New computer model could revolutionize artificial intelligence field

By Katie Landeck

Marsha Gelin/Collegian
Marsha Gelin/Collegian

Hava Siegelmann, an associate professor at the University of Massachusetts computer science department, envisions a newer, smarter kind of computer.

She pictures a computer that is able to think for itself. A machine that constantly evolves as it computes information collected from what it “sees” and “hears” in the world around it using a new type of programming that would mirror the human brain.

She calls it “Super-Turing.”

Super-Turing, according to Siegelmann, is the expansion of the Turing model, which is the model that most personal computers and industrial computers operate with.

Right now, the idea is purely theoretical as no computer has ever been built that operates using the Super-Turing model. But, Siegelmann was recently awarded a grant so that she and a team of scientists at Missouri State University could build the first Super-Turing computer.

“This is the first time I am suggesting doing something really different which is kind of building a new type, a new idea of what the computer is,” she said.

The Turing model – developed by Alan Turing who would have celebrated his 100th birthday this year – accepts finite inputs and outputs. It is limited by its programming which tells it exactly what it can and cannot do.

“There is no way that the Turing model can be creative,” said Siegelmann.

The Super-Turing model would be practically limitless, explained Siegelmann. Its programming would be modeled after the human brain, allowing it to learn.

“The Super-Turing model would have infinite input and output,” said Siegelmann. “The memory from yesterday still affects me today.”

This model, according to Siegelmann, is the hope for artificial intelligence.

Up until this point, everything in the artificial intelligence field has been based on the Turing model. For example, Watson – the Jeopardy-playing super computer – is a Turing model. It has huge databases and intricate programming that enabled it to understand the nuances in a question and generate the best possible answer.

But, it is only as good as its programming.

“You saw that it didn’t learn,” said Siegelmann. “The person said an answer and it didn’t learn that it was wrong. It tried to do it, too.”

If Watson operated as a Super Turing computer, it would have been able to adapt to the game and learn if an answer was right or wrong.

“If you change the rules as it goes or if you tell it this is wrong, it will change itself a little bit,” she said.

Siegelmann stumbled across the idea of Super-Turing in 1993, when she was working on her doctorate in computer science at Rutgers University. Initially, she had set out to prove the theorem that the Turing model was a better model than a neural model for computers.

However, while going through multiple proofs in favor of the Turing model, she kept finding mistakes. She then tried to make her own “really good” proof, but couldn’t do it mathematically.

Then, she changed tactics.

“Without asking my advisor, I went home and said, ‘Well I can’t prove it, so let me try to disprove it,’” said Siegelmann. “And I did.”

After earning her Ph.D., Siegelmann spent the next decade focusing on biological and physical computations, building computer programs that reflected how the brain works.

She found that a lot of her research in this field intersected with her research on the Super-Turing model, and convinced her that is was possible to build a Super-Turing machine.

While the hardware for the new analog computer is being developed at MSU, Siegelmann is developing the algorithms which will be different than what is in the average computer.

“I think it will have two levels of programming. One will be the initial programming which will be very similar and be kind of like what we have,” she said. “And then one will be the other kinds of options of adaptivity allowed.”

Siegelmann imagines this new type of programming will not replace current computers, but be a new type of computing that will be best utilized in robots. She hopes that this type of artificial intelligence will be used to help people with disabilities such as Alzheimer’s patients or the visually impaired.

Katie Landeck can be reached at [email protected]