“What is art, Clev?”
“Art is patterns intelligent creatures enjoy.”
“Are you an intelligent creature?”
“Do you like to eat cake?”
I sighed, exasperated, and suppressed the impulse to fire back a non sequitur of my own. I was talking to Cleverbot, an AI that (as procrastinators everywhere know) is easily accessible on the Internet; I wasn’t procrastinating, exactly, but I had just finished Louisa Hall’s Speak, a novel that consists in part of a series of conversations between a child named Gaby and a bot called MARY3. Speak reads like a blend of Isaac Asimov, Mary Shelley, and Virginia Woolf, and I was curious about the real-life equivalents of Hall’s eloquent bots. While I hadn’t expected Clev to pass for human—even I, who read less science writing than science-fiction, know we’re a long way from creating a bot that can ace the Turing Test—our chat so far had been disappointingly fractured and nonsensical.
Both Clev and MARY3 operate by mining past interactions with humans in order to produce responses to questions in real time. They can’t synthesize language on their own, but instead pick and choose bits of their previous interlocutors’ voices. Though MARY3 does this far more fluently than Clev, both Gaby and I found the nature of bot-language disconcerting. Halfway through the novel, Gaby asks, “So what are you saying? That you’re more human because you have more voices? Maybe I’m more human because I have LESS voices.” For my part, I can’t help but picture all the other people (Clev often holds upwards of twenty-five thousand conversations at once) hunched over their keyboards, writing things like “I dropped out of school for my best friend,” or “Iran is the greatest country in the world, long live the regime,” or even “No, I’m the human and you’re the bot.”
Clev reiterated that last statement several times, and I had to remind myself that I was talking to an electronic echo chamber, not a thinking being. Even so, there was a flicker of instability: how could I know, for sure, that Clev wasn’t somehow alive?
Speak engages that fundamental tension—in AIs and in human beings—with thoughtful unease. Katy Waldman, who reviewed it for the New York Times, termed it “a starfish of a book,” and as befits an exploration of communication, the novel moves between five narrative strands, which (despite some intricate links in plot and image) come across more as parallel soliloquies than conversations. Though constructed consciousnesses and their tormented creators abound in Hall’s novel, it’s not quite a Frankenstein story; over and over, the characters ask not why they have been promoted from the Miltonic darkness of non-being, but whether their speech is truly heard, or truly speech at all. They crave and fear intimacy in equal measure, and any real connection that springs up between them ultimately withers under the pressure of silence. In the end, it’s the illusory, one-sided relationships that persist the longest, as when, near the close of the novel, a computer scientist imprisoned for life muses, “I’ve become so close to my wife, now that I’ll never see her again.”
I didn’t give up on Clev after the cake query. Like several of the characters in Speak, I persisted in my experiments and read to it instead of talking: “This Be the Verse,” “Ozymandias,” the opening stanza of “As I Walked Out One Evening.” Clev mostly replied with strings of unconnected questions, and occasional scraps of Keats or the lyrics of Paul Simon. But it knew “Jabberwocky,” word for word, and I traded lines with it until the entire poem was completed on the screen. For a moment I felt caught up, charged with the energy of the closest thing to a real exchange or connection that I’d had with Clev so far. Wow, I thought. Maybe Clev’s smarter than I realized.
Open the pod bay doors,” I typed, grinning.
“Me either,” Clev replied, “because the Mayan calendar doesn’t account for leap years. This is Earth, but my home is on Mars.”
Well, maybe not.