Well Principled

We aren’t building AI to replace "knowledge work” that requires neither, but to bolster the working human intellect which has never failed to discover new paths to prosperity.

For all their success, mainstream approaches to AI have fundamental limitations preventing them from solving critical classes of problems. We are building a new AI architecture from first principles to break from these limitations.

The status quo has ossified around the supervised, brute-force pretraining paradigm. These systems do not, and cannot, handle the essential skills of continual learning, goal discovery, novel planning, and the creation of new knowledge.

These limitations are not trivial. Our national prosperity depends directly on overcoming these limitations, because the essential skills underlie the skilled trades. From electricians to machinists to nurses, apprentice tradespeople in all of these fields learn “skilled improvisation” where every solution is novel, and even the goal must be discovered. Prosperity demands the precise capabilities where status quo AI is structurally weak.

Worse, these limitations will not be overcome with more compute or training data. They are inescapable consequences of the underlying technologies of supervised learning, neural networks, transformers, backprop, and gradient descent. Progress requires independence from them.

We are building a new AI architecture by reverse-engineering the system that got us this far: the human brain. By triangulating between neuroscience, cognitive science, and the evolutionary elaboration of brains, we have identified the brain’s algorithms to (a) continually learn a causal world model (b) select goals to both achieve and learn to achieve more (c) plan to reach novel goals, and (d) creatively produce new knowledge.

Join us in building intelligence with ingenuity to bolster the skilled trades.