New AI project aims to push beyond neural networks

An ambitious new effort called the Thousand Brains Project aims to develop a new AI framework that its founder says will work on the same principles as the human brain — but will be fundamentally different from the underlying principles of deep neural networks that dominate artificial intelligence today. With funding from the Gates Foundation, the open source initiative aims to collaborate with electronics companies, government agencies and university researchers to explore potential applications for its new platform.

In today’s artificial neural networks, components called neurons are fed data and cooperate to solve a problem, such as recognizing images or predicting the next word in a sequence. Neural networks are called “deep” if they possess multiple layers of neurons.

Deep neural networks currently match or beat human performance in many tests, such as identifying skin cancer and playing complex games, however, they are plagued by a host of problems. For example, as they grow in size and power, they become more power-hungry—to train OpenAI’s GPT-3, a 2022 Nature The study suggested that the company spent $4.6 million to run 9,200 GPUs for two weeks. Neural networks also often prove unstable, with slight changes in the data they receive leading to wild changes in results. For example, previous research found that changing a single pixel in an image can make an AI think a horse is a frog.

To overcome these challenges, the Thousand Brains Project aims to develop a new artificial intelligence platform by reverse engineering the neocortex, which makes up about 80 percent of the mass of the human brain.

“Today’s neural networks are based on basic neuroscience from 80 years ago. We’ve learned a lot about neuroscience since then, and we want to use that knowledge to advance AI,” he says. Jeff Hawkinswho co-invented the Palm Pilot in the 1990s. Hawkins is the co-founder of AI company Numenta in Redwood City, California, which launched the Thousand Brains Project in June 5 at Stanford University’s Human-Centered Artificial Intelligence conference.

The Thousand Brains Targets Project

The name of this project is inspired by the structure of the neocortex; it is composed of thousands of so-called cortical columns, each divided into multiple layers of neurons. “The human brain has about 150,000 cortical columns, and each is essentially its own learning machine.” Hawkins shows IEEE Spectrum.

Deep networks essentially generate a single model of the world, processing data step by step from simple features to complex objects, Numenta researchers have argued. In contrast, the company “The Thousand Brains Theory of Intelligence” proposes that the brain’s many cortical columns generate multiple maps of the world, as if each human brain were actually thousands of brains working in parallel.

Jeff Hawkins says the Thousand Brains Project offers a way forward for AI.Numenta

“Once we learn how to build a cortical column, we can build as many as we want,” Hawkins says.

The project aims to mimic this neuroscience structure in AI with many cortical column-like units that can each perform a sensorimotor task, such as operating a robotic finger. These units can then communicate with each other using connections that are very similar to the long-range connections seen in the neocortex. Hawkins believes this modular structure will make his approach easily scalable.

“The human brain grew very quickly in evolution by replicating the cortical column many times, and we hope we can do the same,” says Hawkins.

The role of sensorimotor learning

The project is also based on the role of the neocortex in sensorimotor learning. While deep neural networks currently learn from giant static databases, the neocortex learns in a dynamic way: it perceives its surroundings using the senses, explores how things work using body movements, and builds models of the world from this feedback. sensory as well as motor.

Hawkins argues that this difference between AI strategies is profound. Creating and labeling the datasets from which deep networks learn is a costly, laborious endeavor, and these systems are unable to continuously learn from new data; instead, they must be retrained on the entire database. In contrast, the neocortex is able to actively learn and can quickly adapt to new information.

“We can build machines that function like the neocortex in terms of sensorimotor learning, and they’re essentially robotic,” Hawkins says. “I think our work is not only the future of AI, but also robotics.”

In addition, the project is developing AI based on the support of the neocortex in reference frames. In the mammalian brain, the so-called place the cells help with encoding memories of locations and grid cells help map out locations in space. The neocortex uses these frames of reference to store and understand the continuous stream of sensorimotor data it receives.

“The way the brain structures data—in 2-D and 3-D reference frames—replicates the structures of objects in the real world,” Hawkins says. “When you look at deep networks, they don’t fundamentally understand the world, which is why if you change just one small feature of an image, they often don’t recognize it. In contrast, reference frames can help the brain understand how its models of objects may change under different conditions.

Potential applications for this new AI platform could include sophisticated computer vision systems that can use multiple cameras to understand what’s happening in scenes, or advanced touch systems to help robots manipulate objects, says Hawkins. “The Gates Foundation is interested in sensorimotor learning for global health. Think of ultrasound, which moves a sensor through space to build a pattern, such as an image of a fetus. This is basically a sensory-motor problem.”

The Thousand Brains open source project is developing a software development kit so others can build on its work. The initiative also pledges not to claim its patents related to the Thousand Brains approach.

Funding from the Gates Foundation

The Gates Foundation is providing the Thousand Brains Project with a minimum of $2.69 million over two years. (The Gates Foundation declined to comment for this article.) “We also hope to announce agreements with government agencies around the world soon,” Hawkins says.

The project aims to create a complete software version of a cortical column. It will then connect multiple units for a complex process, such as seeing or hearing. “Then we want to cross modalities — say, bring vision and touch together — and ultimately build hierarchies, with a model of the world composed of objects, which are composed of objects, and so on,” Hawkins said.

Although the Thousand Brains Project is focused on building software, it is collaborating with researchers such as John Shenprofessor of electrical and computer engineering at Carnegie Mellon University in Pittsburgh, who is designing devices based on the Thousand Brains concept.

“Neuroscience research has made significant advances in understanding how the brain works and how it is constructed since the first neuromorphic brain-mimicking device was created,” says Shen. “What we’d like to do is take the best of what we know now about neuroscience and combine it with the best of silicon to make a new kind of computer. We’ve really embraced the theory of a thousand brains, and this summer we’re talking to chip companies to see if anyone has any interest in partnering with us.”