How do you know your AI is smart?

Posted by ljmacphee on January 30, 2007 under topics in artificial intelligence | Be the First to Comment

One of the more difficult tasks artificial intelligence creators face, is how do you know it is intelligent? There is not even have an agreement as to what constitutes intelligence in people let alone what constitutes intelligence in a machine.

There is some agreement on what is required for artificial intelligence: pattern recognition; the ability to create or learn, the ability to keep and access stored information; problem solving ability; communication ability; the ability to form intentions i.e. self awareness.

Much progress has been made using neural nets to do pattern recognition. The ability to create, learn, keep and access stored information has been partially done with black board systems and neural nets. Communication is lagging behind the other areas. Some speech recognition has been done with neural nets, but this is pattern recognition, not actual two-way communication. The ability to form intentions is hotly debated. What exactly this means is not clear. What constitutes self-awareness in a machine is also hotly debated and a definition or test has not yet been agreed upon.

It is not yet clear whether creativity is a part of general intelligence or a separate entity. The main traits of creativity are generally agreed to be: a lack of conventionality and the willingness to question status quo; the ability to recognize connections be they similar or dissimilar; and an appreciation and skill in any of the arts. Persistence or motivation is often considered a characteristic as well. It is necessary to have knowledge in the field as well. A lucky guess that is not understood is hardly creative.

It is clear there is such a thing as ‘general intelligence’ or ‘the G factor’. We can all readily identify intelligence, or lack thereof in people we come across. However, it is also clear that intelligence tests measure one’s ability to take tests and education, rather than one’s intelligence. General intelligence remains fixed over the life of an individual. Education grows as a person learns more whether through self education, academia or other methods. Some areas of the brain can be damaged with out harming a persons intelligence, instead only costing the person some memories or skills.

Although psychologists have tested for self awareness by placing dots on the foreheads of sleeping beasts and determining they were or were not self aware if they recognized the dot in the mirror as being on them. It is now clear that self awareness, like intelligence and education is a matter of degree and not a have or have not part of beings.

It is also becoming clear that consciousness is a dynamical ongoing process in the brain. It is not something that can be found in the pieces of the brain but only in the operation of the brain. How a piece of software works can not be determined by dismantling the computer into circuits and chips, nor can consciousness be understood by only looking at pieces of the brain.

The level of intelligence of a species is related to a constant multiplied by brain size divided by body size. Male human brains average 1371cc, female brains average 1216 cc. Normal IQ scores have been documented for brains between 735cc and 1470cc. Before anyone gets confused, remember that it is excess neurons above and beyond what are needed for body maintenance that matters, since male bodies are usually much larger than female bodies more neurons are needed for general maintenance.)

See Also:
The Random Test - AI Depot
The Streeb-Greebling Diaries: Some ingredients of intelligence

Agents, Bots, and Spiders

Posted by ljmacphee on January 29, 2007 under bots, source code, topics in artificial intelligence | Be the First to Comment

‘Bot’ is short for ‘robot’ and refers to s software robot. A bot goes out into the Internet and pulls back data. Bots are used to handle repetitive tasks, to index the web for search engines, by games to be your avatar and by any program that sends out ‘robots’ to perform tasks. Java and Python are the preferred language for them.

A spider is a specialized bot. A spider starts on a given web-page and is restricted to a given domain or set of domains. The spider traverses the area by collecting links on the page and going from link to link.

Information gathering agents are all over the Internet now. They are the spiders that index web pages for you to find. There are bots that traverse databases and pullout information to serve up to users. Since computers do not understand language teaching bots to gather and sort information is quite a challenge. Several techniques are discussed here.

WebMate uses multiple TF-idF vectors each one in a different domain of interest to the user, as well as ‘Trigger Pair’ word searches in documents, and WebMate learns from watching the user. It runs between the browser and the HTTP server monitoring transactions. WebMate learns the classifications rather than have the user select and feed them to the program. The program learns incrementally and changes as the users interests change. The learning algorithm is run whenever the user flags a document as useful.

Information gathering agents may use the SIMS architecture. Each agent is a specialist in a different subject. These agents use KQML as the communication language between them, and LOOM as the knowledge representation language.

One agent is created, then others are instantiated to become experts in different areas of knowledge (flight schedules, hotel locations and rates, etc) and an area of domain (type of database, physical location, etc). A network of these agents is then put together in an acyclic graph.

Link Checker Spider (JAVA)

More information:
Alice
Alpha Works
BotSpot

See also:
Cognitive code develops software personal assistant using Lua
FBI’s bot roast finds compromised computers
AI promoting ‘Flatland’
Personal agents may reach your phone before they reach your computer