Bots

Dateline: September 21, 1997

"IN one form or another, bots have been around since the early 1960s," writes Andrew Leonard in his new book BOTS: The Origin of New Species. A bot is "an intermediary between the digital and the biological," in the same way that mythological daemons of the spirit world were intermediaries between mortals and the gods. Plato wrote:

The divine will not mingle directly with the human, and it is only through the mediation of the spirit world that man can have any intercourse, whether waking or sleeping, with the gods.

Increasingly, it is only through the mediation of bots—digital daemons—that, awake or asleep, we enjoy the blessing, and suffer the curse, of computers.

"Bots are cool. They stoke our imaginations with the promise of the universe populated by beings other than ourselves, beings that can surprise us, beings that are both our servants and, possibly, our enemies. Bots, which are here, now, and growing in number and power every day, are advance scouts from the future."

But they don’t have to be benign, says Leonard. "Bot misbehavior doesn’t have to be accidental. Bots can be instructed to do whatever their creators want them to do, which means that along with their potential to do good they can also do a whole lot of evil." Bots aren’t just cool: "They’re trouble."

Microsoft, Intel, IBM, Apple and other giants are spending millions on intelligent agent—advanced bot—research, in "top-down" fashion. But as Leonard points out, "Bots, meanwhile, just happen," percolating from the ground up, "a grassroots phenomenon, as likely to be conceived by bored teenagers as by PhDs from Stanford or MIT." The evolution of bots was "accidental, unscripted, and unpredictable. They are the spawn of the Net’s anarchy and decentralization, the product of a thousand different hackers writing code on a thousand different computers."

This article summarizes what bots are, what they do, how they arose, and where they are headed.

What is a Bot?

Robot means "forced labor" in Czechoslovakian. It was first used to describe mechanical beings by Czech scifi writer Karel Capek. Bot is a contraction of softbot, which is in turn a contraction of software robot." A bot is a software version of a mechanical robot, "Strings of code written by everyone from teenage chat room lurkers to topflight computer scientists," says Leonard. Like a mechanical robot, a bot is guided by algorithmic rules of behavior—if this happens, do that; if that happens, do this. However, bots are just beginning to be endowed with heuristic capabilities: if this (that) happens, as a rule of thumb you should probably do this (that), other things being equal (which they never are). You decide.

MIT psychologist and online interaction guru Sherry Turkle defines bot as "a small artificially intelligent program." Leonard defines it as "a supposedly intelligent software program that is autonomous, is endowed with personality, and usually, but not always, performs a service." There are several names for bots: daemons, agents, softbots. Purists may point to subtle technical or functional differences among them, and "there is no consensus on what particular sequence of encoded ones and zeros truly classifies a bot," but they are fundamentally the same. They differ from ordinary computer programs such as WordPerfect in that they are designed to run on their own. They don’t just sit there, waiting for you to enter a command.

What do Bots do?

"Bots are variously designed to carry on conversations, act as human surrogates, or achieve specific tasks—in particular, to seek out and retrieve information. Bots entertain, annoy, work, and play," writes Leonard.

"Mailbots filter electronic mail, preventing junk mail and spam advertising from clogging up our online mailboxes. Chatterbots carry on whimsical conversations in online, real-time text environments, such as chat rooms or MUDs. Cancelbots seek out unwanted expression and erase it from electronic bulletin boards. Gamebots populate computer game environments with believable characters and wily foes. Web robots explore the hyperlinked reality of cyberspace, mapping out and indexing the vast quantities of information available through the World Wide Web."

The service aspect of bots is "crucial, a character trait running at cross angles to issues of autonomy, personality, or intelligence."

Today’s bots are the "precursors to the intelligent agents that many visionaries see as indispensable companions to human in the not-too-distant future. Intelligent agents are software programs designed to help human beings deal with the overwhelming information overload that is the most obvious drawback to the information age. The two categories are not exactly equivalent. Agents do not require the accoutrements of personality." (But neither do spambots, cancelbots, etc.!)

"True botness makes a computer program or a computer environment more approachable, more entertaining, more user-friendly—drastically important considerations for those who wish to create successful agent prototypes, either for academic research or for the consumer marketplace."

A mailbot (sometimes called a mailer daemon) is a postmaster. It hangs around a mail server, looking at the addresses on messages sent and received by users and at the at the state of their mailboxes. It will tell you if a message you send is undeliverable for some reason (incorrect address, no such person at that address, the address is not receiving mail). It will tell you when your mailbox is full, if your mail service provider has put a size limit on your mailbox. It will automatically forward incoming mail to another address, if you tell it to.

HTTP daemons hang around Web servers, intercepting your requests to look at a Web page and sending you whatever text, graphic, sound, and video files make up that page.

Print daemons hang around network file servers, putting users’ print jobs in a queue and making sure they all get taken care of eventually.

All of these bots work quietly and invisibly in the background. They work automatically and autonomously—on their own. This is what makes a bot different from a standard computer program, which needs to have you tell it what to do at various points.

Where do Bots come from?

Significant events in the Bot timeline are these:

1958: Pandemonium, concept by Roger Selfridge for a program that would use "demons" to wait for discrete events and automatically respond to them.

1963: The first actual bot. Fernando Corbato created a daemon (automatic program—a bot) for automatically saving files on a mainframe computer after users had worked on them.

1966: Eliza, written by Joseph Weizenbaum. A "chatterbot" that talks to humans like a Rogerian psychotherapist, and the first to achieve a level of impersonation that could (and still can) sometimes fool humans into thinking they were talking with a real human, not with a bot.

1972: Wumpus, a "gamebot," a character (a monster) that lurked initially in MUDs and MOOs and later (c. 1988) graduated to IRC (Internet Relay Chat; virtual rooms where you can meet and chat with other Internet users).

1977: Adventure, an interactive computer game featured bot dwarves and monsters.

?: Descent, a "thiefbot" incorporated into Adventure’s successor, Zork. Descent would steal a player’s weapons and ammunition.

c. 1990: Julia, written by Michael Mauldin, the first service-oriented bot. She could not only chat, but also provide factual answers to user questions on specific topics. Julia can be considered the first "intelligent agent."

1993: World Wide Web Wanderer, created by Matthew Gray, the first widely recognized Web robot, whose purpose was to contact all the Web servers in the world so they could be counted.

1993: WebCrawler, created by Brian Pinkerton, went a huge step further than Wanderer in that it not only visited every available Web server but also retrieved and indexed every Web page on the servers, noting the topic and URL, thus enabling a WebCrawler user to quickly search through the index for information and then link to the server.

1993: ARMM, written by Dick Depew, was a "cancelbot" designed with good (though debatable) intentions—to prevent anonymous users posting messages in his Usenet newsgroup. Usenet purposely allows messages to be canceled so that posters can change their minds about leaving a message up. Cancelbots automate the process based on some aspect of the message; its sender, its topic, etc. Unfortunately, a later version, ARMM5, had a bug that caused it to "spew" cancellation messages by the truckload into the very newsgroup—news.admin.policy—set up for ensuring the smooth running of Usenet!

1996: Bartender bot on AlphaWorld, one of the first 3-D virtual worlds. Code was nearly identical to Eliza’s.

1996: Scooter, brainchild of Louis Monier of DEC/AltaVista and the engine behind the AltaVista Web search service. It indexes not just the titles of Web pages, but the full text, making it more impressive than WebCrawler by orders of magnitude. "Smarter than Lycos, faster than WebCrawler, more robust than Wobot," says Leonard.

1996: RoverBot, from GlobalMedia Design, was the first mass-marketed "spambot," a bot that would visit Web servers, cull any email addresses it found there, and construct a mailing list for "spammers"—companies that want to send you junk email advertising their product or service. As Leonard rightly says, spambots are "perhaps the most hated genus of bots in the entire bot catalog."

How do they evolve?

The key evolutionary variable in terms of determining fitness, says Leonard, is habitat. "The better a species is fitted to its habitat, the more likely it is to prosper and reproduce." The bot habitat is limitless cyberspace, "the sum total of all the interconnected computers in the world—the ever-growing, ceaselessly squirming, restlessly twisting Internet, as fertile today as the primeval Earth."

"Natural selection is not at work on these nonorganic beings in a state of not quite nature [i.e., bots]. Humans both propose and dispose. Bots cannot upgrade themselves yet. To survive, bots must fulfill some purpose for humans, as a tool or as a plaything, amid the bot environment."

Spambots are a case in point. "Like pests that evolve immunities to each new onslaught of deadly insecticide, new spambots, more powerful, wily, and obstinate than their predecessors, mutate in response to each new obstacle placed in their way," says Leonard. But they don’t do this spontaneously; they evolve only in response to human programming and greed memes.

According to Leonard, Prof. Ken Schweller, of Buena Vista University and designer of MrChat, a user-programmable chatterbox, "estimates that some three to four hundred [chatterbots] currently hold forth in various nooks and crannies of the Net."

By the mid-1990s, IRC "boasted bots that could spellcheck channel conversations, spy on chats and report back to the absentee bot owners, or simply keep logs of everything that happened in a given channel. Virtual bartender bots mixed imaginary drinks and made idle chitchat. Bots played Scrabble, Jeopardy, Boggle, and other games. Help-menu bots offered automated access to information on topics: they could tell you how to prevent a rape, steer you to the latest warez—[aficionado slang for] pirated illegal software—or even deliver that software to you automatically, no questions asked. Bots could link channels together on different IRC networks or convert Fahrenheit temperatures into Centigrade and back. In channels frequented by devout Christians, biblebots responded to keywords with chapter and verse from the King James Bible—in multiple translations."

"In the annals of bot evolution, IRC in the mid-nineties will be remembered by paleobotologists as a modern-day equivalent to the Cambrian explosion—a relatively short period of time 570 million years ago that spawned more new species of life than ever before or since."

Where are bots going?

"The most recent advances in genetic algorithm programming promise an imminent world in which computer programs do spawn offspring and mutate out in the wild, but such creatures have yet to leave the research laboratory to swim free on the net," says Leonard.

"We’ll wake up one morning and ask Jeeves the Bot to make coffee, check the mail, and summarize the news headlines. And we will expect that if Jeeves has any questions, he will ask us for answers; if he has any doubts, he will share them; if he has noticed something unusual, he will alert us, in common English (or Chinese or Hindi or whatever), without stuttering or hesitation (unless appropriate). We will expect intelligent conversation." But today’s chatterbots just don’t make the grade, he says.

Allan Alford's bot Chico pretended to be a foreign exchange student, in which guise it could "get away with making syntactical errors that would otherwise betray [its] robotic DNA", wrote Andrew Leonard. The lesson here, as with Eliza, is that we are easily fooled, and need to be careful about assuming intelligence in machines, soft or hard. As Leonard puts it, "Believability comes cheap" when the human imagination is involved.

"We want to believe in them so much that we’re willing to cut them an immense amount of slack, but by and large we’re fooling ourselves. Despite thirty years of programming practice and countless online recitals, exposing a bot as a mindless string of dumb algorithms is child’s play." This is a little ungenerous. Thirty years is nothing on the evolutionary timescale, and Leonard himself recognizes that bots may not long remain mindless strings of dumb algorithms: "Heuristically determined artificial intelligence is a state-of-the-art research area for search engine operators and a prime requirement for tomorrow’s Web robots," he wrote.

Julia’s creator Michael Mauldin has noticed that "some people find talking to a robot better than talking to no one at all." This was my observation, too, when my colleagues and I created a multi-user version of Eliza and gave her a chat channel on our BBS back in 1990.

"Bots benefit mightily from the innate human tendency to personify." Game or bot designers should care "less about ‘intelligence’ and ‘reality,’" because "When you strip away the protective cover of text and enter a fully 3-D realm of real-time interaction, it becomes more difficult for bots to garner that necessary suspension of disbelief," says Leonard. This may be true, but the way around it is to make the 3-D and the intelligence of the bot so realistic that suspension of disbelief is no longer necessary!

The capabilities of "the primitive Web robots scuttling from hyperlink to hyperlink today," writes Leonard, "are as dust before the avalanche to come. Thousands of researchers all over the world are hard at work constructing new protocols and languages that will allow autonomous programs to communicate with each other, make decisions, react to environmental changes, and carry out tasks. The Net will soon experience the debut of `distributed agents’: swarms of bots acting in concert, updating each other and correlating events all over the Net. These agents are intended to run air traffic control systems, to monitor the flows of capital through financial networks, and to keep factories and telecommunication nodes operating smoothly."

When can we expect that avalanche?

The pace of change

The coming avalanche, the accelerating pace of change, can be seen in the growth of the Web (numbers culled from Andrew Leonard’s book):

May 1993: 100 servers

Nov 1993: 600 servers

Jan 1994: 2,000 servers

Jan 1996: 100,000 servers

Jun 1996: 200,000 servers

Jan 1997: 400,000 servers

The growth in importance of at least one form of bot, the searchbot, can be seen in growth in the number of Web queries conducted via AltaVista (numbers again from Leonard):

Jul 1996: 13 million queries a day

Aug 1996: 16 million queries a day

Sep 1996: 20 million queries a day

Jan 1997: 26 million queries a day

As of January 1997, the AltaVista index contained the full text of 31 million documents and 4 million Usenet news articles.

One last example of the pace of change, this time within IRC, a choice habitat for various types of bot:

1988: IRC servers existed only on three machines in southern Finland.

Mid-1992: 500 people could be found using IRC at any given time of the day or night.

End of 1994: Raise that to 5,000.

End of 1995: 15,000.

1996: 20-30,000, using innumerable IRC servers scattered across the globe.

Nobody planned or organized this. It evolved from a few memes scattered on the Net, principally one meme that took shape out of other memes in the mind and memory banks of Jarkko Oikarinen, the originator of IRC. The same kind of evolution holds for the Internet as a whole. "What rules there are," says Leonard, "occur almost of their own accord, through processes of bottom-up evolution rather than as laid down by fiat from above."

Speaking of rules: At the beginning of this article, we noted Leonard’s statement that "Bots aren’t just cool. They’re trouble." The troublesome side of bots (and AI in general) is discussed in a series of separate articles beginning here.

Until next week,

NEXT WEEK: Machina sapiens and Human Morality.

Previous Features