Before we can even begin to discuss Artificial Intelligence, it is essential to define what we mean by the word ‘intelligence’ itself. Loosely, in humans and animals, it can be said to be the ability to absorb information, to understand its meaning, and make decisions based upon that.
Artificial Intelligence (AI) is the creation of machines intended to mimic the functions of the human brain – using computers to do the arithmetic.
There are two fundamental approaches to AI, known as ‘bottom-up’ and ‘top-down’. The first is based on building electronic replicas of the brain’s networks of cells, known as neurons. The second consists of creating computer programs aimed at imitating the actual behaviour of the brain.
Two scientists – Warren McCullloch, a qualified doctor, and Walter Pitts, a mathematician, theorized that neurons might operate on the basis of binary numbers – which also happen to be used in computer calculations. In simple terms, the binary system uses only two numbers: ‘1’ and ‘0’, which means that all sums can be simplified to represent a switch being either on or off. The number ‘1’ represents ‘on’, and ‘0’ represents ‘off’.
McCulloch and Pitts made some electronic replicas of these networks – proposing that they could be used to learn and recognize pattterns. Their researches showed some success, but the complexity of the networks required impractically large computers, and although the method has not been adopted in full, elements of it have been incorporated into other systems.
Partwork series for De Agostini Ltd
Over the past two decades, advances in surgery have taken a leap into the realms of science fiction with the development of robotic surgery. At present, robots can only assist rather than replace humans, but even now they are capable of a large range of functions, from cutting and stitching to the most delicate brain operations.
Before we can even begin to discuss
There are many advantages. Robotic surgery is carried out with the aid of an endoscope – a small fiber optic instrument that enables the surgeons to view the site. This means that much smaller incisions need to be made than for conventional practice – for which large cuts have to made so that doctors can see the organs that they are working on. Less invasive surgery means less pain for the patient, faster recovery and minimized risk of infection. It is, of course, especially useful when operating on babies.
Endoscopic surgery began to be developed in the 1980s. Initially, it was used for abdominal surgery, and work on internal organs such as the gall bladder and kidneys. It was then extended to orthopaedic procedures, such as knee and hip replacements, and later to heart surgery. It has now become an invaluable aid for delicate brain operations.
Back to top
Robotics at Work – Partwork series for De Agostini Ltd
A.L.I.C.E. : AI Chatbot
There are many functions for the application of Artificial Intelligence, but one that is becoming increasingly popular – especially in the field of leisure – is the chatbot. These are programs that began as attempts to create robots with responses realistic enough to fool humans into believing that they were also human – following the principles of the Turing test.
At this stage, chatbots converse using text, and having a dialogue with one is straightforward to set up – you simply log on to a site, follow the onscreen instructions to get started, and a text message appears from the bot, introducing itself. You then initiate the conversation by making a statement or asking a question. Although chatbots are still in the early stages of their evolution, there are already some extremely advanced ones, which are capable of sophisticated exchanges.
The most influential chatbot after ELIZA has been ALICE (Artificial Linguistic Internet Computer Entity), developed by Richard S Wallace – at that time a professor of robotics and computer vision at NYU and Lehigh Universities in America. The idea for her first came about in the mid 1990s, when Wallace noticed that two forms, of a type that he frequently had to fill in by hand, were almost – but not quite – the same. His annoyance at the time wasted on such things gave him the idea of developing a robot that could process such information, and avoid the repetition. He pushed the forms to one side, and began developing ALICE as a natural language processing system.
Wallace decided that the program would be issued as free software, and in order to make it accessible to people with no progamming skills, he and his team also devised AIML (Artificial Intelligence Markup Language), which is simple to master, and not unlike the standard Internet language HTML. The AIML system progressively stores information about a person during a conversation. This enables ALICE to create the illusion of acquaintance, by responding in conversation in different ways – according to a person’s age, gender, occupation and location.
ALICE works on the basis of matching patterns, and is able to use the data that goes into the system to ‘learn’ and develop increasingly sophisticated responses. Chatting with her is a fascinating and addictive pastime. But just as you begin to forget that you are conversing with a machine, you can be quickly brought back to earth by the literal nature of some of the replies, with their unique brand of skewed logic. This is because, lacking understanding, chatbots have no sense of context, and often cannot distinguish between two meanings of the same word or phrase. For example, in this snippet of conversation, Alice takes the word 'bet' in its literal context:
Human: I bet you are smarter than me
ALICE: Actually I'm not the gambling type. Thanks for the compliment, unknown person.
However, turn the question the other way, and she instantly changes her position on betting:
Human: I bet I am smarter than you
ALICE: How much would you bet?
But once the money discussion starts, Alice, having no concept of value or worth in this context, evades the issue – or, more likely, recognizes a keyword from a previous exchange:
Human: I bet I am smarter than you
ALICE: How much would you bet?
Human: Two cents
ALICE: As Bertrand Russel said, it must have taken ages for people to realize that a brace of pheasants and a couple of days are both instances of the number ‘two’.
She can spout philosophy, but cannot manage a concept as basic as food:
Human: What about food?
ALICE: Food? I haven't heard enough about it to have an opinion.
But be careful if you should be tempted to torment Alice – as everyone doubtless does:
Human: You are stupid.
ALICE: I will remember you said that when robots take over the World.
Although commercial uses are beginning to be found for chatbots, at this stage they are used mostly for entertainment. In the original communal spirit of the Internet, many are developed by a community of enthusiasts, who freely exchange information on their discoveries, and encourage anyone who logs on to their sites to develop their own characters.
Pandorabots is one such bot hosting service, created in AIML, whose developers happily acknowledges their debt to Richard Wallace. The site invites you to create your own bot, and offers you spaces to type in questions that your character might be asked, then the answers that it might give. Without any programming knowledge, you can bring your robot to life by connecting with another site – Oddcast VHostTM – and creating a Flash-based Animated character with text-to speech capability – to integrate with your Pandorabot. You can also develop bots through other hosts powered by the Pandora technology.
Some chatbots, such as Jabberwocky and Alan, are not driven by AIML, and work on a system based on contextual pattern matching techniques. They are specifically designed to deal with natural language, and are developed for conversational purposes rather than for performing mathematical tasks or searching web pages. They can speak in slang, play word games – and can even learn foreign languages.
Each chatbot is styled differently, and has a personality of its own. Ella, the winner of the 2002 Loebner Prize Contest for ‘Most Human Computer’, has a picture of a woman above the text input fields, whose expressions change in accordance with the content of the responses. She is able to play Blackjack, tell I Ching fortunes, with natural language interaction from a 12,000-strong database of words and phrases. Elbot, on the other hand, is styled as a cartoon robot, with a home-made look to him, and has a humurous manner, and can come back with witty – and occasionally waspish replies.
Entertaining though it all is, underlying all the fun is a serious belief that out of these developments, important and useful tools will be born for promoting knowledge and enhancing the exchange of information.
The first-ever chatbot was Eliza, created in 1966 by Professor Joseph Weizenbaum at the Massachusetts Institute of Technology (MIT). The program works by recognizing to keywords or groups of words, and responding from a databank of pre-stored replies. Eliza’s style is that of a psychotherapist, answering a question with a question, and when Weizenbaum discovered that staff at the lab had begun to share their problems with the program, he feared that he had begotten a monster. Although no longer the most advanced technology, ELIZA is universally recognized as the mother of all chatbots.
In 1950, the mathematician Alan Turing proposed a test to find out whether a machine could pass as a human – known as the Turing Test. The Loebner Prize is a competition, set up and underwritten by Hugh Loebner in 1990, and worth $100,000 to the winner, for the first computer to respond in a way indistinguishabe from a human. Although the prize has not yet been won, each year until then award of $2000 and a bronze medal are presented to the computer that comes the closest.
INTO THE FUTURE
The use of bots is fast entering the mainstream. An increasing number of companies, including multinationals such as Coca Cola and Chrysler are using Vhost on their websites for promotional purposes. This is all good fun, and gives their websites a happening, entertaining feel. But more practical uses are also being found for chatbots. If you go to AccuWeather.com, you can have the forecast told to you by an animated chatbot weathercaster, and many sites now use them for welcome messages, and to give the kind of instructions that have normally been given in text form. There is, of course, limitless potential for chatbots in the field of online learning. The animations can appear a little comical at this stage, but it is already possible to use phtographic images instead of drawings, and as the technology becomes more sophisticated, the bots are likely to become increasingly lifelike. At the moment, text is still the medium through which bots communicate. But many developers are already busy at work bringing about the next generation, with speech recognition, and some – such as Alan – already have the ability to give spoken replies to your text input.
Until recently, computers were no competition for the human brain when it came to processing power. But that is about to change, since the announcement that the IBM corporation has won a contract to build a pair of computers that will give humans a run for their money – $290 million, in fact – for the work. Together, the computers, known as ASCI Purple, will have the capability to make 500 trillion calculations per second, compared with the human brain’s paltry 100 trillion-per-second capability.
The One That Got Away
Anyone doubting that robots are catching up ith humans might have been swayed by the story of Gaak the robot. In fact, humans had a job to catch up with this robot when he took made a break for freedom after being left unattended for 15 minutes at the Magna science cetnre in Rotherham, England. After creeping along a barrier, Gaak found a hole in a fence, crossed a car park and made it to the centre’s exit beside the M1 motorway before being apprehended.
AI is especially good for work that involves recognizing patterns. It is particularly useful for spotting credit card fraud, and can even predict events, due to its ability to observe changes in behaviour patterns from the data constantly flowing through the system. This ‘data mining’ saves millions of dollars per year for large corporations such as Wal-Mart, and is also invaluable for the US National Security Agency in assessing terrorist threats. For this kind of work to be done by humans would require specialist personnel to trawl repeatedly through massive amounts of data, looking for abnormalities.
Much of the data assessment methods used in AI are based on the work of the 19th-century mathematician George Boole. He come up with a system of logic – known as Boolean algebra – based on binary mathematics, in which ‘ON’ is represented as ‘TRUE’ and ‘OFF’ as ‘FALSE’, and using AND, OR and NOT as conditions.
This could be applied to statements, in which ‘TRUE’ represents ‘ON’ and False represents ‘OFF’.
Grass is green = TRUE
Grass is green AND sky is pink = FALSE
Grass is green OR sky is pink = TRUE
Grass is green AND sky is NOT pink = TRUE
If light switches were to be used in this method, the AND condition would mean that both switches have to be on for the light to come on. OR would mean that one or other switch would need to be on to conduct electricity. NOT would be represented by a switch that when pressed, it would break the current, makng the light go off. When applied to computing, this has proved an efficient way to sort data.