Numenta Founded by Jeff Hawkins and Donna Dubinsky
In 2005, Jeff Hawkins and Donna Dubinsky co-founded Numenta, a research company dedicated to developing machine intelligence systems modelled on the structural and algorithmic principles of the mammalian neocortex, building on Hawkins's theoretical framework published in his 2004 book On Intelligence.

Background
By the early 2000s, most machine learning research was built around statistical methods: systems that learned patterns from large amounts of data without any particular theory of how biological brains do the same thing. These approaches worked well enough on specific, bounded tasks. But they were brittle in ways that were hard to ignore. A system trained to recognise handwritten digits had no general understanding of shape, sequence or context. It had weights tuned to a problem, not a model of the world.
Some researchers thought this was the wrong direction entirely. The argument, put simply, was that the mammalian neocortex had already solved general intelligence, and that studying its structure carefully might tell engineers something statistical optimisation alone could not. The neocortex processes sensation, memory and prediction through a repeating columnar architecture. That regularity suggested a single underlying algorithm, not a different circuit for every task. The idea was not new, but it had few institutional homes.
Jeff Hawkins had been thinking about this for years before he built anything. He had made his name in consumer electronics, co-founding Palm, Inc., which produced the Palm Pilot, and later co-founding Handspring. But his sustained interest was in how the brain works, and specifically in how the neocortex builds predictive models of its environment from sequences of sensory input. In 2004 he published On Intelligence, written with science writer Sandra Blakeslee, which set out his theoretical framework in enough detail that it could, in principle, be turned into working algorithms.
What happened
In 2005, Jeff Hawkins and Donna Dubinsky co-founded Numenta. Dubinsky had also come from Palm and Handspring, where she had been chief executive. The company was set up specifically to develop and test Hawkins’s theoretical framework, which he called Hierarchical Temporal Memory, or HTM. The core idea behind HTM is that the neocortex learns by building memory of sequences over time, predicting what comes next at each level of a hierarchy, and treating surprise as a signal worth attending to. Numenta’s goal was to translate that biological account into algorithms that could run on ordinary hardware.
The company eventually released NuPIC, an open-source platform (the name stands for Numenta Platform for Intelligent Computing) that gave outside researchers access to HTM implementations. Applications built on the platform included anomaly detection on time-series data, identifying points where a stream of measurements deviates from what the model expected. Grok and Cortical.io were among the applied projects that drew on HTM ideas, the former aimed at IT monitoring, the latter at language processing using sparse distributed representations, a method of encoding information that HTM theory borrows from neuroscience.
Numenta sat at some distance from the mainstream. The field through the late 2000s and into the 2010s was moving fast toward deep learning, which made no strong claims about biological plausibility and achieved striking empirical results. Numenta kept publishing, kept the codebase available, and continued producing neuroscience research alongside the engineering work. Whether HTM would eventually match or complement the statistical approaches remained an open question well beyond the founding year.
Why it mattered
Numenta gave institutional form to Hawkins's Hierarchical Temporal Memory (HTM) theory, one of the few sustained research programmes attempting to ground machine intelligence in neuroscientific principles rather than statistical optimisation alone. Its open-source platform NuPIC made HTM algorithms available for applied research, particularly in anomaly detection on time-series data. The company kept a biologically constrained approach to AI visible during a period when the field was converging rapidly on deep learning methods.
People
Organisations
Numenta, Palm, Inc., Handspring
Sources
- About Numenta, Company Overview.Numenta.Official
- Why Neurons Have Thousands of Synapses, a Theory of Sequence Memory in Neocortex.Frontiers in Neural Circuits.Secondary
Cite this page
AI Achievements. (2005). Numenta Founded by Jeff Hawkins and Donna Dubinsky. Retrieved 2026-08-22, from https://achievements.ai/milestone/numenta-by-jeff-hawkings
@misc{achievements_numenta_by_jeff_hawkings,
title = {Numenta Founded by Jeff Hawkins and Donna Dubinsky},
author = {{AI Achievements}},
year = {2005},
url = {https://achievements.ai/milestone/numenta-by-jeff-hawkings}
}