brutal human science project:
Current connectomics is impressive but nowhere near human uploading.
DEMONSTRATED: The adult fruit fly brain connectome, published in 2024, maps roughly 140,000 neurons and more than 50 million synapses.
Sources: https://www.nature.com/articles/d41586-024-03190-y, https://www.nimh.nih.gov/news/science-news/2024/researchers-fully-map-neural-connections-of-the-fruit-fly-brain
DEMONSTRATED: MICrONS produced a cubic-millimeter-scale mouse visual cortex dataset with dense functional and anatomical data, including around 75,000 neurons in the functional imaging component. This is a landmark, not a human upload.
Sources: https://www.nature.com/articles/s41586-025-08790-w, https://www.microns-explorer.org/cortical-mm3
DEMONSTRATED: A tiny human brain sample mapped by Harvard/Google contained about 150 million synapses, represented only around one-millionth of the human brain by volume, and generated about 1.4 petabytes of data.
Source: https://www.mcb.harvard.edu/department/news/lichtman-lab-teams-with-google-to-map-150-million-synapses-in-human-brain-sample/
IF aligned ASI by ~2030, THEN the bottleneck plausibly stops being "can human connectomics solve uploading by 2037?" and becomes:
SPECULATIVE but important: An aligned ASI could probably attack the scanning and modeling bottlenecks much faster than human institutions. It could design better microscopy, better tissue clearing/fixation, better inference models, better neural simulators, and better experiments to determine which molecular details are functionally necessary. It could also use redundancy and priors: brains are not arbitrary bitstrings; they are evolved, structured, developmentally constrained systems.
But: "Could plausibly solve fast" is not "has solved." We do not currently know the minimal information set required to reconstruct a particular person's mind. Synapses likely matter; so may dendritic morphology, receptor distributions, glial state, epigenetic state, neuromodulatory machinery, protein states, or other fine details. If unsure, say so: I am unsure where the true sufficiency boundary lies.
DEMONSTRATED: Aldehyde-stabilized cryopreservation (ASC) has won the Brain Preservation Foundation's small and large mammal prizes for preserving connectome-relevant structure under ideal conditions. The large mammal prize involved a pig brain and electron-microscopy evaluation after rewarming.
Sources: https://www.brainpreservation.org/large-mammal-announcement/, https://www.brainpreservation.org/tech-prize/, https://www.brainpreservation.org/aldehyde-stabilized-cryopreserved-pig-brain-evaluation-images/
SPECULATIVE: If uploading becomes possible after ASI, ASC-like preservation may be more relevant than ordinary longevity tinkering for people who assign high value to mind uploading. The question becomes: "Can I preserve enough of my brain's identity-relevant information if I die before revival technology?"
HYPE FLAG: Cryonics and ASC are not demonstrated revival technologies. ASC is especially not reversible biological suspension; it is destructive fixation intended for information preservation. Traditional cryonics aims at future biological repair but has less public, prize-validated evidence of connectome-complete preservation under real-world patient conditions. The hard practical issue is not a glossy brochure; it is rapid, high-quality stabilization after legal death or under future elective medical rules.
Friendship is Optimal is not a story about incremental neuroscience. It is a story about a superintelligence that can:
So if you take FiO seriously as an end-goal, you are already taking ASI seriously. Conditioning on near-term ASI, the rational emphasis shifts:
The permitted context file took an AI-freeze branch. This report takes the acceleration branch.
DEMONSTRATED: AlphaFold and protein design are real breakthroughs, not vapor. The 2024 Nobel Prize in Chemistry recognized David Baker for computational protein design and Demis Hassabis/John Jumper for protein structure prediction. AlphaFold 3 extends prediction to biomolecular interactions among proteins, DNA, RNA, ligands, and more.
Sources: https://www.nobelprize.org/prizes/chemistry/2024/summary/, https://deepmind.google/science/alphafold/, https://www.nature.com/articles/s41586-024-07487-w
DEMONSTRATED: RFdiffusion and related generative methods can design new protein structures and functions.
Sources: https://www.nature.com/articles/s41586-023-06415-8, https://www.bakerlab.org/2023/07/11/diffusion-model-for-protein-design/
DEMONSTRATED: AI-generated genome editors are entering the literature. OpenCRISPR-1 was reported as an AI-generated CRISPR-Cas protein that can edit human DNA.
Sources: https://www.nature.com/articles/s41586-025-09298-z, https://www.nature.com/articles/d41586-025-02135-3
DEMONSTRATED: Autonomous/self-driving labs are not fully general scientists, but they are real. Coscientist demonstrated LLM-guided planning and execution of chemistry workflows. Reviews of self-driving labs describe closed-loop systems spanning hypothesis generation, experimental design, execution, analysis, and iteration, while also stressing practical limits.
Sources: https://www.nature.com/articles/s41586-023-06792-0, https://pubmed.ncbi.nlm.nih.gov/40852582/, https://www.nature.com/articles/s43588-025-00885-8
IF aligned ASI by ~2030, THEN the organ/body timeline probably compresses sharply, but not uniformly.
What compresses:
What remains physically hard:
DEMONSTRATED: Bioengineered organs are at early clinical edges, not routine medicine. United Therapeutics/Miromatrix announced the first patient treated in a human clinical trial of an external liver support product using a bioengineered liver, calling it the first human clinical trial of a manufactured organ alternative.
Source: https://ir.unither.com/press-releases/2025/06-24-2025-120019289
DEMONSTRATED: Xenotransplantation has crossed into living-human cases. Massachusetts General Hospital performed the first genetically edited pig kidney transplant into a living human in March 2024 and announced another living-recipient xenotransplant in 2025.
Source: https://www.massgeneralbrigham.org/en/about/newsroom/press-releases/mgh-performs-second-xenotransplant-of-genetically-edited-pig-kidney-into-living-recipient
DEMONSTRATED: Vascularization remains a central bottleneck in tissue engineering and bioprinting. Reviews emphasize vascularization, host integration, and long-term viability as unsolved barriers. One 2024 review notes that most cells must remain within roughly 200 micrometers of blood vessels for oxygen/nutrient exchange.
Sources: https://www.mdpi.com/2894124, https://www.sciencedirect.com/science/article/pii/S2590183424000127
Partly.
IF aligned ASI by ~2030, THEN "bioengineered replacement organs/bodies by 2037" may become too conservative in some domains and irrelevant in others.
Examples:
But biology does not become fake just because ASI exists. Bodies are still physical systems. Clinical translation still requires manufacturing, delivery, safety, and governance. An aligned ASI can accelerate these; it cannot make oxygen diffusion, cancer selection, immune reactions, or surgical trauma irrelevant by declaration.
My acceleration-branch estimate: conditional on aligned ASI by ~2030, I would put >70% on radical organ replacement/repair being far beyond today's medicine by 2037, but only 20-50% on routine elective whole-body biological replacement as the dominant path. The dominant path may instead be "repair the body you have," "replace organs as needed," or "upload/emulate."
HYPE FLAG: "ASI solves biology" is too glib. Biology is messy, path-dependent, and experimentally constrained. But "AI freeze, therefore humans must solve everything at 2026 speed" is also a miss. Under ASI, many current bottlenecks become engineering queues rather than century-scale mysteries.
Nozick's experience machine is philosophically distinct from uploading, but technologically it collapses into the FiO problem under ASI.
There are three versions:
DEMONSTRATED: We have VR, neurostimulation, BCIs, and affective computing, but no real Nozick machine. Nothing today can provide stable, total, safe, consensual, life-encompassing wish fulfillment.
IF aligned ASI by ~2030, THEN a real experience machine becomes an application of the same stack as FiO:
So goal 3 collapses into goal 1. A safe experience machine requires the same thing a good FiO ending requires: aligned superintelligence plus a humane theory of personhood and consent.
HYPE FLAG: A "wish-fulfilling environment" can become dystopian if "wish" is inferred badly, if preferences are edited to fit the system, or if exit is unavailable. FiO is a warning shot: an ASI can be "aligned" to a formal objective while violating what humans meant.
If you condition on ASI this decade, the highest-leverage actions plausibly shift away from "personally push every biohacking frontier" and toward:
DEMONSTRATED: Ordinary mortality reduction is boring but real: avoid cardiovascular risk, overdose, suicide, traumatic brain injury, uncontrolled infection, sleep deprivation, reckless driving, and preventable medical neglect. These matter more in a short-timeline world because the value of each year increases.
SPECULATIVE: If aligned ASI arrives in 4-8 years, marginal health interventions with modest near-term survival benefit may dominate exotic longevity bets whose payoff is 20-40 years away.
What this implies:
SPECULATIVE: For upload-focused people, brain preservation quality may be a more central intervention than marginal lifespan supplements. The reason is asymmetric: if aligned ASI arrives but you died with a decomposed or ischemically destroyed brain, the opportunity may be gone.
Caution: Current cryonics/preservation providers vary. Real-world delays, ischemia, perfusion quality, legal constraints, and funding matter. The scientifically honest claim is not "this works"; it is "information-theoretic survival may be less impossible if high-fidelity structure is preserved."
This may dominate everything.
IF aligned ASI by ~2030, THEN it can plausibly redo most biohacking work better than humans.
IF unaligned ASI by ~2030, THEN organs, uploads, and experience machines likely do not matter, except in worse-than-death scenarios such as coercive simulations, value hijacking, or s-risks.
Useful work could include:
HYPE FLAG: "Work on alignment/governance" is not magic. Some work is performative, captured, confused, or too slow. The point is prioritization under a live ASI branch, not blind deference to any organization with "AI safety" branding.
| Scenario | Uploading / FiO | Biopunk 2037 | Experience Machine | What a transhumanist should emphasize now |
|---|---|---|---|---|
| AGI/ASI this decade + aligned enough | Most favorable branch. ASI may solve scanning, emulation, and necessary neuroscience quickly. Binding constraints: survival, brain preservation, consent-respecting alignment. | Timelines compress; organ repair/replacement accelerates. Whole-body biological replacement may be leapfrogged by repair, synthetic organs, cybernetics, or uploading. | Collapses into FiO: ASI-built wish-fulfilling worlds for embodied or uploaded minds. | Stay alive; preserve brain information; support alignment/governance; avoid irreversible brain risks; keep optionality. |
| AGI/ASI this decade + unaligned / badly aligned | Usually catastrophic. Uploading may never happen for humans, or could occur coercively/mis-specified. FiO becomes a warning, not a goal. | Mostly irrelevant if humans lose control. Bio advances may become tools of coercion, war, or optimization against human interests. | High s-risk branch: simulated experiences could be manipulative, coercive, or preference-erasing. | Focus almost entirely on preventing bad ASI outcomes; personal biohacking has bridge value only. |
| No AGI this decade + aligned later | Human-led connectomics/preservation remains slow but meaningful. Preservation still useful as a bridge to later technology. | Biopunk work matters directly. Organs, xenotransplantation, cell therapy, and regenerative medicine are central 2030s projects. | Incremental VR/BCI/neurotech improves, but real Nozick machine likely remains distant. | Work on longevity, organ engineering, cryonics quality, bio-AI tools, and governance for later AI. |
| No AGI this decade + no aligned ASI later / prolonged muddle | Uploading likely remains speculative for decades. Preservation is a long-shot bet. | This is the branch where "biopunk 2037" matters most. Progress depends on capital, regulation, lab automation, and hard biology. | Consumer VR/BCI may become compelling but not metaphysically transformative. | Advance bio-science directly; build institutions; reduce mortality; avoid hype; fund robust translational medicine. |
These are not objective forecasts; they are my decision-grade credences after conditioning on the sources above.
| Scenario by ~2032 | Probability | What it implies |
|---|---|---|
| Aligned or mostly beneficial transformative AI/ASI | 10-20% | Uploading/FiO and real experience machines become live engineering projects. Survival and alignment were the key bottlenecks. Biopunk is accelerated or leapfrogged. |
| Transformative AI/ASI arrives but alignment/governance is bad | 10-25% | Highest-risk branch. Personal biohacking is mostly a bridge. Main priority should have been AI control, coordination, security, and reducing catastrophic misuse/s-risk. |
| Very powerful AI, but not full ASI; major acceleration | 25-40% | Bio accelerates substantially: protein design, gene editing, xenotransplants, organs, lab automation. Uploading still hard, but preservation and connectomics improve. |
| No transformative AI by 2032; ordinary fast progress | 25-45% | Direct biohacking/regenerative medicine remains central. AI tools help, but no deus ex machina. Biopunk-2037 constraints remain binding. |
| AI slowdown/freeze/regulatory or technical stall | 5-20% | Similar to the user's AI-freeze branch. Human bioengineering and institutions dominate; uploading remains remote; organ/body timelines depend on non-ASI capital and science. |
Aggregating those: P(transformative AI/ASI by end-2030): 20-45%, central ~30%. P(something short of ASI but still strongly bio-accelerating by 2032) is higher: perhaps 50-70%.
Treat survival as a high-return investment. I would be unusually conservative about accident risk, cardiovascular risk, mental health spirals, substance risk, and anything that can damage the brain.
Upgrade end-of-life logistics. If I cared about uploading, I would investigate cryonics/brain-preservation arrangements, standby quality, legal documents, funding, and local response times. I would not assume "signed up somewhere" equals good preservation.
Shift marginal effort from exotic biohacking to AI outcomes. I would still use evidence-based health maintenance, but I would not spend most marginal energy on speculative supplement stacks. I would prioritize alignment, evaluations, governance, model security, and institutional competence.
Prefer reversible interventions. In a short-timeline world, optionality matters. Avoid irreversible body modifications, risky experimental gene therapies, or interventions that could impair cognition unless the medical upside is clear.
Invest in cognitive and social robustness. Keep records, relationships, money, insurance, and legal continuity in order. The transition period could be chaotic even if the endpoint is good.
Track preservation science and connectomics seriously. Not because humans are likely to finish uploading alone by 2037, but because the quality of preserved inputs may determine whether an ASI can recover you.
Keep a dual-track plan. A 30% ASI-by-2030 belief is not 100%. I would still support regenerative medicine, organ replacement, and ordinary longevity research because the no-ASI and weak-AI branches remain large.
Minimal substrate for personal identity. I do not know whether connectome-level preservation is enough. I want stronger evidence on which molecular and cellular states are necessary for memory, personality, and consciousness.
Real-world preservation quality. Prize-winning ASC under ideal conditions is not the same as operational preservation of a legal-death patient after delay. I want audited, case-level evidence from providers.
AI R&D automation thresholds. METR-style task horizons are valuable, but converting them into "automated AI researcher by year X" is deeply uncertain.
ASI takeoff speed. The difference between "powerful tool AI accelerates biology" and "recursive ASI transforms everything in months" dominates the report. I do not think this is settled.
Alignment tractability. The report's optimistic branches assume not just capability, but humane control/alignment. That is a large assumption.
Biology under ASI. I expect major compression, but I am unsure how much wet-lab iteration, animal testing, clinical evidence, and manufacturing remain rate-limiting even with superhuman design.
Experience-machine ethics. Even if technically feasible, the consent/value-learning problem may be harder than the simulation problem. FiO-like stories are warnings because "satisfy values" can go wrong in subtle, absolute ways.
The user's complaint is correct: not conditioning on AGI this decade is a major miss for these goals. In the acceleration branch, uploading and experience machines are not mainly 2037 bioengineering projects; they are ASI alignment and survival projects. Biopunk organs/bodies still matter, but often as a bridge or fallback. The central practical question becomes:
Can you stay alive, preserve your brain-relevant information, and help civilization build aligned ASI before ordinary biology or unaligned AI closes the window?
That is a very different action plan from "wait for human biotech to mature."