There was a palpable excitement in the auditorium when Andrew Ribeiro began speaking about artificial intelligence. People were interested in learning about what may be one of the next frontiers.
Andrew is a co‑founder of Knowledge Exploration Systems and co‑organizer of Danbury AI. He studied computer science at Western Connecticut State University. Regarding his interest in artificial intelligence, he explained that his own interests revolve around philosophy, psychology, mathematics, and computer science. “There is no field other than artificial intelligence that brings these subjects together … [in] such a cohesive and exciting manner.”
Andrew defined artificial intelligence (AI) in two ways:
- AI researcher John McCarthy defines AI as “the science and engineering of making intelligent machines, especially intelligent computer programs. It is related to the similar task of using computers to understand human intelligence, but AI does not have to confine itself to methods that are biologically observable.”
- Andrew’s own definition is simply “the investigation of intelligence with mathematics.”
He further broke down AI into two subcategories:
- strong AI that would “have a mind in exactly the same sense human beings have minds”, according to AI researcher John Searle.
- weak AI that “simply mimics a human and does not have internal processing isomorphic to human minds”.
Andrew categorized the uses of AI into three broad areas:
- cognitive science applications, such as machine learning, genetic algorithms, and neural networks
- robotic applications, such as visual perception and navigation
- natural interface applications, such as speech recognition and virtual reality
Andrew made it clear from the outset that his understanding of artificial intelligence (AI) is based on underlying fundamental principles and both pure and applied mathematics. He told us a great deal about the fundamental principles, including:
- the notion of “complex systems”
- “black box intelligence”
For instance, he explained “constructionism” and quoted Seymour Papert:
“From constructivist theories of psychology we take a view of learning as a reconstruction rather than as a transmission of knowledge. Then we extend the idea of manipulative materials to the idea that learning is most effective when part of an activity the learner experiences as constructing a meaningful product.”
More examples of underlying fundamental principles include logic, language, symbolic reasoning, and rules of deduction that produce intelligent behavior.
Regarding mathematics, Andrew described “mathematical modeling” and said it is key to the study of AI. Mathematical modeling is the effort to “find parallels between the broad universe of mathematical abstractions and instances of those abstractions in the real world”. Specifically, given a “real world instance of structure”, one can develop a “space of mathematical abstractions” that models the real world. This is mostly applied mathematics and some pure mathematics. Andrew intentionally avoided showing any actual math in this introductory presentation, but he made it clear that a real understanding of AI would be based on extensive mathematical tools, including linear algebra, calculus, and stochastic systems.
Andrew stressed that an understanding of AI involves multi-disciplinary learning. He mentioned a few examples of connection between AI and other fields:
- Electrical engineering and AI both include the study of “control theory”, which is “the control of continuously operating dynamical systems … using a control action in an optimum manner without delay or overshoot and ensuring control stability.”
- Engineering and AI both use cybernetics, which is “the scientific study of control and communication in the animal and the machine.” (quoted from Norbert Wiener)
- Cognitive psychology and AI both have parallels to behaviorism.
Andrew described neural networks in depth. An “artificial neuron” (a model of a biological neuron) has weighted inputs, a method of aggregating these inputs (including summing, threshold/bias, and activation bias), and output(s). He explained that neural networks are the core building blocks of a biological nervous system; they function in AI as “amazingly useful” mathematical models to “describe the connectivity of distinct things”. He detailed several neural network topologies, with neurons arranged in various ways.
I started this review by writing that audience members were interested in learning about AI because it may be one of the next “frontiers”. It was clear from Andrew’s talk that AI is not a new frontier but rather an already highly developed field!
Andrew Ribeiro’s presentations slides can be found in both PDF and Google Docs form on the DACS Downloads page.