"WE KNOW BOOKS," promises a welcome banner atop the home page of Bookish.com, which launched last week after months of delays and three CEOs. This motto is, to a certain extent, demonstrably true. The venture is backed by Hachette Book Group, Penguin Book USA, and Simon and Schuster, and its editors' resumes include stints at those and other publishing houses. But it's less clear what Bookish is, exactly. It's partly an e-commerce site, selling books published by the above three, and partly a lit mag, offering author interviews, original essays, and reader reviews; the awkwardness of this relationship is made apparent by the fact that the site's declaration of editorial independence can be found ... on its "Corporate" page. But Bookish's main draw—or so it hopes—is its recommendation engine, which is fueled by "proprietary technology to provide a book-centric, contextual and personalized experience": "We serve smarter book recommendations," the site claims. Karen Sun, Bookish's "recommendations data scientist," says it's "an exercise in big data."
So is their algorithm the holy grail of book recommendations? At registration, you're asked to pick genres ranging from "Fiction & Literature" to "Mind, Body & Spirit," reveal some basic personal information, and then the search is on. You can select the authors you "love," and rate books on a five-star scale, thereby building up a database of your reading history and preferences—the idea being that the algorithm will (eventually) churn out titles that you haven't read but which match your sensibilities. In the blithely promotional words of CEO Andy Khazaei, who took over four months ago, "this will be a better way for people to discover new books, because recommendations from friends are not necessarily the best way to find a match for a reader's tastes." His unspoken hope, of course, is that better recommendations mean more book sales.
Bookish is hardly the first site to lure readers with the promise of perfectly calibrated recommendations (though it is unique in its employment of editors—real humans—who are empowered to supply recommendations as well, though its not clear in what capacity). Amazon, perhaps the gold standard in the industry because of its unparalleled data set, has long offered recommendations in the form of matching books that were "Frequently Bought Together" and the "Customers Who Bought This Item Also Bought" carousel. Assiduously tracking its customers' habits, Amazon may know more about my purchasing patterns than my boyfriend does—and let's keep it that way—but its recommendation engine has two crucial failings: Amazon does not know about books I buy or obtain in other capacities, and it assumes that I'm voracious for more from authors I've already purchased. Anyone who has ever read one Anthony Trollope novel knows this cannot possibly be true.
Amazon knows, for instance, that in the past I've purchased Middlemarch, and Jon Klassen's This Is Not My Hat, and Barbara Kingsolver's The Bean Trees. It also knows that I've recently searched for James Salter, and Nathaniel Hawthorne, and Vladimir Nabokov. What Amazon doesn't know is that Middlemarch was a replacement for a never-returned older copy, Klassen was a one-off gift for my nephew, and that I wanted to throw Kingsolver's book out the nearest window. Yes, I searched for Salter because I think he's a literary genius and want to read every word that man has written, but the Hawthorne was for research and the Nabokov was simply to ooh and ahh at the pretty new covers. So Amazon keeps asking if I'd like to purchase Pnin and Lolita, both of which I already own, and only one of which I care to read. What Amazon does not (and cannot) do, is understand how keenly I envisioned George Eliot's small British town on the cusp of industry in Middlemarch, and thus place in my hands Elizabeth Gaskell's North and South, a novel that struggles with the same ideas, and may have informed Eliot's sensibility. While it may contain a library of sorts, Amazon is not, alas, a librarian.
Six-year-old Goodreads, which claims to be the Internet's "largest site for readers and book recommendations in the world," is the web's most popular place for dedicated readers to share and browse shelves, and to (thrillingly, voyeuristically) dip into the reading lives of our friends and neighbors. Its tagline is "Meet Your Next Favorite Book," and the site implores you, "Tell us what titles or genres you've enjoyed in the past and we'll give you surprisingly insightful recommendations." I visit Goodreads regularly, though perhaps not in the way its makers intended: For me it's a catalog of sorts, a record of what I read and when—a virtual catalog, essentially, of my dusty library.1 For many users, though, Goodreads is a digital neighborhood bookstore, a place where chatter—and sharing book recommendations, specifically—is encouraged.
By many measures, Goodreads is a delightful site, peopled by thoughtful readers who offer careful analysis and clear-eyed reviews.2 But its recommendation engine relies on such broad, sweeping categories as to render it nearly useless. Over the past four years, I've entered over 500 of my reads into its system, yet the site can still only tell me that I enjoy "Classics," and thus, it foists Turgenev and Defoe upon me at every turn. No matter how many opportunities I give Goodreads to better understand my taste and desires, it sees the Brontes and Melville as contemporaries and therefore correlative. And for reasons I cannot even begin to understand, the site is determined that I read every word that spilled from George Gissing's pen.
As brick and mortar bookstores continue to shut their doors—even the oft-reviled Barnes & Noble will be missed—it's inevitable that more literary communities will emerge online. Bookish, with its original content, retail functionality, and social-media hyperconnectivity, could become a hub for book lovers—but not with its current, pitiful recommendations engine, which offered me Hilary Mantel's historically precise and vivid Bring Up the Bodies when I wanted something similar to Richard Ford's placid Canada. Other than the books' lovely, award-winning writing, it's hard to see any similarities. I entered Jane Eyre and Bookish spat back the AIA Guide to New York City. Unless one considers Jane Eyre's Thornfield Hall scenes an homage to architecture, I'm not certain what the correlation could be. As for Animal Farm, Bookish smartly (though with little originality) recommended One Flew Over the Cuckoo's Nest, A Clockwork Orange, and Lord of the Flies. But their last recommendation left me a bit confused: Matar un Ruiseor—that's Harper Lee's To Kill a Mockingbird in Spanish. I don't speak Spanish.
The irony of these book recommendation engines is that they seek to solve a problem that society doesn't have. Smart, high-functioning recommendation engines already exist, and they're readily available online and in print: They're called Dwight Garner, and Robert Silvers, and that long wooden table at the entrance to Kramerbooks, or whatever your local independent bookstore is named. I have two personal engines of my own, my colleagues Chloe Schama and Leon Wieseltier, each of whom contains millions of bits of bookish knowledge and knows my tastes and curiosities. It is because of them, and not some algorithm, that I've delved into Salter, and Eliot's essays, and Muriel Spark. No engine can ever know, as Schama and Wieseltier do, that I bizarrely prefer books about cold places, and will always pass up a novel that takes place too close to the equator. Khazei, Bookish's CEO, has said that my friends and relatives "won't be able to know about as many relevant books as our tool can." That may be true. No human brain can retain and immediately access that amount of information. But the point is moot: One doesn't need scores of recommendations, one needs the right recommendations. Online recommendation engines are not inherently useless. They are indeed fast and convenient, and some more than others provide a certain community. They're also inevitable. Much of our societal dialogue has moved online, and book chatter is no exception. Much of our commerce has moved online, too, and there's no bigger example of that than Amazon. It's only natural that, as happened before the Internet, we congregate to chat about books in the same places where we buy books. Reading remains one of the most solitary human acts, but now, we needn't wait for our roommate to come home to share our thoughts about the languid beauty of David Guterson's prose, or the sheer terror of Sarah Waters's The Little Stranger. Sites like Bookish offer immediate gratification: Love the author, rate her book, and start a conversation about it—and then, yes, search for a recommendation.
But there's a reason book clubs are going strong—offline. No matter how heartily an algorithm endorses, say, Thornton Wilder, it remains limited by its composition. Data is entirely a collection of externalities; it can collect and sort millions of user preferences and similarities, but it can never move beyond the what to the why. Data has no imagination. When it comes to book recommendations, attempts to sort or streamline or mathematize them necessarily dehumanize the process. The very nature of the endeavor, much like digesting Ulysses, requires an infinitely more complex machine: the human brain.
A record of what I want to read, meanwhile, is kept on my iPhone: I take pictures of book covers, and then flip through the photo album when I'm ready for a new book—hardly efficient, I know.
Well, usually. Twilight has an average rating of 3.59 out of 5, while Virginia Woolf's Jacob's Room—a novel that arguably ushered in Modernism—has a 3.53.
Hillary Kelly is the digital media editor at The New Republic.