The first agents were designed by General Magic, a small high-tech company financed by Apple, Matsushita, Philips, and Sony. The company has developed Magic Cap, a graphical interface for controlling teams of agents working in networks. Computer manufacturers such as Apple have created personalized interactive icons on their multimedia computers. The user can set the parameters of the interface, choosing the voice, sex, and physical appearance of the agent, designing a synthetic character, a robot, that appears on the screen as an icon and speaks, its lips moving with the words. The agent program understands human speech and responds the same way that computer icons do when clicked by a user.
My favorite agents are Oliver and Sarah, two computer creatures who work closely together, and are products of the fertile mind of John Evans, president of News Electronics Data, an American subsidiary of the Rupert Murdoch group. Evans's objective was to make it possible for any user to order a custom-made newspaper, the Daily Me, by using agents. But many other applications have been developed, such as assistance with hotel and transportation reservations or research in databases. These agents are now available in commercial software.
Oliver is a yellow Labrador retriever. He regularly appears on the screen looking for something to do, independently of the functions being performed. If you do not give him a specific task, he gets bored, scratches himself, plays with balls of paper, and goes and sits in his corner. When you want him to search for references in international databases by key word, all you have to do is write the retrieval formula for him and he will disappear into the networks. At lightning speed, Oliver dials the numbers of dozens of databases, enters the codes, types the key words, and collects the information and stores it in files, and returns with the results of his research, wagging his tail. If you're satisfied with his work, you give him electronic biscuits; if you're not satisfied, you send him to the doghouse. Oliver's expert program remembers the user's reactions and takes them into account when the next request is made by pursuing certain strategies and eliminating others. Thus there is reinforcement and learning as there is in training a dog.
Sarah is an ethereal character, a kind of fairy, represented by a vague, mysterious graphic. You tell her your travel preferences (type of hotel, location, price, meals, airlines, etc.), using a series of cursors similar to those in simulation games. Then, all you need to give her are brief instructions, such as, "Organize my meeting in London with Jim Stuart for ten o'clock on January 15. I would like to take him to lunch and return to Paris in the evening." On the basis of these instructions, Sarah goes to work. She knows your habits. No dawn flights, so she'll make a hotel reservation in your favorite area. You will have to leave Paris after your last appointment (Sarah manages your datebook). Car rental, restaurant reservation (Sarah knows Jim Stuart's preferences, which are recorded in the contacts file), reservation for the return trip. Taxi to Charles de Gaulle airport, as usual. With all the constraints determined, Sarah sends Oliver to check plane schedules with the travel agent and book the flights, book the rental cars and hotel, and make restaurant and taxi reservations. Oliver comes back with a series of automatic confirmation numbers. All Sarah has to do now is print your detailed schedule for the evening of January 14 and the following day. You will communicate your response to her, and she will update her reference files. Here too, the expert program of the agent learns and evolves over time.