Category Archives: AI

book machine

Philip M. Parker, a professor at Insead, the international business school based in Fontainebleau, France, has written 85,000 books and counting. He’s like a machine. In fact, he has a machine that writes them for him. The Guardian has more.
Most, if not all, of these books can be found on Amazon. Sifting through them felt like a bad riff on “The Library of Babel.” I felt like I’d stumbled upon a weird new form of bibliographic spam -? thousands of machine-generated titles gumming up the works, jamming the signal, eroding the utility of the library. Matt Kirschenbaum, who forwarded the link, said it recalled the book machines in Italo Calvino’s great meta-novel, If On A Winter’s Night a Traveler:

He has you taken into the machine room. “Allow me to introduce our programmer, Sheila.”
Before you, in a white smock buttoned up to the neck, you see Corinna-Gertrude-Alfonsina, who is tending a battery of smooth metallic appliances, like dishwashers. “These are the memory units that have stored the whole text of Around an empty grave. The terminal is a printing apparatus that, as you see, can reproduce the novel word for word from the beginning to the end,” the officer says. A long sheet unrolls from a kind of typewriter which, with machine-gun speed, is covering it with cold capital letters.

Prices are often absurdly inflated, up to the many hundreds of dollars. While, on Amazon, you can’t peek inside any of the books, the product descriptions read like prose recycled from free government business or health leaflets (stuff that usually feels like it was written by a machine anyway). There seem to be a few dozen tropes which are repeated with slight variations ad nauseum. A few sample titles:
-? The 2007 Report on Wood Toilet Seats: World Market Segmentation by City (330pp., $795)
-? The 2007-2012 Outlook for Lemon-Flavored Bottled Water in Japan (140pp., $495)
-? Avocados: A Medical Dictionary, Bibliography, and Annotated Research Guide (108pp., $28.95)
-? Brain Injuries – A Medical Dictionary, Bibliography, and Annotated Research Guide to Internet References (244pp., $28.95)
In fact, there’s a whole trope of titles that are guides to “internet references,” which makes me wonder if Parker’s machine is just scraping the entire Web for content.
Odd.

google flirts with image tagging

sawyer fence.jpg Ars Technica reports that Google has begun outsourcing, or “crowdsourcing,” the task of tagging its image database by asking people to play a simple picture labeling game. The game pairs you with a randomly selected online partner, then, for 90 seconds, runs you through a sequence of thumbnail images, asking you to add as many labels as come to mind. Images advance whenever you and your partner hit upon a match (an agreed-upon tag), or when you agree to take a pass.
I played a few rounds but quickly grew tired of the bland consensus that the game encourages. Matches tend to be banal, basic descriptors, while anything tricky usually results in a pass. In other words, all the pleasure of folksonomies — splicing one’s own idiosyncratic sense of things with the usually staid task of classification — is removed here. I don’t see why they don’t open the database up to much broader tagging. Integrate it with the image search and harvest a bigger crop of metadata.
Right now, it’s more like Tom Sawyer tricking the other boys into whitewashing the fence. Only, I don’t think many will fall for this one because there’s no real incentive to participation beyond a halfhearted points system. For every matched tag, you and your partner score points, which accumulate in your Google account the more you play. As far as I can tell, though, points don’t actually earn you anything apart from a shot at ranking in the top five labelers, which Google lists at the end of each game. Whitewash, anyone?
In some ways, this reminded me of Amazon’s Mechanical Turk, an “artificial artificial intelligence” service where anyone can take a stab at various HIT’s (human intelligence tasks) that other users have posted. Tasks include anything from checking business hours on restaurant web sites against info in an online directory, to transcribing podcasts (there are a lot of these). “Typically these tasks are extraordinarily difficult for computers, but simple for humans to answer,” the site explains. In contrast to the Google image game, with the Mechanical Turk, you can actually get paid. Fees per HIT range from a single penny to several dollars.
I’m curious to see whether Google goes further with tagging. Flickr has fostered the creation of a sprawling user-generated taxonomy for its millions of images, but the incentives to tagging there are strong and inextricably tied to users’ personal investment in the production and sharing of images, and the building of community. Amazon, for its part, throws money into the mix, which (however modest the sums at stake) makes Mechanical Turk an intriguing, and possibly entertaining, business experiment, not to mention a place to make a few extra bucks. Google’s experiment offers neither, so it’s not clear to me why people should invest.

kurzweil’s techno-narcissism

Ray Kurzweil looks into the future and sees the singularity gazing back full of love. It whispers. It seduces. “Ray, take care. Preserve yourself. It will be another 50 years yet. Go. Preserve yourself with vitamins, fruits, infusions. Keep your body tender and vital, and soon enough you will be subsumed, you will transcend. The singularity is near!”
singularity.jpg Kurzweil’s book is out and it’s as big as a dictionary. A good friend of mine was given it as a gift a couple of nights ago for his birthday. After dinner, as we rode the crosstown bus toward a game of cards, I read the first few pages. Try holding this goliath in one hand! The bus was crowded and we were standing in the aisle, gripping the handles on the top rail. The bus lurched, and I cursed my physiognomy. If only I could download the damn thing into my brain! If only the singularity were here now!
Kurzweil’s theory, or rather, his unshakeable conviction, his messianic belief, is that we, the human species, are nearing the point (he predicts around 2045) when our tools will become more intelligent than us and we will merge – mentally, biologically, spiritually – with them. Computer processing, artificial intelligence, biotechnology are all developing at an exponential rate (the law of accelerating returns), and are approaching a point of singularity, an all-encompassing transformative power, that will enable us to eliminate poverty, eradicate hunger, and “transcend biology.”
The reason Kurzweil is preserving his body – “reprogramming his biochemistry,” as he puts it – is because he is convinced that in about a generation’s time we will be able to ingest millions of microscopic nanobots into our neural pathways that will turn our brains into supercomputers, and engineer ourselves to live as long as we please. We will become, to borrow a conceit from an earlier book of Kurzweil’s, “spiritual machines.”
I would like to say that I will take the time to read his book and engage with it in more than a passing (and admittedly reactionary) way. Perhaps we’ll make a project of reading Kurzweil here at the institute as a counterpoint to Neil Postman (see recent discussion). But I’m not sure how much of his flaming narcissism I can take. Kurzweil’s ideas of “transhumanism” are so divorced from any social context, so devoid of any acknowledgment of the destructive or enslaving capacities of technology, and above all, so self-involved (the fruit and vitamin regimen is no joke – and there is probably a black monolith at the foot of his bed), that I’m not quite sure how to have a useful discussion about them.
As an inventor, Kurzweil has made many valuable contributions to society, including text-to-speech synthesis and speech recognition technology that has greatly aided the blind. It is understandable that his success in these endeavors has instilled a certain faith in technology’s capacity to do good. But his ecstatic, almost sexual, enthusiasm for human-machine integration is more than a little grotesque. Kurzweil’s website and book jacket are splashed with approving quotes from big name technologists. But I don’t find it particularly reassuring, or convincing, to know that Bill Gates thinks

Ray Kurzweil is the best person I know at predicting the future of artificial intelligence.

For a more reasoned, economic analysis of the possible outcomes of accelerating returns, read John Quiggins’ “The singularity and the knife-edge” on Crooked Timber. Another law – or if not a law, then at least a common sense suspicion – is that if the engine keeps accelerating and heating it up, it will eventually fall apart.

“finally, I have a Memex!”

There’s an essay worth reading in the ny times book review this past sunday by Steven Johnson about a powerful semantic desktop management and search tool recently released for Macs. vannevar.gif The software (called DEVONthink) not only helps organize and briskly sift through readings, clippings, quotes, and one’s own past writings, but assists in the mysterious mental processes that are at the heart of writing – associative trains, useful non sequiturs, serendipitous stumbles. In effect, we now have a tool resembling the Memex device described in the seminal 1945 essay, As We May Think by visionary engineer Vannevar Bush. Working with the cutting edge technologies of his day – microfilm, thermionic tubes, and punch, or “Hollerith,” cards – Bush pondered how technology might help humanity to manage and make use of its vast systems of information. His recognition of the basic problem is no less relevant today: “Our ineptitude in getting at the record is largely caused by the artificiality of systems of indexing.” Fast forward to 2005. Now, the holy grail of search is the Semantic Web – moving beyond the artificiality of crude content-based queries and bringing meaning, relevance, and associations into the mix.
“Consider a future device for individual use, which is a sort of mechanized private file and library. It needs a name, and to coin one at random, “memex” will do. A memex is a device in which an individual stores all his books, records, and communications, and which is mechanized so that it may be consulted with exceeding speed and flexibility. It is an enlarged intimate supplement to his memory.” – Vannevar Bush

memex-1.jpg

It’s quite suggestive that DEVONthink’s semantic search function can to an extent be trained, taking the obnoxious little puppy on Windows search toward its full potential – a sleek, truffle-tuned hound. When Johnson loads his body of work onto the computer, the hound picks up the distinctive scent of his writing, which in turn suggests affinities, similarities, and connections to other materials – truffles – that will find their way into later works.
20truf-dog.jpg Says Johnson on his latest blog post, which goes into much greater detail than the Times piece:
“I have pre-filtered the results by selecting quotes that interest me, and by archiving my own prose. The signal-to-noise ratio is so high because I’ve eliminated 99% of the noise on my own.”
But it is significant that DEVONthink is not useful for searching entire books (the author’s own manuscripts notwithstanding). Currently, the tool is ideal for locating chunks of text that fall within the “sweet spot” of 50-500 words. If your archives include entire book-length texts, then the honing power is diminished. DEVONthink is optimal as a clip searcher. File searching remains a frustrating enterprise.
Johnson makes note of this:
“So the proper unit for this kind of exploratory, semantic search is not the file, but rather something else, something I don’t quite have a word for: a chunk or cluster of text, something close to those little quotes that I’ve assembled in DevonThink. If I have an eBook of Manual DeLanda’s on my hard drive, and I search for “urban ecosystem” I don’t want the software to tell me that an entire book is related to my query. I want the software to tell me that these five separate paragraphs from this book are relevant. Until the tools can break out those smaller units on their own, I’ll still be assembling my research library by hand in DevonThink.”
Another point (from the Times piece) worth highlighting here, which relates to our discussion of the networked book:
“If these tools do get adopted, will they affect the kinds of books and essays people write? I suspect they might, because they are not as helpful to narratives or linear arguments; they’re associative tools ultimately. They don’t do cause-and-effect as well as they do ‘x reminds me of y.’ So they’re ideally suited for books organized around ideas rather than single narrative threads: more ‘Lives of a Cell’ and ‘The Tipping Point’ than ‘Seabiscuit.'”
dog.gif And what about other forms of information – images, video, sound etc.? These media will come to play a larger role in the writing process, given the ease of processing them in a PC/web context. Images and music trump language in their associative power (a controversial assertion, please debate it!), and present us with layers of meaning that are harder to dissect, certainly by machine. It is an inchoate hound to be sure.