Scientist predicts digital brain by 2023

It may be possible to simulate a human brain down to the cellular level by 2023, according to Henry Markram, the director of the Brain Mind Institute at the Ecole Polytechnic Federale De Lausanne.

Read this

HP figures out how memristors work, predicts robot use

Read more

Markram made his prediction on Monday, in the Simulating the Brain — The Next Decisive Years keynote speech at the International Supercomputing Conference in Hamburg on Monday. He told attendees that biological scientists are "facing a massive tsunami of data". They need their computing resources to rise to the challenges of simulating a brain from its basic data processing and communications unit — the neuron — upward.

"We'd need about an exascale [one exaflops, 10¹⁸flops] to get to the human brain at the cellular level [by 2023]," Markram said. "To do this you have to do a global integration, you have to integrate all the science from the genes up to the phenotype, database the information and apply reverse engineering to look for rules and patterns."

Great co-ordination is needed, he said, noting biotechnology projects lag behind international physics efforts such as those run by Cern.

2023 is only a realistic date if academic funding models and computing resources continue to develop as anticipated, Markram noted.

Hidden benefits of biology-led computation

However, unlike other fields of computation, where systems become more complex, they become more difficult to design and simulate, the reverse is true in biology.

"From the macroscopic constraints you can specify microscopic detail," he said. "This will make it faster and easier to specify the specific detail in a cell."

In other words, the more biologists learn about the general structures of the brain, the easier it becomes to simulate it at a fine level of detail.

When a brain is eventually simulated at some point in the 2020s, tools will already have been built to allow researchers to conduct experiments on the simulation in real time, he said. The researchers at Lausanne are developing a system to allow global researchers to access both the HPC (high-performance computing) system that will underpin the simulation, and the brain itself.

There is also a plan to parcel the basic task of analysing the cellular structure of the brain out to other computers, in the style of the distributed SETI@Home or Boinc projects, he said.

Factoring into the future of computing

Once built, the simulation should be able to contribute to general-purpose computing and developing fields, such as robotics.

The brain is 30W, a million kilometres of fibres and a thousand trillion synapses. The energy efficiency is incredible, astronomical.

– Henry Markram

"The brain is 30W, a million kilometres of fibres and a thousand trillion synapses," Markram said. "The energy efficiency is incredible, astronomical."

Through studying the deep structure of the brain, the researchers hope to extract high-level principles of computation that will be useful for algorithms that mimic complex processors, such as character recognition, he said.

"One thing to be aware is that even if we can build very intelligent neuromorphic architectures... I think this will be thousands of times simpler than a human brain but thousands of times more intelligent than anything we will have in robotics," he said. "Robotics is still in the dark ages, [because] they do not have a brain."