Perhaps Allan Turing will be called the father of AI whose contribution is tremendous in many fields of computer science. Proudly known for his trademark definition of Turing Machines, Turing in 1950 devised a test known as the Turing's Test as a method to effectively conclude that an artificial life is truly intelligent. Those were the times when robots, droids, cybots, cyborgs were quite popular in sci-fi and many people including him believed that the Turing Test will be broken before the end of the 20th century. But it didn't happen. In fact Turing Test is a feat computer scientists across the globe believe that will take many more decades to accomplish (many hold the belief that it will never be broken).
Actually this test measures only one aspect of AI, considered the first step in machine intelligence, Natural Language Processing (NLP). NLP includes those algorithms and tools which make a machine interact in human or natural languages as human beings do. For example you ask a machine or a program to describe yourself in one sentence, this is NLP. The complexity of NLP lies in the conversion of grammar rules to machine processing and vice versa, which is a difficult problem given the ambiguity of grammars of languages existing in the world. Many algorithms are being used for NLP, from Neural Networks to Neuro-Fuzzy logic but still the answers received from machines seem to be dumb.
To conduct the Turing Test, we need two people and the machine to be evaluated. One person plays the role of the interrogator, who is in a separate room from the computer (or on network) and the other person. The interrogator can ask questions and receive typed responses. However the interrogator knows them only as A and B and tries to determine which is the person and which is the computer. The goal of the computer is to make the interrogator or fool the interrogator into believing that it is the person. If the machine succeeds then this will coincide with the fact that computer can think and act like a human being (at least linguistically).

So for example if the computer is asked the question what is 122486 times 19 then it can wait for sometime and return a wrong answer, perfectly valid since a human in these circumstances would be expected not to act like a calculator. The more serious issue, however, is the amount of knowledge the machine would need to pass the Turing Test. Turing gave an example of this by asking the computer a tough question about a famous poem. If the computer can answer these kinds of questions then it means that it has as much knowledge as a human being.

In 1966 Joseph Weizenbaum, a scientist of MIT, wrote a program Eliza (named after the central character of the famous movie My Fair Lady, who spoke in a foolish way) in a Lisp-like language called MAD-Slip, way back in pre-Unix days. Eliza is called the first AI and simultaneously the first NLP program, a chat bot (actually it was more a psychoanalyst). It is pretty easy to fool Eliza; in fact you only need a few sentences chatting with it to conclude that it is artificial. It used string matching techniques to generate responses.

But suppose we are willing to settle for less than a complete imitation of a person. Taking this fact in view a famous prize (as revered in the AI community as the X-prize of Aerospace technology) called the Loebner Prize has been conducted.

In 1990 Hugh Loebner agreed with The Cambridge Center for Behavioral Studies to underwrite a contest designed to implement the Turing Test. Dr. Loebner pledged a Grand Prize of $100,000 and a Gold Medal for the first computer whose responses were indistinguishable from a human's. Each year an annual prize of $2000 and a bronze medal is awarded to the most human computer. The winner of the annual contest is the best entry relative to other entries that year, irrespective of how good it is in an absolute sense.

Since then Alice the famous bot now, has been winning for many years this revered prize. Alice follows the strategy of keyword matching but enhancing it much more. Suppose we separate the Inference Engine and Knowledge Base of an NLP program like an Expert System (another AI domain). Alice does exactly that by keeping the knowledge base and Inference Engine (the think tank of AI) separate. The Knowledge Base is encoded in a markup language called Artificial Intelligence Markup Language (AIML). The Inference Engine is coded in java. Now there is a plethora of bots, mostly chat bots on internet competing for Loebner Prize but yet very far from beating the actual Turing's test.

An important part of NLP or any AI program is its learning capability. Although Eliza didn't have any learning inherent in it, the latest bots do have. Learning in AI is of different types and use different algorithms and heuristics. The simplest and easy to code is the Rote Learning which is nothing but caching the program parameters in main memory. Interestingly this increases the efficiency as well as effectiveness to nearly 80%. Other learning methods include Learning by taking Advice, Learning by Problem Solving, Learning by Analogy, Learning by Examples and the most interesting one -Learning by Discovery. The application of any of these methods to NLP programs or bots requires a lot of work, for example computer scientists have yet not been able to implement Learning by Discovery in bots.

Learning and growing the way a human child does seems a good analogy to be implemented in AI but its not that simple. Curiosity is what drives the human mind, curiosity which makes him question. Heuristics that make a program thirsty for knowledge ensure his Knowledge Base expansion very easily. Feeding simple text files to NLP programs to add to their KB (Knowledge Base) has been accomplished. So you could feed an electronic encyclopedia to a program and test its intellect over that. Or Internet Relay Chat (IRC) can be used to connect the program online to messengers and chat rooms so that it interacts with users online from them (great risk involved in this, seriously).
This goes suffice to say as far as knowledge is concerned. But what about analysis, judgment and reasoning? Unfortunately this is where bots fail to perform. Reasoning and analysis is better done by Expert Systems confined to a narrow domain and using extensively vast amount of knowledge in that field. Then are they able to say that these symptoms suggest this viral infection. Language is vaster than any scientific domain because it is in common day to day use and being enhanced with every passing moment. Slangs which were prevalent in 18th century maybe unheard of in 21st century. Moreover language is ambiguous, that's why some scientists suggest the use of better natural language grammars for machine interactions, say Sanskrit. But truly speaking how many people in the world speak Sanskrit?
The parts of the inference engine can be KB language specific or the way the author codes but basically just as a developer would develop a compiler for a language is the way a NLP program is built. The difference is that in a compiler the input language is a computer language and the output language is also a computer language (mostly assembly code). But in case of a natural language processor the input is a natural language and output is pragmatic analysis of what the sentence means. Therein lies the complexity of the problem because natural language grammars are not so hard and fast as computer grammars, so how to precisely analyze a sentence?
A problem of fooling around with a computer is an interesting one. Suppose that a user instead of seriously chatting with the programs tries to fool it and provides it misinformation or rumors. How to know what the mood of the user is through his input? Moreover how to respond when he is trying to say that he once went to sun. A human being can understand this very easily but for a computer to understand this is very difficult.

AI is a vast field and NLP is not the only intriguing domain in it. AI includes robotics (the more enhanced one is called cybernetics) and searching. Whereas searching is much simpler than the earlier two, Robotics is a field which has still to see the creation of an artist by technology.

Say we talk of a droid and with droid I mean robots like the one concealed by pentagon (which can supposedly swim, talk, and even shoot, is it myth or reality?) and not the one's like Asimo being displayed year after year by companies like Honda. What should this Artificial Life form possess to be called intelligent? First of all if it can talk it has NLP. Then comes speech processing (which is in its nascent stages of development) and recognition, for computer vision it has to have static and dynamic visualization and 3D (or at least 2D) recognition of objects and sense of touch (force or heat sensor can do this) leaving apart the cybernetic and electro-mechanical components of the droid. Sense of smell and taste have not yet been started to be created artificially.

So making of this droid involves so many difficult and complex tasks which are still being understood, leave apart the feat shown in Spielberg's AI. Well that even deals with Ontology (the part of AI that deals with computer consciousness), conscience (the ultimate intellect) and artificial tissues.

By taking into account the progress made by AI since its inception we can ask questions like - has science lived up to science fiction? Well it is always easier said than done but still when does the reader think the Turing Test will be passed, if ever. And is Artificial Intelligence intelligent enough by 2005 or is it just dumb machines doing specialized jobs?