Steve Omohundro, Ph.D.

It has been my privilege to be a scientist, author, university professor, software architect and entrepreneur. My research has explored the boundary between mind and matter and my hope is that deeper understanding will promote compassion and cooperation in the world.

My first company, Om Sonic Systems, designed and built custom analog music synthesizers. At age 16 I went to Stanford University to study physics and learn about the nature of the universe. I also wanted to understand how manipulating symbols could reveal natural truths, so I studied mathematics voraciously. I was a member of the Stanford team which won fourth place in the William Lowell Putnam Mathematical Competition, the best that Stanford had ever done. I learned how to prove formal properties of programs from John McCarthy and had great discussions with Doug Hofstadter as he wrote “Godel, Escher, Bach”. Leonard Susskind advised a friend and me in writing the first two theoretical physics honors theses at Stanford. My honors thesis, “Some Topological Excitations in Physics” explored the use of algebraic topology in understanding physical phenomena. I graduated Phi Beta Kappa from Stanford University in 1980 with Honors and Distinction in Physics and with Distinction in Mathematics.

I went to U. C. Berkeley for graduate school because of the vibrant interaction between the physics and mathematics departments. Allan Kaufman served as my physics adviser and Alan Weinstein as my mathematics adviser. I was privileged to participate in two revolutions in mathematical physics: the understanding of chaotic dynamical systems and the discovery of the geometric Hamiltonian structures underlying all areas of physics. I wrote papers on “A Geometric Approach to the Equations of Plasma Physics” with Alan Weinstein, “On the Global Structure of Period Doubling Flows” with John David Crawford, “Bifurcations and Chaos in the Lorenz Equations Book Review”, and “Geometric Hamiltonian Structures and Perturbation Theory“. I gave many talks including: “Geometry of Averaging“, “Hamiltonian Approach to Perturbation Theory“, “Symmetry, Its Breaking, and Hierarchical Levels“, and “Ten Ideas From Nonlinear Dynamics“. At 560 pages, my dissertation was the longest ever written in the Berkeley physics department and served as the basis for my book Geometric Perturbation Theory in Physics. I received my Ph.D. in physics in 1985 from the University of California at Berkeley. While at Berkeley, I also did work on cellular automata and machine vision, publishing the papers: “Modelling Cellular Automata With Partial Differential Equations” and “Unique Recovery of Motion and Optic Flow Via Lie Algebras” with Peter Blicher.

After finishing my physics Ph.D., I decided that I could make the biggest contribution by focusing on artificial intelligence. Danny Hillis recruited me to his new company Thinking Machines Corporation which was doing research in artificial intelligence and was building a massively parallel computer called the Connection Machine. They had  amazing consultants including Richard Feynman, Marvin Minsky, Jack Schwarz, and Stephen Wolfram (here is a photo of the company a few months after I joined). I worked on machine vision and software and algorithms for the Connection Machine. I developed parallel Connection Machine algorithms for a variety of important computational areas such as linear algebra. Cliff Lasser and I developed the programming language StarLisp which became the first language delivered for the Connection Machine. My talk “Parallel Programming on the Connection Machine” describes a variety of parallel algorithms for the Connection Machine and some novel applications that I was involved with creating.

In 1986, I joined the computer science faculty at the University of Illinois at Champaign/Urbana and co-founded the Center for Complex Systems Research with Stephen Wolfram and Norman Packard. I taught courses in machine vision, algorithms, and graphics and was selected to a list of teachers ranked as excellent by their students. I was one of the first to realize the importance of machine learning in building models for machine vision. I started the Vision and Learning Group and my students and I developed many geometric learning algorithms. I supervised 4 Masters theses and 2 Ph.D. theses. I published “Efficient Algorithms With Neural Network Behavior” which introduced new learning algorithms and used computational geometry techniques to provide dramatic speed-ups. Other papers included “Fast Texture Recognition Using Information Trees” with Darrell Hougen, “Fundamentals of Geometric Learning“, and “How Can Slow Components Think So Fast?“.  In 1988 Apple Computer held a contest to design the computer of the year 2000. Stephen Wolfram and I advised a team of 5 students and we came in first with the paper “TABLET: Personal Computer in the Year 2000“. It is remarkable how many things that we predicted have come to pass. We also wrote the paper “Academic Computing in the Year 2000“. I served on the editorial board of the journal “Complex Systems”.

I wrote the three-dimensional graphics portion of Wolfram Research’s Mathematica program as one of the seven original developers. I developed new algorithms for object-based hidden surface removal, three-dimensional function plotting, and spline-based contour mapping.

In 1988 Jerry Feldman recruited me to the International Computer Science Institute in Berkeley. I developed a variety of novel neural network and machine learning algorithms and with students and colleagues built systems which learned to read lips, control robots, and learn grammars. I published papers on “The Delaunay Triangulation and Function Learning“, “Geometric Learning Algorithms“, “Bumptrees for Efficient Function, Constraint, and Classification Learning“, “How Receptive Field Parameters Affect Neural Learning” with Bartlett Mel, “Building Faster Connectionist Systems with Bumptrees“, “Best-First Model Merging for Dynamic Learning and Recognition“, “Toward a Synthesis of Symbolic AI and Connectionism“, and “Five Balltree Construction Algorithms” and short notes on “Floyd-Steinberg Dithering“, “An Equilateral Triangle Recognition System“, “Bayesian Segmentation of Dot Pictures“, and “Boxtrees for Fast Image Feature Search“. I gave many talks including: “Beyond Symbolic AI“, “Geometric Learning Algorithms“, “Geometric Learning Algorithms for Vision, Robotics, and Graphics“, “Learning and Recognition by Model Merging“.

Subutai Ahmad and I worked on understanding how biologically realistic neural models could use selective attention to solve the variable binding problem which plagues connectionist systems. We published: “Equilateral Triangles: A Challenge for Connectionist Vision“, “A Network for Extracting the Locations of Point Clusters Using Selective Attention“, and “Efficient Visual Search: A Connectionist Solution“.

Andreas Stolcke and I worked on using best-first model merging to learn stochastic grammars from data. We published: “Best-first Model Merging for Hidden Markov Model Induction“, “Hidden Markov Model Induction by Bayesian Model Merging“, and “Inducing Probabilistic Grammars by Bayesian Model Merging“.

Chris Bregler and I worked on using “manifold learning” to build systems that learned complex geometric domains such as reading lips. We published: “Surface Learning with Applications to Lipreading“, “A Hybrid Approach to Bimodal Speech Recognition“, “Nonlinear Image Interpolation using Manifold Learning“, and “Nonlinear Manifold Learning for Visual Speech Recognition“.

I led an international team in developing the object-oriented programming language Sather which was featured in O’Reilly’s History of Programming Languages. The language developed over time as may be seen from the specifications for Sather 0.2, Sather 1.0, and Sather 1.1 with David Stoutamire. I published papers on “Sather’s Design“, “The Differences Between Sather and Eiffel“, “The Sather Language and Libraries” with Chu-Cheow Lim, “Sather Provides Nonproprietary Access to Object-Oriented Programming“, “CLOS, Eiffel, and Sather: A Comparison” with Heinz Schmidt, “The Sather Programming Language“, “Engineering a Programming Language: The Type and Class System of Sather” with Clemens Szyperski and Stephan Murer, “The Sather 1.0 Implementation“, and “Iteration Abstraction in Sather” with Stephan Murer, David Stoutamire, and Clemens Szyperski. I gave many talks describing and promoting Sather around the world.

I served on the editorial board of “Eiffel Outlook” and on the program committee of the 1991 conference on Neural Information Processing Systems.

In 1995 I joined the NEC Research Institute in Princeton to do research on artificial intelligence. I wrote papers on “Family Discovery“, “Learning Visual Motion Models for Lip Reading” with Chris Bregler, “Model Merging for Hidden Markov Model Induction” with Andreas Stolcke, and  “Probabilistic Models of Verbal and Body Gestures” with Chris Bregler and others. Peter Blicher and I designed and wrote the RCL C++ library for probabilistic modelling and machine vision.

A group of us developed the PicHunter image database search engine. We published “Target Testing and the PicHunter Bayesian Multimedia Retrieval System“, “PicHunter: Bayesian Relevance Feedback for Image Retrieval“,  “Toward Optimal Search of Image Databases“, and wrote a patent on “Multimedia Database Retrieval System Which Maintains a Posterior Probability Distribution That Each Item in the Database is a Target of a Search“.

I served on the program committee for the 1997 International Workshop on AI and Statistics, the 1996 and 1995 conferences on Neural Information Processing Systems, the 1996 International Conference on Pattern Recognition, and the 1996 Conference on Computational Learning Theory. I gave numerous lectures including talks on “Nonlinear Manifold Learning for Visual Speech Recognition” and “The Family Discovery Learning Algorithm“.

In 1997 I returned to Silicon Valley to try to have a bigger impact with my work. I did trainings in entrepreneurship and business development and consulted for InterTrust Technologies Corporation on digital watermarking, tamper-resistant software, and formal digital contracts, for Xerox Palo Alto Research Center on Bayesian information retrieval, for Fuji-Xerox Palo Alto Laboratory on automated video conferencing, and for Ask Jeeves Inc. on grammar learning. I started Olo Software to work with a number of startup companies which unfortunately haven’t survived. I was a co-founder of Molecular Objects which aimed to use innovative computational chemistry to develop self-assembling nanotechnology but it also did not survive. I was pleased that Scientific American’s “Mathematical Recreations” devoted a column to a game theory puzzle I invented.

In 2003, I started Self-Aware Systems to work on self-improving systems, a new approach to artificial intelligence that I had been thinking about for some time. As the work progressed, I realized that it could have important social consequences. I gave talks and wrote papers on this aspect of the work to stimulate a broader discussion of important issues. I joined the advisory boards of the Singularity Institute for Artificial Intelligence and the Lifeboat Foundation. I wrote papers on “The Nature of Self-Improving Artificial Intelligence” and “The Basic AI Drives“. I gave talks on “Creating a Cooperative Future“, “Self-Improving AI and the Future of Computing“, “AI and the Future of Human Morality“, “Co-opetition in Economics, Biology, and AI“, “Self-Improving AI and Designing 2030“, “Evolution, Artificial Intelligence, and the Future of Humanity“, and “The Science and Technology of Cooperation” among others.

In 2009, Self-Aware Systems launched the Semantic Computing Initiative and the Cooperative Technology Initiative. These aim to create immediate value and to lay down foundations that will be needed for future human benefit.