Distributed computing systems consist of many autonomous computers that communicate through a computer network, or of many autonomous processes that occur on the same computer. To be considered autonomous, units on a distributed system have to consist of separate entities with their own local memory and must communicate via message passing. Parallel computing is very similar to to distributed computing in that many calculations are simultaneously executed as part of a larger function. Parallel computing, however, is much more controlled, and may merely consist of multiple cores controlled by the same central processing unit.
Distributed computing systems deal with a number of issues which require that they are flexible and adaptive. Autonomic distributed computing systems seem to be filling in where artificial intelligence (AI) research left off. After years of developing highly advanced robots and AI systems that act human, it has become very clear that acting human and acting independently are very different things for computers.
This shouldn’t really come as a surprise. After all, people came into being in a completely different way than computers did, and have completely different motivations. Do computers have feelings? Why should they?
There is something of a conceptual duality between AI machines which are programmed to mimic human behavior, and programs which are programmed to think autonomously. Watson (IBM’s AI Jeopardy! champion) showed a high degree of exploratory creativity by being able to process massive amounts of cultural information in a human-like way. It had to zero in on a few essential key words, executed thousands of language analysis algorithms in parallel, then would buzz in when the possibilities had been narrowed down to an answer for which it was sufficiently “confident.” Watson may be considered a highly developed example of artificial intelligence according to his mastery of human-like thought and language.
Contrast AI machines like Watson, which process information in a predefined manner, and autonomous computing methods, which may rewrite their underlying software or independently decide their own directives. Watson did not in fact learn or change anything in its underlying programming. The Watson project did show that super-computers are getting so powerful that they are able to process information quickly enough to result in human-like output at exceptional speeds.
In 2001 IBM started the “Autonomic Computing Initiative” which was geared towards making highly complex networked systems more self-stabilizing and self-managing. They aim to make distributed system which can automatically configure internal components, discover and correct faults, optimize functioning by monitoring and controlling resources, and engage in proactive identification and protection from arbitrary attacks.
So while traditional AI frameworks resulted in machines which are little more than tools or puppets, dynamic and self-regulating systems are arising which can autonomously act or produce content. It may be awhile before we see a digital entity that we recognize as being self-aware, yet computers that think for themselves are already among us.
The potential for self aware distributed computing seems awesome, but I’m trying to think about how systems like SETI@home fits into this paradigm. Is that the sort of system that your referencing, or is that different in that it’s just employing end user machines for raw computational power and not really having machines act very independently?
Depending on one’s definition of “self-aware,” a self-aware distributed computing system doesn’t seem likely in the foreseeable future (unless you believe the technological singularity is near). SETI@home follows the structure I’m talking about, but autonomic distributed systems are a lot more dynamic. SETI@home is more akin to a massive parallel processor, while creative systems would need to incorporate machine learning and self-management.