Former NASA Chief Develops Brain-Like Chips

KnuEdge Chief Executive Dan Goldin, NASA Administrator from 1992-2001, 

at an agency symposium in 2010. 

PHOTO: NASA/BILL INGALLS

Dan Goldin’s startup, KnuEdge, has been working in secret for 10 years on a new kind of computing that mimics the human brain

Dan Goldin spent nine years as chief of the National Aeronautics and Space Administration, overseeing complex projects like the international space station. Now he’s preparing for another challenging launch: a startup that has been working in secret for 10 years on a form of brain-like computing.

His San Diego-based company, KnuEdge, has developed an unusual processor chip and related hardware and software, aiming to bring dramatic speed improvements to tough chores like finding patterns in images, sounds and financial data. KnuEdge is disclosing its plans for the first time Monday, along with a parallel effort in software to help recognize speech in noisy environments.

Many companies are working in related areas, often loosely categorized under the term artificial intelligence. Most are using software that runs on time-tested varieties of computer hardware, not designing chips from scratch.

But there is a widening recognition that conventional processors may not be the best tools for all computing jobs. Nor are they increasing in speed as quickly as they once did, in part because of diminishing returns from miniaturizing circuitry under the pattern known as Moore’s Law.

In one high-profile effort, Google Inc. recently disclosed it is using an internally designed chip for chores related to machine learning, a branch of artificial intelligence, along with chips from Intel Corp. and graphical processing units from Nvidia Corp. International Business Machines Corp. also opted to design its own processor for specialized tasks that emulates aspects of the brain.

Mr. Goldin, who led the space agency from 1992 to 2001, had a similar inspiration. He started by studying with Nobel laureate Gerald Edelman, a prominent theorist on the workings of the brain who died in 2014.


“I was sick and tired of hearing the world is ending, Moore’s Law is dead,” he said. “We need to bring in neurobiology.”

Any kind of chip startup is a rarity these days, as venture capitalists prefer investments that take less time and money to produce a return. The 75-year-old Mr. Goldin, KnuEdge’s chief executive, said he managed to raise $100 million from individuals with the patience to wait for a payoff. KnuEdge hired expert talent slowly and selectively, he said, but the company now has 100 employees.

Conventional chips outperform humans at chores like high-speed mathematics and sifting through large databases. However, humans are far better at tasks like distinguishing one face or voice from another, partly because of the brain’s parallel structure.

Where a typical Intel chip might have one to a dozen or so processing cores, Mr. Goldin estimated that people have a couple of hundred billion neurons—the cells that process and transmit information. Each may be able to communicate to 10,000 to 100,000 others dispersed through the brain, he said.

KnuEdge’s first chip has 256 processor cores, each of which can be working on a different program—unlike some other chips with many cores that execute the same kind of instruction at once, Mr. Goldin said. It also developed a companion communication technology that can link up to 512,000 of the chips at extremely high speed.

“It is the ability to scale to unprecedented levels that makes us different,” Mr. Goldin said.

Mr. Goldin isn’t disclosing customers that have received the first circuit boards containing the KnuEdge chips.

But he expects financial-service firms may be particularly interested, for purposes such as analyzing patterns of transactions to prevent fraud. KnuEdge expects to begin selling boards and chassis based on the technology near the end of the third quarter.

Few outsiders have seen the technology in action. But Jim McGregor, an analyst at Tirias Research briefed on the company’s plans, said he expects such novel chip designs will play an important future role in tasks like machine learning.

“I think Dan is a pioneer of the inevitable future architecture trends in computer engineering,” said Larry Smarr, director of the California Institute for Telecommunications and Information Technology, which expects to include KnuEdge’s hardware in a lab that will study pattern-recognition technology. “I think all major companies will be doing this.”

Don Clark

wsj.com

Google has developed a 'big red button' that can be used to interrupt artificial intelligence and stop it from causing harm

The Future of Humanity Institute, University of Oxford

Stuart Armstrong is a philosopher at the University of Oxford and one of the paper's authors.

Machines are becoming more intelligent every year thanks to advances being made by companies like Google, Facebook, Microsoft, and many others.

AI agents, as they're sometimes known, can already beat us at complex board games like Go, and they're becoming more competent in a range of other areas.

Now a London artificial-intelligence research lab owned by Google has carried out a study to make sure that we can pull the plug on self-learning machines when we want to.

DeepMind, bought by Google for a reported 400 million pounds — about $580 million — in 2014, teamed up with scientists at the University of Oxford to find a way to make sure that AI agents don't learn to prevent, or seek to prevent, humans from taking control.

The paper — "Safely Interruptible Agents PDF," published on the website of the Machine Intelligence Research Institute (MIRI) — was written by Laurent Orseau, a research scientist at Google DeepMind, Stuart Armstrong at Oxford University's Future of Humanity Institute, and several others.

The researchers explain in the paper's abstract that AI agents are "unlikely to behave optimally all the time." They add:

If such an agent is operating in real-time under human supervision, now and then it may be necessary for a human operator to press the big red button to prevent the agent from continuing a harmful sequence of actions — harmful either for the agent or for the environment — and lead the agent into a safer situation.

The researchers, who weren't immediately available for interviewing, claim to have created a "framework" that allows a "human operator" to repeatedly and safely interrupt an AI, while making sure that the AI doesn't learn how to prevent or induce the interruptions.

The authors write:

Safe interruptibility can be useful to take control of a robot that is misbehaving and may lead to irreversible consequences, or to take it out of a delicate situation, or even to temporarily use it to achieve a task it did not learn to perform or would not normally receive rewards for this.

The researchers found that some algorithms, such as "Q-learning" ones, are already safely interruptible, while others, like "Sarsa," aren't when they're off the shelf, but they can be modified relatively easily so they are.

"It is unclear if all algorithms can be easily made safely interruptible," the authors admit.

srf

University of Oxford philosopher Nick Bostrom

DeepMind's work with the Future of Humanity Institute is interesting: DeepMind wants to "solve intelligence" and create general purpose AIs, while the Future of Humanity Institute is researching potential threats to our existence.

The institute is led by Nick Bostrom, who believes that machines will outsmart humans within the next 100 years and thinks that they have the potential to turn against us.

Speaking at Oxford University in May 2015 at the annual Silicon Valley Comes to Oxford event, Bostrom said:

I personally believe that once human equivalence is reached, it will not be long before machines become superintelligent after that. It might take a long time to get to human level but I think the step from there to superintelligence might be very quick.

I think these machines with superintelligence might be extremely powerful, for the same basic reasons that we humans are very powerful relative to other animals on this planet.

It's not because our muscles are stronger or our teeth are sharper, it's because our brains are better.

DeepMind knows the technology that it's creating has the potential to cause harm.

The founders — Demis Hassabis, Mustafa Suleyman, and Shane Legg — allowed their company to be bought by Google on the condition that the search giant created an AI ethics board to monitor advances that Google makes in the field.

Who sits on this board and what they do, exactly, remains a mystery.

The founders have also attended and spoken at several conferences about ethics in AI, highlighting that they want to ensure the technology they and others are developing is used for good, not evil.

It's likely that they will look to incorporate some of the findings from the "Safely Interruptible Agents" paper into their work going forward.

Sam Shead

businessinsider.com