Andrew Leonard is yet another Wired reporter who has written a book about the computer software that will take you into the next millennium. He discusses bots, long-lived software processes with some decision-making capability, in their native habitats of IRC, usenet, MOOs and the web.
While it is safe to guess that bots, under the guise of autonomous software agents, will be major players in the computing world, in order to get a book's worth of material, Leonard had to define the class very broadly. As a result virtually anything from IRC eggdrop bots to that little dancing paperclip in Microsoft Word qualifies. The stretch becomes particularly visible when he reaches back into history to discuss the origin of bots and comes up with early backup software and 'Eliza' [Note to younger readers: Eliza was a program which faked human conversation, badly. It has been implemented in every programming language you can imagine]. In chapter 4, Leonard actually describes the Wumpus of "Hunt the Wumpus" as a bot, about as ludicrous an argument as you could imagine. [Note to younger readers: Hunt the Wumpus was a very simple, very stupid game that was played on university mainframes and the early home computers of the 1970s. You wandered (textually) through a finite network of caves. Each time you moved, the wumpus moved too, randomly. You could shoot arrows into adjoining rooms. If you hit the wumpus, you won. If you wandered into the wumpus, you lost. Look, Doom wouldn't be invented for another 20 years.]
Once he's defined 'bot' so broadly, Leonard has to contend with the universe of daemons and faceless applications which infest any modern operating system. Unfortunately, most of these are not very exciting and Leonard focusses on software which is more visible, and ideally anthropomorphic. This means that all his modern bots fall into a small number of classes: usenet monitoring programs (including cancelbots), IRC bots, MUD and MOO bots, and web spiders. This puts him in an awkward position -- this book is clearly intended for the mass market, but the vast majority of the discussion regards systems which his readers will never use.
Leonard very much wants to draw trends and lessons from the evolution of bots in these areas. Unfortunately for him, the universe of bots he chooses to discuss has been so short-lived that he can draw only the most banal conclusions. "Poorly tested bots can get into infinitely recursive conversations with each other." "AI bots do a poor job of mimicking human beings." "When evil bots are programmed, good bots are usually created to fight them. Both groups are then reprogrammed repeatedly in attempts to outsmart each other."
This book avoids the typical Wired error of quoting a bunch of "friends of Wired" as experts on whatever topic is at hand. However, it does slip into the magazine's absurd typography. Many paragraphs (selected randomly, as far as I can tell) start with an initial letter which is dramatically larger than the surrounding text, rotated sideways, and rendered in a different font. How hip.
The book is delightfully cerebral, drawing from Plato and Darwin, Gibson and Asimov. [Note to younger readers: Plato for his moral "demon", Darwin for the theory of evolution by natural selection (which, if you ask me, clearly does not apply), Gibson for the AIs in Neuromancer, and Asimov for the "Three Laws of Robotics"] The research is admirable, and Leonard tracks down the authors of an awful lot of the software he describes. I used MUDs a few times back in 1990 or so (and honestly never saw the point). Chapters 1 and 5 describe in amusing detail the troubles caused by bots at various MOOs, including an extended discussion of "The Barney Problem," or the 1993 swamping of Point MOOt by sloppily programmed Barney Bots singing the "I love you" song.
The discussion of Bot politics on IRC was instructive. I've been on EFnet for almost ten years now, but have always tried to avoid the undying politics of IRC-abuse and server control. As a result, I missed the inside scoop on why Alternet formed and why Nickserv went away, and so forth. Leonard fills in the gaps. Would this be as interesting to somebody who doesn't use IRC, or who uses it so much that they already know the stories? Probably not.
The material in chapter 3 on the failure of AI could form the core of its own book, a book about why AI looked so promising in 1980, the brilliant people who devoted their careers to it, and why it failed nonetheless.
"In part, the AI community doomed itself. Its own bold promises and early success led to a breathless boom period in the 1980s. Corporations rushed to adopt so-called expert systems -- programs that specialized in particular domains of knowledge and were supposed to represent the accumulated wisdom of hundreds of human experts. Unfortunately, most expert systems ended up requiring even more human resources than they replaced, and they often failed to work as promised" [stern: give me examples! juicy ones!]
"A sorry record of broken promises and the demise of the cold war dried up most AI funding and sent the artificial intelligence community reeling. Attendance at the premier artificial intelligence conferences declined. Morale sank to its lowest point when aspiring AI workers discovered that just putting the words artificial intelligence in a grant application guaranteed the kiss of death."
Those two paragraphs, on page 45, could be the first two paragraphs of a book about the past failure of AI and new methods being tried today, especially on the web. That book would probably be better than the one which Leonard has written.
Leonard writes well, and his research can not be faulted. I look forward to reading his future books. This particular book should be of interest to people already familiar with (and curious about) robo-moderators on USENET, web spiders, IRC or MUDs/MOOs. If you do not fall into one of those categories, don't waste your time here.
back to the books page