From Minsky's Critique to
Deep Cognition Labs

DeepCog.ai is a tribute to one of the most important debates in AI history — and to the prophet who was right all along.

MIT's Cog Project (1993–2003)

In the early 1990s, the Humanoid Robotics Group at the Massachusetts Institute of Technology launched one of the most ambitious experiments in artificial intelligence history. Led by Rodney Brooks and Lynn Andrea Stein, the project was called Cog — a humanoid robot built on a bold hypothesis: that human-level intelligence could only emerge through physical embodiment, social interaction, and the kind of sensory experience that human infants acquire over years of living in the world.

Cog was designed to see, hear, move its arms, track faces, and respond to social cues. It was, in the language of its creators, meant to grow up rather than simply be programmed. The project attracted enormous attention — Cog appeared on ABC's Brave New World in a segment drumming with They Might Be Giants, and its earliest popular science appearance was a 1994 Wired article titled "Bringing Up RoboBaby."

For a decade, Cog stood as one of the most recognized faces of humanoid robotics research. As of 2003, all development of the project had ceased. Today, Cog is retired to the MIT Museum.

Wikipedia: Cog Project → MIT CSAIL: Cog Project → Wired: Bringing Up RoboBaby →

Marvin Minsky's Prophetic Critique

🧠

Marvin Lee Minsky August 9, 1927 – January 24, 2016

Harvard- and Princeton-trained mathematician, cognitive scientist, and co-founder of MIT's AI Laboratory alongside John McCarthy. Widely considered one of the "fathers of artificial intelligence," Minsky received the ACM Turing Award (1969), the Japan Prize (1990), and the Benjamin Franklin Medal (2001), among many other honors. Author of The Society of Mind (1986) and The Emotion Machine (2006).

Wikipedia: Marvin Minsky →

"Robotics research is really a software problem."

—  Marvin Minsky, Turing Award laureate & co-founder of MIT's AI Laboratory — on the Cog Project at MIT's Humanoid Robotics Group. Source: Wikipedia: Cog (project)

Not everyone was convinced that building a physical robot was the right approach. Minsky's dissent was not idle skepticism — it was a principled, deeply considered position from someone who had spent his entire career studying how minds work. His foundational works — The Society of Mind, The Emotion Machine, and his landmark theory of frames — all converged on the same thesis: intelligence is fundamentally a matter of computation and representation, not metal and motors.

The problem of intelligence, Minsky argued, was a problem of ideas, not anatomy. Steel and servo motors were a distraction from the real challenge. Cognition was a software problem — and it needed a software solution.

From Physical Robot to Software Intelligence

Minsky's critique proved prescient. The arc of AI history vindicated his position completely.

1969

Minsky receives the ACM Turing Award

Recognized for foundational contributions to AI, Minsky's theoretical framework — that intelligence is a computational phenomenon — becomes central to the field. Source →

1986

Minsky publishes The Society of Mind

Argues that human intelligence emerges from the interaction of many small, simple computational processes — a software architecture for the mind. Source →

1993

MIT launches the Cog Project

Rodney Brooks and Lynn Andrea Stein begin building a humanoid robot at MIT's Humanoid Robotics Group, based on the embodied cognition hypothesis. MIT CSAIL →  ·  Brooks & Stein (1993) →

1994

Cog reaches popular media — "Bringing Up RoboBaby"

David H. Freedman's landmark article in Wired brings the Cog project to global attention, reporting that AI consciousness was part of the original goal. Wired →

2003

Cog development ceases

After a decade of research, the project concludes. Cog is retired to the MIT Museum. The embodied intelligence hypothesis has not produced human-level cognition. Source →

2010–2015

A Herndon, Virginia makerspace attempts to restart Cog

An attempt was made between 2010 and 2015 by a makerspace group in Herndon, Virginia, to restart the Cog Project. Marvin Minsky was contacted for guidance. A series of Medbots were designed as Cog robots, ranging from hospital assistants to a phlebotomist robot named Vispra.

2012–2017

Deep learning transforms AI — in software

The ImageNet breakthrough, AlphaGo, and transformer architectures demonstrate that the path to general intelligence runs through software, data, and computation — exactly as Minsky predicted.

January 24, 2016

Marvin Minsky passes away, aged 88

Minsky does not live to see large language models reshape the world — but the arc of his argument has been vindicated completely. Source →

2022

Deep Cognition Labs — DeepCog.ai — is founded

After successive attempts at embodied and robotic AI failed to achieve human-level cognition, DeepCog.ai is born from the software-first philosophy Minsky advocated — building medical intelligence in code, not metal.

DeepCog.ai — Minsky's Vision in Software

Deep Cognition Labs, operating as DeepCog.ai, takes its name as a deliberate homage to MIT's Cog project and to Minsky's enduring insight. The "Cog" in DeepCog is not accidental. It is an acknowledgment that the project Minsky criticized — the attempt to build cognition from the ground up — was asking exactly the right question, even if the physical robot was the wrong answer.

Where Cog reached for cognition through mechanical arms and camera eyes, DeepCog reaches for it through language, reasoning, and medical knowledge encoded in billions of parameters. Where Cog aimed to learn by interacting with humans in a lab, DeepCog learns from 42 million peer-reviewed papers, clinical trial records, genomic databases, and expert medical reasoning chains.

The vision is the same. The medium is software — exactly as Minsky said it should be.

  • OpenBioLLM Achieves 91.2% on MedQA, surpassing GPT-4 on seven of nine medical benchmarks. Trained on 42M+ PubMed abstracts with DPO alignment.
  • GenomicLLM Trained on 180M genomic sequences from NCBI and Ensembl for variant interpretation and gene-disease association mapping.
  • ClinicalReasoner Fine-tuned on de-identified EHR data and clinical pathways for differential diagnosis and evidence-based decision support.
  • DrugDiscovery-LLM Models ADMET properties across 10M+ compounds from ChEMBL and PubChem for early-stage drug discovery acceleration.
  • PathologyVision-LLM Multimodal AI for histopathology slide analysis, tumor grading, and automated radiology report generation.

These are not robots. They have no arms, no cameras, no social stimuli to respond to. They are, in the purest sense, what Minsky called for: the intelligence problem solved in software, at scale, for real human benefit.

The Name Carries the Legacy

"Marvin Minsky did not live to see large language models reshape the world. But the arc of his argument has been vindicated completely. DeepCog.ai stands as a tribute to his foresight — to the idea that cognition is a software problem, that minds can be built from mathematics and data, and that the most important frontier in intelligence research is not the body that carries it, but the reasoning it performs."

The name carries the legacy. The work carries the mission.

Sources & Further Reading