Build Your Own Bot, But Don't Create A Monster

[ Second of two parts ]

By Birrell Walsh

      Software that communicates. Programs that carry out your chores while you drink coffee. The idea delights some with its power and makes others tremble with memories of the sorcerer's apprentice. This month, we look at programs that take on human functions, and find bot-making programs you can download and use now.

      Intelligent agents, or "bots," are semiautonomous programs that do tasks for you. They are not globally intelligent, as we pretend to be. Instead, they are clever and flexible at a single task. Turned loose on the Internet, they act for all the world like simple, specialized living beings. Whether the idea of (somewhat) self-directing software makes you grin or tremble, such products are here and at work among us.
      Last month, we looked at the kind of pre-built agents that one can meet on the Internet. Search engines are full of bots. "Spiders" go out to the Web and discover what sites exist.
      Other bots create indexes of the information that the spiders have found, and still others use those indexes to create a presentation page with your answers. Shopping bots compare prices among sites. "Alerts" remember what you want to know, repeat the searches and notify you periodically of the results.
      This month, we examine just two of the many products that let you create your own bots. It's fun and perhaps very profitable, but there is a serious side as well. When you create a bot, you become accountable for its actions, you have a duty to take care of others. When you use agents that others have created, you need to watch out only for yourself.
      Whenever you sic an agent on the world, it is wise to ask just what its effect on others will be. Will it give correct information? Will it overuse the resources of someone's Web server? Will it accidentally pick up and redistribute copyrighted material? When you create an agent, it is your responsibility to consider such issues.

Alice The Chatterbot
      In the 1950s, Alan Turing made good conversation the test of computer intelligence. While other sorts of agents have gone from concept to front-line service, conversational agents or "chatterbots" remain in their infancy. A half-century after Turing proposed his test, chatterbots have not passed it.
      That very failure means there is a great opportunity for linguists and programmers to make a buck. Graduate linguists need to find some way to pay for their difficult-to-sell Ph.D.s
      If you want to experiment with chatterbots for free, try downloading "Alice." She is a talking engine (invented by Richard Wallace of San Francisco) that you can modify and train to respond as you see fit. There are open-source versions of Alice in several languages, and a community of users that share experience and ideas. Many of the sites in the Resources box have network versions of the program, so you can see how she performs.
      Alice works by taking input speech and running it through a set of filters, coded in a dialect of Extensible Markup Language (XML) called AIML, the Artificial Intelligence Markup Language. Alice is searching for patterns, and when she finds a match for one, she responds from a template.
      Both patterns and templates can have a wild card in them, so if you say "My name is Joe," Alice remembers it. Alice can store a few facts about the user, the conversation and herself. She can carry on a discussion for a while before her forgetfulness forces a subject change.

Click to view larger

      The designers of Alice emphasize that her structure is minimalist. The program represents its filter system as a huge spiral through which utterances travel, seeking a pattern that will trigger a response. There are currently about 24,000 pattern-response categories in the basic coil. The program could probably handle cocktail-party conversation —— though that might not be a very high standard.
      AIML allows you to add to and modify the patterns to which Alice responds. The engine will keep track of conversations Alice has, so you can see what parts of her vocabulary, her "responsory," need work.
      Alice is not intended to be a final chatterbot that interacts with the public. Instead, the Alice engine and AIML are an experimental arena. Designers can try out ideas and get their own inspirations before attempting to create the Great American Chatterbot. And did I mention that Alice is free?
      NQL: A LANGUAGE FOR AGENTS
      At Internet.com's recent San Francisco Bots 2001 show, I found a language designed for the building of bots. Priced from $495 at the low end to $10,000 for the enterprise edition, it is definitely not free — though you can try it for 60 days for no charge.
      It is called "NQL," the Network Query Language, a deliberate play on "SQL," and it facilitates the creation of very powerful agents to roam the Internet in your service.
      The bad news is that NQL is a language. It is not a Web page with a couple of buttons you push to get a result. To use it, you must not be intimidated by programming. The good news is that NQL is a language. It is the creation of NQL Inc. of Santa Ana, CA, a company that has been designing and building agents for clients for years.
      There are more than 550 functions in the NQL toolbox. The functions get data from the Internet (Web pages, FTP and e-mail), process it in complex ways, and publish it to your intranet, and then out to the world as mail or new Web pages.
      In a first experiment, I created a script to search incoming e-mail for keywords and report the result (also by e-mail) to another address. That took about an hour from the first time I opened the program, and made me want to try something more strenuous.
      I then tested it on a tax project. A large library was being donated to a university and the owner wanted a book-by-book valuation.
      He had a spreadsheet that had the name of the author and the title of each book. The problem was to design an agent that would go on the Internet, research book prices and give an average price for each volume to justify the donation amount to the IRS.
      NQL can speak to Excel through Object Linking and Embedding (OLE) automation, so I used OLE commands to open the spreadsheet and read the title into one NQL variable and the author's name into another. Next I used NQL's "Browser Session Recorder" to automate much of my script writing.
      I was able to complete a script to find an average price for each book in just a few hours, working only in moments stolen from other projects.
      The program has been used to create a number of complex multi-agent systems. NQL sells a rich application called ContentAnywhere that allows corporate users to summon Internet data through its NQL product and insert it into other programs. There is also a rumorbot, a program to search for online mention of a corporation's products, created with NQL by Agence Virtuelle of Geneva.
      IT professionals could create very powerful Internet agents and integrate them with their company's knowledge-management system.
      A talented consultant could perform seeming magic for a client who did not know the power of the language. The variety of possible applications for NQL is staggering.

Learning About Bots
      To stay current with developments in the intelligent agent world, you can subscribe to Internet.com's Bot Spot newsletter. For a more academic and advanced user, it is worthwhile to join the University of Maryland Baltimore County Agent Web.
      So where do we stand on agents? As intelligent machines begin their long companionship with us, the opportunities — and the tools —for clever and thoughtful bot makers are better than ever.