"Artificial intelligence is still
science fiction" says an article recently published in the online edition of the Orange County Register. It describes how enthusiasm for a real "thinking machine" has fizzled over the past 30 years, between "2001: A Space Odyssey" and "A.I." Michael Pazzani, professor at the University of California, Irvine's Artificial Intelligence Lab, says that "... we're down to incremental, evolutionary steps. We're tackling smaller problems than the thinking machine." It doesn't mean that AI research hasn't produced real-world results: Pazzani's new company, AdaptiveInfo, is an excellent example of this practical approach. However, quite a few researchers simply refuse to admit the limitations of the current technology. This article will give a brief overview of projects aiming to build intelligent machines, capable of communicating in natural language and acquiring new knowledge.
Artificial Intelligence NV (Ai), an international company split between Boston and Tel Aviv, claims a breakthrough with a computer that has learned to talk like a toddler. They recently passed the first important milestone when Hal successfully finished an adapted Turing Test for 15 month olds. A child development specialist was given transcripts of Hal's conversations, and declared him a healthy, normal little boy. She wasn't told that Hal is a computer program running on a regular Windows PC. The philosophy behind the whole project borrows from Bayesian statistical analysis, predicting what output will be rewarded based on the computer's prior experience. Contrary to the traditional approach of using a set of hardwired rules and a vocabulary database to approximate human conversation, researchers at Ai follow the Turing's idea of the child machine. This behaviorist approach involves just a simple set of learning algorithms and develops speech by talking to humans. It focuses on reinforced training: language is regarded as a skill that is similar to any other behavior, so reinforcement, imitation and successive approximations are required to mature performance.
The Cyc project was started by AI pioneer Douglas Lenat at MCC (Microelectronics and Computer Technology Corporation) in Austin, Texas as a research project in 1984. At that time the project was estimated to be completed in 10 years at a cost of approximately 50 million dollars. The approach used to capture common sense information for the project was to acquire all of the knowledge - both implicit and explicit - in a hundred randomly selected articles in the Encyclopaedia Britannica. Lenat's method is characteristic for the branch of AI commonly named GOFAI - or Good-Old-Fashioned-Artificial Intelligence. It is also in direct contrast to the approach taken by another famous AI researcher and his former student, Rodney Brooks, who is developing embodied intelligent machines - take Cog for an example - mainly based on a sensory input and without complicated methods for planning and knowledge representation. Cycorp recently set up an independent organization, OpenCyc.org, to disseminate and administer the open version of their technology. The premier release of OpenCyc will offer 6000 concepts - an "upper ontology" for all of human consensus reality - and 60000 assertions about those 6000 concepts. Besides that, it will include a compiled version of the Cyc Inference Engine and the Knowledge Base Browser, a suite of tools for rapidly extracting knowledge from domain experts, a specification of CycL and Cyc API, etc.
WebMind, Inc. (aka Intelligenesis) has spent the last 3 years developing the Webmind AI Engine, "the world�s most advanced AI system", intended to, when complete, display true human-level intelligence. It was scheduled for completion in mid-2002, but unfortunately the company announced bankruptcy earlier this year when a key investor abruptly canceled an expected multimillion-dollar round of funding. WebMind has already developed a number of Internet products providing unique functionality, deployed on a variety of public sites and corporate intranets. A substantial percentage of the development team is still working: you can reach them through the Real AI Institute. An excellent essay on the history of WebMind Inc., written by Ben Goertzel in early April 2001, explains the reasons of its unfortunate demise.
<Starlab, another private blue sky research lab in Brussels, Belgium, Europe, went bankrupt in June. The reason was almost the same as in the WebMind's case: a vital investor pulled out of a bridge loan agreement at the last minute. Starlab was the home for prof. Hugo de Garis and his Starbrain - artificial brain project. The original aim of the "CAM-Brain Project", as stated at its beginning in 1993, was to build an artificial brain with a billion artificial neurons by the year 2001, using evolved cellular automata (CA) based neural circuit modules. In reality, this number was adjusted to 75 million neurons and 64,000 modules.
Several large scale AI projects harvest the power of the Internet to gather information from millions of users. The Mindpixel Digital Mind Modeling Project claims to be the world's first program attempting to become artificially conscious by talking to Internet users. Its goal is to collect and validate 1 billion MindPixels - simple true or false statements - over the next 10 years. Its founder, Chris McKinstry, thinks that quality control is crucial for such large-scale distributed projects. His system starts by asking users for a mindpixel, and continues with a request for validation of 20 statements entered by other volunteers. The validation process ensures the correctness of the provided information.
Another similar effort, Open Mind Commonsense, enable large number of people to work together, giving computers the millions of little pieces of ordinary knowledge that constitute "common sense". Everyone has common sense, so everyone could participate. Non-expert users answer simple questions, fill in the blanks, describe the relation between two words, describe pictures, all in order to build a database of common sense facts. Open Mind Web site now encourage visitors to teach a computer to recognize speech or handwriting.
The archetypal idea of the "global brain" - an emergent, distributed worldwide intelligence - is gaining more popularity these days. An earlier article on this topic also describes a Web system called the Principia Cybernetica Web that continually rebuilds the links between its pages, adapting them to users' interests and actions.
Natural language understanding remains one of the toughest obstacles in human-computer interaction. Two smaller companies recently launched Natural Language Operating Systems, trying to break the traditional paradigm of today's operating systems by allowing users to have a natural conversation with their machines using spoken language. Verbal OS promises to provide an open platform for developing interactive applications that use Natural Language Processing (NLP), Voice Recognition (VR), Text To Speech (TTS), and similar technologies. Contrary to some specialized products that are restricted to a specific knowledge domain, Verbal OS promises to grow and match the user's needs by simply installing the additional packages. <Brainhat, on the other hand, is an NLP executive that can dispatch tasks, handle natural language events, evaluate ideas, and ask and answer questions. Because Brainhat is knowledge representation-based, it can digress within a conversation and answer questions outside the script.
While we are at this topic, it is interesting to see the most innovative applications of AI today. The Thirteenth Annual Innovative Applications of Artificial Intelligence awards have just been announced, covering a number of different AI techniques including natural language, case-based reasoning, cooperating multi-agents,
planning, and more. Also, Artificial Intelligence NV just announced the global
Learning Machine Challenge to promote original research in artificial intelligence. The goal is to create a computer program that can win at a game without knowing either the game's rules or the identity of its opponent. The creator of the winning program will win a $2000 prize. The deadline for entries is October 30, 2001, and more information about will be available from their Web site.
