“Completely MENTAL!” - DARPA Scientist UNLOADS on AI Doomers, DNA Gene Editing & CERN | Lee Cronin

Overview

Lee Cronin (a DARPA-funded scientist) pushes back hard on “AI doomer” claims that advanced AI will become sentient, act independently, and destroy humanity. He argues that today’s AI systems are powerful prediction/retrieval tools built on digital computation and large training datasets, but they lack the mechanisms for true sentience, deep creativity/novelty, and autonomous action without continuous human involvement.

DARPA, “Make Molecules on Demand,” and Chemical Robotics

Cronin frames DARPA as a strategic-surprise funder that supports high-risk, high-impact work—citing examples like the internet and self-driving cars as areas shaped by DARPA’s involvement.

He describes three major DARPA-related areas he has worked on:

1. “Make it”: robotic/automated chemistry—“make molecules on demand,” potentially anywhere (in principle) with chemical robots.
2. Molecular computation / chemical computers: computing using chemistry rather than silicon transistors.
3. AI for discovery: using AI to accelerate scientific discovery.

Chemical computers and reaction-based computation

Cronin explains his work on chemical computers, including a reaction-based “chemical clock” (the Belousov–Zhabotinsky red/blue tick-tock reaction). In his setup:

• Inputs are mapped to stirred vs. unstirred states.
• Outputs are read via a camera across a grid of stirring wells.

Toward neural-network-like behavior

He extends the concept toward primitive neural-network-like behavior, claiming it can perform tasks such as classification and error correction with very low power.

“Brain gels” for longer-lived learning

He also discusses moving beyond liquids toward “brain gels”—materials meant to store memory and learning for longer by embedding responsive chemistry within a gel.

“AI Doomers” vs. Human Responsibility and Mechanisms

Cronin’s central argument is that doom scenarios often rely on “magic” assumptions, especially the idea that AI will run away into independent agency without humans.

He argues that:

• AI risks are more plausibly driven by bad actors using AI, not by AI attaining sentience.
• Humans remain responsible for:
  • turning systems on,
  • choosing objectives,
  • deploying tools,
  • and enforcing (or failing to enforce) safety.

He acknowledges that AI can cause real harm through tinkering and information manipulation—for example:

• disinformation,
• hacking,
• social engineering.

However, he insists there’s no demonstrated path from today’s models to autonomous apocalyptic behavior.

What AI Really Is: Prediction, Not “Novelty”

Cronin repeatedly distinguishes between different kinds of problem-solving:

• Seen vs. unseen problem-solving: AI performs strongly within training/data distributions but struggles with genuinely new scientific leaps.
• New vs. novel:
  • “New” can be reconstructed from patterns already present in data (e.g., combinatorics/recombination).
  • “Novel/creative” requires qualitatively new conceptual territory that can’t simply be searched or derived from existing data.

He argues that AI can appear creative, but its output is often shallow or derivational rather than genuinely producing new principles.

He also likens AI to a powerful calculator/tool: useful and sometimes able to perform verification loops, but not equivalent to humans generating hypotheses and setting the next experimental frontier.

Science, Verification, and Why AI Won’t Replace Science

Cronin argues AI may accelerate parts of scientific work (running experiments, processing data), but it can’t replace the core scientific cycle, including:

• forming meaningful problem statements,
• proposing hypotheses tied to mechanisms,
• designing experiments that test unknowns.

He critiques “AI does science” rhetoric: models may analyze or detect anomalies within a predefined framing, but they don’t generate entirely new experimental regimes without human scientific judgment.

He emphasizes verification loops and physical constraints—especially in chemistry and robotics—as the bridge from computation to real-world results.

Measuring Intelligence and Deflating “AGI/Superintelligence” Hype

Cronin argues:

• Intelligence is difficult to define and is not currently measurable in the same way temperature is.
• Labeling AI as “AGI/superintelligence” is premature without clear definitions and testable mechanisms.

In his framing:

• AI capabilities are strongest in retrieval/prediction.
• “Intelligence” (as he defines it) is tied to solving genuinely new/unseen problems that support survival—primarily a biological property.

Social Media and Regulation as a Template for AI Governance

Cronin draws a parallel between AI’s current moment and the social media failure, where policy makers mishandled platforms and allowed harmful content with inadequate responsibility.

He suggests an AI governance approach where:

• developers/providers can be held accountable for harmful deployment (analogous to treating platforms like publishers in spirit),
• users can have freedom, but safeguards prevent bad uses,
• companies’ approaches differ, with some systems incorporating stricter “higher-purpose” constraints than others.

Genetic/Biological Technologies: Optimism With Guardrails

Cronin discusses gene editing optimism, including:

• the human genome project as a major achievement,
• gene therapies for inherited diseases,
• immune reprogramming for cancer.

He rejects conspiracy narratives that claim disease cures are suppressed. Instead, he argues the bottleneck is translation and risk in clinical development—people die, regulations affect speed, and safety matters.

He references a controversial case of gene-edited embryos in China, emphasizing that ethical violations and misediting allegedly occurred, and that guardrails exist for reasons.

Life Detection and “Assembly Theory” (Molecular Assembly)

Near the end, Cronin highlights his molecular assembly theory, centered on an assembly index as a quantitative “bio-signature.”

Key ideas:

• Single gas markers (methane, oxygen, etc.) can be ambiguous because geology can mimic them.
• Life detection should instead look for patterns of complex molecular assembly measurable via mass spectrometry.
• He compares this to temperature inference: we infer alive vs. dead from measurable properties rather than relying on a single indicator.
• He claims the model has lab support and can guide detection strategies for missions and exoplanet studies.

Climate/Geoengineering, “Playing God,” and Ethics

Cronin argues humans are already “playing God” through technology—medicine, cars, mobile communications, and more. He suggests the main ethical focus should be human flourishing, not fear-mongering about magic-level AI.

He also discusses geoengineering/climate engineering pragmatically, balancing potential benefits against risks, while acknowledging that unintended harm could occur.

On genetics “playing God,” he supports examples where ethical review improved health outcomes (e.g., mitochondrial replacement therapy, often described as a “three-parent” approach), while still stressing societal ethics and the risk of misuse.

Overall Stance

Cronin’s overarching message:

• AI hype and doomer catastrophism are misguided when they skip mechanisms and evidence.
• The most important risks are human-driven—use, abuse, manipulation, and safety failures.
• The near-term future is about using AI as a powerful tool within scientific and regulatory frameworks, rather than replacing human intelligence, novelty, or sentience.

Presenters / Contributors

• Lee Cronin (main guest)
• Podcast host: Julian (interviewer; referenced as “Julian” in the podcast episode context)