Why We're Living in a Biological Singularity With Ben Lamm | MOONSHOTS Live #297
Summary
Peter introduces Colossal as now worth over $10 billion, while Alexander Wissner-Gross notes more than four spin-offs. Lamm says his lack of formal biology training is an advantage: he hires people “significantly smarter than you” and asks questions; 90% may slow meetings, but 10% make experts say, “We never thought of it that way.”
Lamm says AI’s near-term biological value is less autonomous discovery than connecting fragmented research, standardizing language, and designing experiments that laboratories still must validate. Inconsistent gene names and protocols can produce 40–60% replication failure rates; Dave Blundin adds that the dataset may be more valuable than the model itself, especially when comparative genomics spans thousands of genomes.
Lamm rejects literal resurrection as the end state: synthetic biology can produce close analogues, but incomplete fossils, epigenetics, environment, and development make exact restoration implausible. Fewer than 100 tyrannosaurs have been found despite billions said to have lived, so humanity is likelier to build “programmable biological systems” than reconstruct the entire historical biosphere.
The constraints on engineered life may become social, ethical, and regulatory as well as technical. Lamm says Colossal does not use its technologies for humans and relays advice to avoid applying them to people too soon because of FDA “code-freeze mode.” Species selection weighs ecological function, extinction cause, Indigenous relationships, animal welfare, and educational value rather than following a universal algorithm.
Lamm expects the capability to engineer key phenotypes, synthesize or multiplex-edit organisms, and develop them outside the womb in fewer than 10 years. Colossal moved from celebrating 20 edits to consistently exceeding 300 at more than 90% efficiency and is testing 1,000; Lamm nevertheless expects large-fragment DNA synthesis to overtake multiplex editing.
Lamm forecasts a fully extrauterine mammalian birth within 24 months, from conception to birth without a surrogate; Peter calls that timeline very likely. Separately, Colossal is mapping genotype to phenotype across size, coat color, folds, scales, feathers, tusks, and patterns. The first woolly mice generation shown publicly used eight base edits in one delivery.
The broader infrastructure thesis is to “back up life” before Lamm’s projected loss of half of global biodiversity over the next 25 years. Colossal offers conservation technologies openly through 75 partners, is building government biorepositories, and wants billions in coordinated public funding; despite expecting “scary moments,” Lamm likens them to aircraft turbulence: “everyone lands anyway.”
Deep dive
1. Colossal treats outsider ignorance as an operating advantage
Lamm’s answer to entering synthetic biology without domain credentials is to hire people “significantly smarter than you” and spend meetings asking questions. He estimates 90% merely expose what he does not know; the valuable 10% cause specialists to reconsider assumptions they had stopped seeing.
George Church, whom Lamm describes as the father of synthetic biology, became both co-founder and teacher. Lamm says Church calls him his best student because he “literally bombard[s] him with questions,” often spending holidays discussing what synthetic biology might make possible.
That curiosity extends to cold-emailing world experts and admitting, “I don’t know that.” Lamm’s operating belief is that people asked sincerely for help will provide it “nine times out of ten”—a method for recruiting advisers as much as a personal philosophy.
2. Biological AI begins as connective tissue, with proprietary data as the moat
Colossal initially used early access to OpenAI and Anthropic models as “intelligent middleware,” linking laboratory logs, Jira, reporting, and internal systems without hiring firms such as Deloitte or Accenture. Early LLMs could write papers but were not useful for ancestral-state reconstruction or comparative genomics.
The next step is experimental simulation and study design, not replacing wet labs. Mouse literature may use different names, classifications, and protocols for the same gene, producing 40–60% replication failure; AI can reconcile that language, surface missed papers, and propose experiments that scientists still verify physically.
A single sequence cannot be uploaded and told, “make these six changes” for immortality. Lamm instead wants local “Noah’s Ark 2.0” biorepositories, ATX and T2T sequencing, and thousands of genomes across evolutionary lineages, tracing outcomes such as disease resistance through the tree of life. Dave Blundin’s conclusion is that “the dataset is more valuable than the model itself.”
AstroMech is intended to build “inflectional models” explaining not only what genomic changes occurred, but when, why, and why they did not appear in related clades. With global biorepositories and millions of samples submitted to Mythos, Lamm says these connected datasets could help guide where to focus rather than relying on one magical universal model.
3. De-extinction yields analogues, not perfect historical restoration
Alexander Wissner-Gross invoked Russian cosmism and asked whether technology could eventually restore every organism—or every human—ever alive. Lamm’s answer was no: synthesis and predictive modeling may create close analogues, but the biological and historical record is too incomplete for literal recovery.
His decisive example is the fossil gap: fewer than 100 tyrannosaurs have been found, despite billions said to have lived across time. Humanity may therefore engineer desired life forms that resemble lost organisms, but “we’ll just never know” whether those exact organisms previously existed.
Genetics also cannot restore identity. Cloning Mozart would not reproduce his developmental environment, epigenetics, or life history; with carefully simulated conditions, a “Mozart 2.0” might even outperform Mozart 1.0, but it would not constitute recovery of the original person.
4. The biological singularity is moving from hundreds of edits toward synthesis
Pressed for a biological “CAD” system that accepts a phenotype request and designs an organism, Peter framed the possibility with examples such as Pikachu. Lamm forecast the necessary capabilities in fewer than 10 years: engineering key phenotypes across different clades, synthesizing or multiplex-editing them, and growing organisms outside the womb.
Colossal once celebrated achieving 20 edits, a level Lamm says 99% of biopharmaceutical companies and research institutions would do today. It now consistently performs more than 300 edits at over 90% efficiency and is testing 1,000, though he cautioned that testing is not success and early working results may remain inefficient.
Lamm expects large-fragment DNA synthesis to replace multiplex editing before edit counts simply scale indefinitely. Because companies in the category lack incentives to synthesize beyond certain scales, Colossal has begun building that capacity internally.
Lamm’s nearer forecast is a fully extrauterine mammalian birth within 24 months, from conception through birth without a surrogate. Peter responds that 24 months is very likely. Human use is a separate question governed by acceptance, ethics, rights, investors, management, and regulation—not merely whether the underlying machinery works.
5. Conservation deployment must solve ecology, upbringing, and scale together
Trait mapping already spans coat color, size, skin folds, scales, feathers, tusks, muzzle length, and patterns; an entire AI team works on patterns and stripes. The thylacine program must extrapolate from a fat-tailed dunnart to an animal roughly 1,500 times larger, while the first woolly mice generation shown publicly used eight edits in one delivery.
Animal welfare includes upbringing. Colossal is working on 16 lines and ultimately wants tens of thousands of mammoths, not an isolated specimen; work with Elephant Havens studies synthetic matriarchal herds, while satellites, drones, and AI map elephant corridors and social hierarchies so extrauterine animals can learn how to live as elephants.
George Church’s climate case, as recounted by Peter, is that grass absorbs roughly six times more carbon than trees, albedo rises two- to threefold, and sufficient mammoth density might cool summer permafrost by 6–8 degrees. Other scientists, including some at Colossal, dispute the magnitude; Peter nevertheless argues that the restoration effect on the degraded tundra ecosystem would be net positive.
Lamm’s moonshot is infrastructure for biodiversity: he believes half may disappear within 25 years and says governments must spend billions to “back up life.” Colossal provides conservation technology freely through 75 partners, has programs in Dubai and with the U.S. Fish and Wildlife Service, and—through Celularity’s LifebankUSA example—uses the placenta as an “original boot disk” that society should not casually discard.