Through many Internet discussions there is a usual amount of terminological confusion and fuzziness on part of Artificial Intelligence (AI), Semantics, and Web 3.0 understanding. It would be normal some time ago but not today. Now we come in semantic age and we need to define terms clearly; otherwise semantics will be lost completely. Confucius wrote, “If names are not right, words are misused. When words are misused, affairs go wrong.”

Artificial Intelligence

Tim O’Reilly is very suspicious and concerned about Web 3.0 all being about AI (see discussion http://radar.oreilly.com/archives/2007/10/web-30-semantic-web-web-20.html). Nova Spivack as a great practitioner and Web 3.0 insider defended his position by clearly indicating that we are not doing AI in Web 3.0, but ‘semantics plus’, where ‘plus’ is something which is not semantics and not AI. On another side, there are enthusiastic groups expecting Semantics in AI and Web 3.0.

So, are you confused?

Von Neumann considers a language as accidental. It means the fact that the leg is a leg and the hand is a hand and not vice versa is purely accidental. Terms can have a history during which the meaning may change accidentally as well.

Alan Turing proposed criteria to judge AI: if human cannot recognize he is talking to machine, which he really talks to, than it is an AI system. This criterion is not perfect, since it does not specify a human. Ordinary user can be easily tricked to think that he is talking to the human, while he is really talking to the machine, what was done many times. Real AI for skilful user has been never demonstrated and does not exist yet.

So, what are we talking about when mentioning AI? Oh, it is a bunch of methods, which were implemented in various domains and did not produce AI, but (!) many of them are very promising, useful and widely used; f.e., when you are driving and talking over the cell phone, do you know that the cell handover may be done with a help of AI Fuzzy Logic?

AI has an analogy with Alchemy. Alchemists were trying to find the method of gold production from other metals. They have discovered a lot of useful things: metal alloys, perfumes, acids, etc., but not gold. And whatever they have managed to produce we do not call gold.
AI did not produce “AI” and why we continue to call it AI? It is obviously, because of accidental nature of language.

For us AI grounds on two things:

1. Original intent and easily understandable idea, since everybody is having experience with human intelligence and we have Turing’s criteria for this.
2. Collection of sophisticated methods, which may (or may not) contribute to ultimate AI, but they are not critical methods to produce AI, a real one.

Originally if the method may contribute to AI we classify it as AI method, but how to do it without clear criteria? Is sophistication of the method decisive to relate method to AI? In most contexts AI is a collection of methods.

When Tim concerns that we are doing too much of AI in Web 3.0, Nova gives clear answer that we are not doing “usual” AI (defined in 2.). Now the question is: What about original intent?

Are we doing it? Where is a hope?

You know why alchemists did not succeed in several centuries of efforts? They had no understanding of what the gold is. Nuclear physicists without any intent to produce gold easily pointed to nuclear reaction, which produces gold and they gave commercial evaluation of the method with clear understanding of what the gold is.

AI did not reach the main objective because of emphasis on the logic and missing the language semantics. Semantic Technology is placing semantics into the centre. It provides a better position to achieve AI objectives. Now let us state Intelligence with a simple formula:

Intelligence = Language Semantics + Motivation + Emotions

We should have in mind: motivation grounded on emotions. Well known AI postulate: no emotions no AI.

The most challenging question is in what form Semantics can make machine to be intelligent and the current Semantic approaches do not have an answer to this challenge.

Motivation

Since machine cannot have independent emotional motivation, it cannot be intelligent in the human sense; the Artificial Intelligence is rather historical term and it is nonsense in the straight meaning. More correctly we should talk about Complementary Intelligence (CI).

Cutting emotions from the formula we can state:

Complementary Intelligence = Language Semantics + Motivation

This is strange too. People are asking what would motivate machine and how to model motivation. Really important question is a bit different: do we want to live with independent artificial creatures, uncontrolled by humans, having their own objectives and at least as capable as people but in some respect exceeding them? Motivation of Artificial System is not a game it is a reality with a possible threat. Having so many problems in this world we are just not ready to play a God and create another independent intelligence with potential risk to ourselves.

Expert Systems (ES) were first baby steps of AI and were socially unsuccessful. ES development required experts to cooperate in the creation of ES. Experts were feeling the job security threat and were not willing to cooperate. Expert systems mutated into business rule systems, policy engines, etc. to be less on a threat side.

Most people certainly do not want real AI, hence we do not want to model motivation regardless whether it is possible or not. On other hand, software development virtually has no progress for decades (compare to hardware) and remains pathetically inefficient. The issue of addressing CI is an issue of productivity. There is motivation to create human intellectual helper and make people free of much of a non-creative programming work and duplication of efforts.

Getting to the language part of AI equation will make computers much ‘smarter’, but in full control of humans. However, what it means to put the language into computer? We have natural language parsers, speech generators, analyzers, electronic dictionaries, encyclopaedias, and even machine translators – everything is in the computer. And guess what we are missing? Of course the answer is Semantics.

Collective Intelligence

Excluding motivation, all other components of our Intelligence are collective in nature. Our individual language, knowledge and skills are coming from outside that was always true and have nothing to do with Internet. What Internet is bringing is new environment, where this intelligence can get new impulses for the development, can be accessed in new fashion.

Important thing to notice when praising Internet is that there is some negative part of it as well. Ease of information generation brings information pollution. Significant amount of Internet content has low or no value. So, we need to learn how to rate the Internet content of any type and better understand what we already have. Actually, filtering of information is an important function of social networks.

Semantics

Why semantics is important? It is not because of semantic web, it is not because of Web 3.0, and it is not because of Semantic Technology in general. In present, Semantic Technology is based on so-called ‘semantic stack’, which includes ontology and logical inference. Declarative nature of the ontology and the need for the logical inference made many people to believe that Semantic Technology begins where Artificial Intelligence ends. People already think that like in AI we are not dealing with Intelligence we are not dealing with semantics in most cases in existing Semantic Technology. Where is real semantics?

Semantics is important at least for two reasons:

• as a key to real AI
• as a fundament to the computing like logic or math

Nowadays Semantics is a fundamental thing missing in the computing. This is a unique moment of a historical move in the software industry. There may be nothing more significant happens ever again in this industry. Unfortunately this understanding of Semantics is far from common.

What is about real AI? Have we stopped doing it?

Most of AI efforts were efforts from the top, when objective was in sight. Interesting thing is happening from the bottom, when people are doing AI without suspecting it. What do you think “reuse” and “design for change” mean? These are attempts to give the system quality of real AI system. If we would have real AI systems, reuse and design for change would be implications of other principles. Let see what Nova is telling us about Web 3.0:

“RDF enables something as potentially important as HTML. Just as HTML enabled a universally reusable Web of content, RDF enables the Data Web, a universally reusable Web of data. …. But currently although browsers can render the formatting and layout of data, they don’t know anything about the meaning of the data, unless they are explicitly programmed to do so. The same is true for all applications today — they have to be explicitly programmed in advance to interpret each kind of data they need to use.

The Semantic Web provides a solution for this problem that is analogous to what HTML did for content — RDF and OWL provide a standard way to describe the meaning of any data structure, such that any application that speaks these languages can correctly interpret the meaning without having to have been explicitly programmed to do so in advance.”

Knowledge representation was always and always will stay in the future the problem #1 in AI. Nova is stating that we are improving it in Web 3.0 plus we are adding Semantics to it to reduce explicit programming. We are dealing with language issue here and it cannot be more close to AI.

Summarizing it all: AI in a sense of knowledge representation is the only thing we are doing now, when we are having significant development, the rest is repetitive engineering practices.

Web 3.0

Let see, what Web 3.0 could mean. Web 3.0 has two connotations:

• Software upgrade – new features, less problems.
• Movie sequel (Die Hard 1, Die Hard 2 …) – same actors, new script, capitalizing on previous success, reducing risk and almost guaranteed commercial success.

A lot of positive associations, it would be stupid not to use Web 3.0 term, no matter how questionable it is.

Web 3.0 is a frame to focus human activities toward web progress in an understandable fashion through previously identified approach (Web 2.0). It is a social movement and it will need all attributes of it: industrial shows, venture capitalist activities, magazines, etc. This is about form of Web 3.0.

In terms of content, as Tim mentioned, Web 2.0 was pure application try, not much of new technology. Social web and networking were invented as Web 2.0 differentiating applications. New applications may be the case for Web 3.0 as well, also there is a strong hope that the new applications quality enabled by Semantics will help to implement what was not possible in Web 2.0 and it will be enough for differentiation. This is the most interesting topic, which requires separate discussion.

BTW, Tim also has mentioned that Web 2.0 term was accidental …

Web 3.0 is a placeholder for the applications enabled by the Semantic Technologies. Apparently, there will be some time for the Web 3.0 applications to evolve and demonstrate full potential of the new approach. Initial versions may not exhibit big advantages, at least from the user point of view. Some of Yahoo web sites (for example, http://food.yahoo.com) are already using elements of current Semantic Technology like RDF store.

Conclusion

No matter what we are doing and how we call it in the software (if we are really making a progress) we are doing AI. AI is absolute imperative and it is the only thing we are going to do. Interesting thing in the current development is that semantics is a main missing component of AI, which was escaping our attention for long time.

Recommended book about motivation: “Beetle in an Anthill.” – New York: Macmillan, 1980, ISBN: 0026151200.

© SemPL.net, 2008