WebMind
WebMind
Dateline: 02/09/99
Intelligenesis, a New York based company led by mathematician Ben Goertzel, is presenting a new computer system that promises to end the infancy period of old-fashioned AI systems. Mind you, their ideas are so novel that a dose of profound skepticism is expected at first. On the other hand, they seem so obvious, intuitive and easy to understand. Kind of "deja-vu" effect - only that I can't remember of anything similar.
For a start, let's make a clear definition what WebMind is and what it is supposed to accomplish: it is a complex, self-organizing system, the goal of which is to understand data stored on computer networks, create its own information using this data as a starting-point, and answer questions regarding this information. OK, you've heard similar announcements before. What makes this particular system so different?
To answer this last question, let's first take a look at the current state of our information technology. There is so much information stored on corporate and private computers, database systems and document archives. The only way to extract the interesting information out of it is to use some of the data mining tools, Internet search engines, etc. Long-term memory is "detached" from processing, and this is not how human mind works. There is no way for the traditional systems to generate their own knowledge and answer questions based on the knowledge they generate.
This is where WebMind gets in, breaking this artificial distinction between static data repositories and dynamic search and analysis tools. Recently, I was trying to describe it to my colleagues as a part neural network, part semantic network, and a part genetic algorithm. No wonder nobody understood what I was talking about. Try to imagine a network of software objects ("nodes"), linked to other nodes that can contain either "raw data" such as text or more abstract concepts (so they contain only links to other nodes). This could remind you of neural or semantic networks, but there is more: intelligent software agents are essential part of the process of creating and destroying links as part of the learning process. Semantic networks generally have no learning dynamics, they're just data structures. Neural networks are too fine-grained to be used as data structures for storing real data like texts or data files - instead they're generally used as fancy problem solving techniques. Webmind is in between
semantic networks and neural networks, and hence works well as both a real-world
data structure and an effective learning system.
Learning process goes as follows: when a node receives attention (it can be a TextNode, representing particular information in a text, or more abstract node at "higher level" corresponding to more general topics), it spreads some of its attention to the "meaning-sharing" nodes it knows about by checking its link table. This activation spreading is done by sending agents that are allowed to create new links between it and other nodes. After a certain period of time, the nodes with the highest activation are detected, and returned as the answers to the query. This way, WebMind internal network is constantly refining its structure, which is inseparable from its complex dynamics.
This brief description of Goertzel's mathematical model of mind called the "Psynet model" does not aim to capture many of its very fine points. More elaborate discussion is clearly beyond the scope of our weekly feature articles. If you want to learn more about it, links at the end will give you plenty food for thought. There are just a few more fundamental thoughts that we should tackle here...
One of the key concepts behind the whole project is "compressed" to one simple phrase: The network is the computer is the mind. Although it doesn't aim to emulate human intelligence in an artificial way (maybe the crucial mistake in many earlier attempts to create "real" AI), there are several features of our brains that should be simulated. Brain is "network-based" collection of very heterogeneous systems, heavily involved in communication with each other. Looks similar to the modern computer networks? Exactly. Just supply network-computing hardware base to the whole idea, and you'll have a natural intelligence for the new digital world.
Authors say that it can serve as a search engine, as a Web visualization tool, a data mining engine... There are probably many more areas where WebMind could help. Its ability to extract meaningful business intelligence from text-based data for use in decision support systems is what attracted big Wall Street players. However, there will be no publicly available software releases this year. Intelligenesis is currently pursuing partnerships with major financial firms and Internet information providers. The main ideas is to have WebMind inside most of the future networked information applications.
What's next? Sun recently revealed Java Jini platform, promising that we'll be able to plug devices into network connections without thinking about configurability or device drivers. Jini could gather world's total computing power as one big distributed network of Java processes. The long-term vision underlying Webmind, on the other hand, is that of the world's total computing power as a single global brain, realized as a distributed, self-organizing network of Java processes. Jini will give users unprecedented access to tools and information, but this could create an overload of data and data processing services even greater than that with which computer users are currently presented. In order to render this abundance of information usefully navigable by humans, a network-savvy intelligence will be required. No matter if WebMind serves as Jini Service or Jini serves as mechanism for WebMind I/O, they fit perfectly as two different pieces of the same ambitious, futuristic puzzle.
Similar site(s) of interest: