Neuroscience Pioneer to Receive Top Engineering Award in San Diego

On February 22, Robert Hecht-Nielsen will accept the Outstanding Engineer award, presented annually by the San Diego County Engineering Council (SDCEC). The organization brings together the San Diego chapters of IEEE, American Institute of Aeronautics and Astronautics (AIAA), and many other technical societies.

The award ceremony will be held at the Joan Kroc Institute for Peace and Justice at the University of San Diego. It is one of numerous events in San Diego, including at UCSD, scheduled as part of National Engineering Week 2008.

According to Robert Hecht-Nielsen (pictured above with his cat Zeus, whose responses triggered some of the researcher's theory), thinking is utterly alien in comparison with existing concepts in computer science and information processing. It involves careful, coordinated, deliberate analog control of a group of modules. Just as with the control of muscle groups in movement, thinking involves a graceful, purposeful, analog 'dance' of brain modules. The final result is an intelligent set of conclusions. For example, the (computer-implemented) confabulation architecture illustrated above consists of hundreds of modules (each indicated in the figure by a square - only a few of which are shown) and thousands of knowledge bases (each consisting of all of the knowledge links from the symbols of one specific source module to the symbols of a specific target module; each illustrated in the diagram by an arrow connecting one module to another - again, only a few of which are shown). This architecture allows sets of three consecutive sentences from the same paragraph of a newspaper story to be expressed in terms of their constitutive words (each module on the bottom row has 63 000 neuron groups or symbols representing the most common words and punctuations of English - one symbol is activated in each successive module to represent the corresponding word of the sentence - with the words entered from left to right). Billions of knowledge links (each corresponding to a specific pair of one word or phrase and a second word or phrase that is seen repeatedly) arise as tens of millions of such well-written sentence triples from 1990s-vintage newspaper stories are entered.   Once this architecture has completed this "reading" exposure, and the learning it causes, it is then possible to test the architecture's "intelligence". To do this, a pair of consecutive fresh sentences (i.e., sentences that the machine has not seen during is learning exposure) are entered into the architecture's modules of the first and second context sentences. The modules of the third sentence are then confabulated (illustrated in the figure by a swirling red arrow) to produce a new sequence of words and punctuations. For example, if the two consecutive fresh context sentences (obtained from the Detroit Free Press) entered are: "Several other centenarians at Maria Manor had talked about trying to live until 2 000, but only Wegner made it." and "Her niece said that Wegner had always been a character - former glove model ,  buyer for Macy's, owner of Lydia's Smart Gifts downtown during the 1950s and '60s - and that she was determined to see 2 000." are entered, the word string produced by the confabulation process is: "She was born in the Bronx Borough of New York City." As a second example, if the fresh context sentence pair: "He started his goodbyes with a morning audience with Queen Elizabeth II at Buckingham Palace, sharing coffee, tea, cookies and his desire for a golf rematch with her son, Prince Andrew." and "The visit came after Clinton made the rounds through Ireland and Northern Ireland to offer support for the flagging peace process there." are entered, the word string produced by the architecture is: "The two leaders also discussed bilateral cooperation in various fields." "Clearly, the age of human-level machine intelligence has begun," said Hecht-Nielsen. "What is most striking about this confabulation architecture is its extremely large quantity of knowledge and the effectiveness with which it exploits this knowledge to craft astonishingly intelligent outputs," he added. "All without any traditional 'software', 'rules', or 'algorithms'. This is how neurons think. This is how human and animal 'intelligence' arises. Establishing a complete understanding of this will require a huge new phalanx of research." Hecht-Nielsen envisions that by 2 100, human toil will have permanently ended; he argues that hundreds of billions of confabulation theory-based intelligent machines would do all work that humans choose not to do. "By 3 000, humans and their intelligent machines will have spread throughout the Milky Way galaxy," he said, forecasting that "by the year 20000, humans and their intelligent machines will be intervening to eliminate unnecessary lifeform suffering, strife and tragedy throughout the millions of billions of galaxies of the entire universe."

How thinking works has puzzled scientists for millennia -- from Aristotle's studies on human thought (published in 350 B.C.), to recent computational neuroscience research. Hecht-Nielsen's Confabulation Theory now presents the scientific and engineering communities with what he calls the "first comprehensive and complete theory of cognition," a theory that he believes can shed extensive light on the design of the brain. "Seeing the detailed process of thinking in action," said Hecht-Nielsen, "it is clear that thinking is starkly alien in comparison with past concepts in neuroscience, computer science, and so on."

Hecht-Nielsen postulates in his Confabulation Theory that the human brain's cerebral cortex is divided into about 4,000 separate, small modules; each an individual 'thought processor'. Modules are manipulated in groups, like muscles, to carry out thinking. What thinking involves is taking a starting set of 'assumed facts' (e.g., sensory inputs) and then producing a set of conclusions based upon these starting facts. Billions of learned axonal knowledge links exist within each adult human's cerebral cortex, and often millions of these are used during a single, massively parallel, thought process that takes place in less than a second. The basic information processing operation carried out by each module is called confabulation.

Usually, many confabulations are being carried out in parallel, utilizing millions of knowledge links. The effect of each knowledge link being used is to transmit excitation to specific groups of neurons. During confabulation, those neuron groups with low levels of excitation are slowly shut off, leaving the most excited groups to compete with each other. In the end, each module involved in a thought process has one neuron group active (its confabulation conclusion) and all other neuron groups shut off. [For a more detailed explanation of the theory, see box at right.]

Professor Hecht-Nielsen is Director of the Confabulation Neuroscience Laboratory in the California Institute for Telecommunications and Information Technology (Calit2) at UC San Diego. He is also Adjunct Professor of electrical and computer engineering in UCSD’s Jacobs School of Engineering, where he has taught a popular, year-long graduate sequence on neurocomputing since 1986. This course now serves as an introduction to the neuroscience, and technological applications, of confabulation theory.

That same year, he co-founded HNC Software (originally Hecht-Nielsen Neurocomputer Corporation), one of San Diego ’s early high-tech success stories. HNC went public in 1995 and then merged with similar-sized Fair Isaac Corporation (NYSE: FIC) in 2002. Hecht-Nielsen’s UCSD Confabulation Neuroscience Laboratory is sponsored by the Office of Naval Research.

Hecht-Nielsen received his Ph.D. in Mathematics from Arizona State University in 1974. He is a Member of the Society for Neuroscience, an AIAA Member, an IEEE Fellow, recipient of the IEEE Neural Networks Pioneer Award, and a Member of the Board of Governors of the International Neural Network Society (2007-’09).

He is a founder and member of UCSD’s Institute for Neural Computation, a founder of the UCSD Graduate Program in Computational Neurobiology, and has been honored with the Department of Electrical and Computer Engineering’s Graduate Teaching Award.

Looking to the future, Hecht-Nielsen hopes that his recent book will spark a lot of new “integrative neuroscience research directed at answering the many scientific questions raised by confabulation theory.”

Related Links

Hecht-Nielsen Website  
San Diego County Engineering Council  
Presentation on "Speech Understanding" to Cognitive Computing 2007 

Media Contacts

Media Contact: Doug Ramsey, 858-822-5825, dramsey@ucsd.edu