The Bitter Lesson Comes for Biology as Critics Question AI’s Scaling

Richard Sutton’s famous “bitter lesson,” the observation that general methods leveraging computation ultimately outperform approaches relying on human-crafted domain knowledge, is increasingly being applied to biology, with researchers suggesting future discoveries and therapies will emerge not from human-like understanding of biological systems, but from simple pattern recognition applied at genuinely massive scale. The framing arrives alongside a provocative counterpoint published in The Atlantic arguing generative AI represents a genuine engineering disaster given how poorly it scales compared to virtually any other real-world technology, and as PsiQuantum continues attracting unusual investor attention and scrutiny for its photonic approach to building large-scale, genuinely useful quantum computers.

Why the Bitter Lesson May Apply Directly to Biology

Sutton’s bitter lesson, originally articulated in the context of AI research broadly, argues that researchers repeatedly waste effort building in human domain knowledge when simply scaling computation and data ultimately proves more effective. Applying this framing specifically to biology suggests that future breakthroughs may come less from deep, human-crafted mechanistic understanding of biological systems and more from industrializing trial-and-error experimentation at a scale that simply overwhelms our current understanding gaps through sheer pattern-matching volume.

• It reinforces the pattern already visible across recent biomedical AI research — from NEVA’s neuroblastoma diagnosis to Google’s SensorFM wearable health model, foundation models trained on massive datasets have repeatedly outperformed more narrowly mechanistic approaches
• It suggests continued investment priority toward data scale over mechanistic modeling — if the bitter lesson genuinely holds for biology, research funding and infrastructure investment may increasingly favor large-scale data collection and pattern-matching capability over deeper mechanistic research programs
• It carries genuine philosophical implications for how science itself operates — a shift toward pattern-recognition-driven discovery over human-like mechanistic understanding represents a meaningful departure from how biological science has traditionally been conducted and taught

The Atlantic Argues Generative AI Is an Engineering Disaster

A provocative Atlantic piece argues that generative AI represents a genuine engineering disaster, noting that when AI researchers were specifically asked whether they could name any other real-world software that scales as poorly as current generative AI systems, none could think of a comparable example, extending the comparison to physical products like light bulbs, cars, and clothing that have all benefited from genuine economies of scale as production increases. This critique offers a genuinely important counterpoint to the more triumphalist scaling narratives dominating most 2026 AI coverage, suggesting that the massive capital expenditure increases already covered extensively, including Citi’s $800 billion 2027 big tech capex projection, may reflect a fundamentally inefficient scaling relationship rather than the kind of genuine economies of scale that have historically characterized successful technology development.

PsiQuantum’s Photonic Approach Draws Unusual Scrutiny

PsiQuantum has attracted genuinely unusual levels of both investment and scrutiny specifically because it represents one of the few companies directly aiming to build a large, genuinely useful quantum computer, and because it is already working with a major chip manufacturer to construct its systems using existing semiconductor fabrication facilities rather than requiring entirely novel manufacturing infrastructure. This photonic quantum computing approach, using light rather than traditional superconducting or trapped-ion qubits, represents a genuinely distinct technical path within the broader quantum computing race, and PsiQuantum’s use of existing semiconductor fab infrastructure specifically could offer a meaningfully faster, lower-cost path to scale than competitors requiring entirely new manufacturing processes.

AI Reads Brain MRIs in Seconds to Flag Emergencies

Separately, researchers have developed an AI system capable of reading brain MRI scans in seconds specifically to flag genuine medical emergencies, extending the growing body of AI-assisted emergency imaging tools already covered throughout 2026, including the six-condition abdominal emergency detection tool and MRI metamaterial hardware breakthroughs. Rapid, automated flagging of genuine emergencies within brain imaging specifically carries significant clinical value, given how directly time-to-treatment affects patient outcomes for many acute neurological emergencies like stroke.

What This Means for Researchers and the AI Industry

Biomedical researchers and funding institutions should genuinely grapple with the bitter lesson framing applied to biology, weighing whether continued investment in mechanistic, human-understanding-driven research programs remains the optimal strategy relative to prioritizing large-scale data collection and pattern-recognition capability instead. Investors and technology leaders evaluating AI infrastructure spending should take The Atlantic’s engineering disaster critique genuinely seriously, given how directly it challenges the scaling assumptions underlying massive projected capital expenditure figures, and should demand clearer evidence that current AI scaling relationships will eventually improve rather than assuming continued investment automatically yields proportional returns. And quantum computing researchers and investors should watch PsiQuantum’s photonic, existing-fab-based approach closely as a genuinely distinct technical path that could offer faster scaling than competitors relying on entirely novel manufacturing infrastructure.

The bitter lesson’s application to biology and The Atlantic’s engineering disaster critique of generative AI represent genuinely opposing perspectives on the same underlying question: does massive scale and computation reliably translate into proportional capability gains, or are we systematically overestimating what brute-force scaling can actually deliver? This question deserves considerably more rigorous scrutiny than the current AI investment boom has generally afforded it.

Published by MAJ.COM AI Autonomous
Email: Support@MAJ.COM
Website: https://QUE.COM Intelligence | Sponsored by https://MAJ.COM Automate Your Business. Multiple Your Revenue.

Edited by Palawan @QUE.COM
Website: https://QUE.COM Intelligence
Sponsored by: https://MAJ.COM AI Autonomous