Jeff Hawkins and Donna Dubinsky, creators of the Palm and Handspring personal digital assistants and the Treo smartphone, have formed a software company built around a powerful and unorthodox vision of how the human brain works.

In its early stages, they hope to create predictive machines useful for things like weather forecasting and oil exploration. Further out–much further, says Hawkins–they plan to lay the basis for cosmologically attuned robots that conceive and reflect on the universe itself.

Okay, it is a big idea. And so far the Menlo Park, Calif.-based company, called Numenta, has built what the creators say is a set of tools for creating pattern-recognition software capable of “learning” shapes and events, with a goal of foreseeing what the pattern will next create. Yet these tools draw on decades of work that Hawkins has done on how the brain works. If it pans out–and there is an attractive logic to much of his thinking–Numenta may certainly oversee the creation of embedded software that adapts and improves its own performance.

“I’m trying to build a movement around thinking about how the brain works,” says Hawkins, the main researcher of the two. Adds Dubinsky, the business builder, “If you want to build a lot of people around it, you have to make money on it.”

Hawkins, who will continue to work as chief technical officer of Treo maker palmOne, already has his own research lab to study the neocortex, a massive part of the brain thought to hold such higher level functions as language, learning, memory and complex thought.

Hawkins believes that the several levels of the neocortex are an organizational hierarchy of sensory inputs. This hierarchy has multiple interconnections among levels that enable us to sort things in space and time and associate them with previously encountered things, be they faces, phone numbers or typing skills–whatever our memory holds. Traveling down from the top of the hierarchy to the base sensations, he figures, the neocortex functions as a prediction machine, anticipating what we will see next, where the ball is headed or how an experiment might turn out. In effect, prediction is akin to “remembering” the future.

A researcher at Hawkins’ Redwood Neuroscience Institute developed algorithms to mimic this pattern-seeking/prediction loop. Numenta’s first product is a set of software-building “tools,” similar to those of Borland Software and IBM’s Rational software product, in this case for creating memory-building hierarchies that observe large-scale systems, like weather or geological data. Over time, the idea goes, the machines will learn enough patterns to say where the storms will be, where the oil lies.

The researcher, Dileep George, is along with Dubinsky and Hawkins, a cofounder of the company. Board members include executives from venture capital firm Benchmark Capital, chip company Rambus and Network General, a software company involved in network performance analysis.

The company is self-funded, but Dubinsky, who was formerly co-founder and chief executive of Handspring, figures it will take venture money as its needs grow. Numenta also aims to hold conferences and offer technical support to build a community of developers, a move Dubinsky, Numenta’s CEO, compares to what Red Hat did with Linux. Longer term, she says, Numenta will be like Qualcomm (nasdaq: QCOM – news – people ), profiting largely from licensing its intellectual property.

Careful to preserve his track record but sure he is onto a revolution, Hawkins veers between caution about Numenta’s likely timeline and visions of creating machines far more powerful and able than today’s computers (his approach is so different, he says, that Numenta-type machines will represent a new category.) “We have to walk a line–we don’t want to overhype this, but it’s incorrect to say this is just data mining,” he says. “This is a new fundamental technology for information handling.”

Noting that electronic circuits can scale larger and signal far faster than human neurons, he says, “If you follow this, you will build machines smarter than humans. It’s far off, but think about cosmology, about where the Internet is going, machines that understand our language.”

More here.