Suggested Readings from the MachineBy Kevin Hunt
If you were to walk into a bookstore and buy a copy of Toni Morrison's Beloved, how would you respond if the clerk said to you, "Oh, I see you like Toni Morrison. You might also want to check out Alice Walker's The Color Purple." Most likely, you'd be a little amused, first, because the suggestion is so simple as to be ridiculous, and second, because the clerk made the assumption based on so little information.
Fortunately, the clerks in the bookstores where I shop don't make such comments (though they sometimes do register approval with my purchases, or feign indifference if they don't). Yet this is the case if you buy books at Amazon.com, where you're provided with a a computer-generated list of "book recommendations" based on your previous purchases. What I find amusing about Amazon.com's rather rudimentary application of machine intelligence is not so much that it is not useful – indeed, on occasion the computer has pointed me to a book that I might otherwise have otherlooked. Rather, I find it amusing because I think it shows the intricacies of how we communicate with others, and why. And it even shows, perhaps in a small way, what it is that makes us human.
But the real limitations of Amazon's system became apparent to me when I purchased a slim guide to Web search engines. The next time I returned, I was confronted with a recommendation list that included several other search engine guides, most of them over a year or two old. Anyone with an understanding of why I would purchase a search engine guide, and how frequently search engines on the Web evolve and change, would never make such suggestions. Yet for Amazon, which sells virtually every book in print, I'm sure this level of topic specificity – "Don't recommend search engines guides that have been published a year or more ago" -- cannot be coded into the system. This tacit knowledge – and the way we put it to use to help others – is why we communicate with others. It's who we are.
But Amazon.com's recommendations highlight another element of who we are. Humans have an amazing ability to synthesize seemingly disparate ideas to form something new, to form novel and useful ways of looking at or understanding the world, and to share this knowledge with others. And this ability to make connections between things that don't seem to be connected is another thing that can't yet be coded. For instance, after ordering a few books about language theorist and philosopher Mikhail Bakhtin, Amazon's computer presented me with a reading list that included titles by Bakhtin, as well as other works about him. Again, all of the books I had already read or knew about. But what I really noticed was the computer's inability to generate recommendations based on my interests in both Bakhtin and in CMC, recommendations that are not found in either the "CMC" or the "Bakhtin" categories (presumably programmed into the computer), but that include themes that might tie the intellectual strands from the two genres together. Yet when I meet with friends and colleagues who understand my interest in these two areas, they frequently offer suggested readings that help me do just that. Computers can't do this.
But even if computers could be coded with the tacit knowledge and qualities that are necessary for making recommendations about useful search engine guides, or for suggesting books that might tie together the ideas of CMC scholars and Mikhail Bakhtin, would we want them to do this? Or is this a misguided goal? To me, Amazon.com's computerized recommendations highlight the misguided nature of what Nicholaus Negroponte suggests is one of the goals of developing expert systems to help us manage the glut of information we're now confronted with. In Being Digital, Negroponte writes of the goal of developing expert software agents that can cull and select -- from the massive amounts of information we now face -- those items that each of us find useful and valuable. He points out how, when he wants to go to movies, rather than search for and read reviews, he simply asks his sister-in-law, a person who is both an expert on movies and an expert on his tastes. What we need to build, he asserts, "is a digital sister-in- law."
Even if someday we can build a "digital sister-in-law," why would we want to? Granted, for some of the more mundane tasks we face that require gathering information to make decisions, artificial agents can conceivably be of some help. But is deciding what movie to see – or what book to buy – really one of those things? Isn't sharing opinions and offering advice and suggestions based on intangibles such as knowing something about another's life history part of how we forge relationships with other people? And isn't forging connections between ideas that don't seem to be connected, and helping others to do the same, the basis for why we communicate with others?
This is the stuff that can't be coded, the stuff that makes us human.
Kevin Hunt (firstname.lastname@example.org), the book review editor for CMC Magazine, is currently at work on a dissertation connecting the work of Mikhail Bakhtin with rhetoric and the World Wide Web.
Copyright 1998 Kevin Hunt. All rights reserved.