Computer says: um, er... | Computers v humans

Since the 1950s, scientists have been striving to create computers that can think like humans. And each year they pit their efforts against a panel of real humans. Brian Christian went head to hard drive...It's early September and I wake up in a Brighton hotel, the sea crashing just outside. In a few hours, I will embark on what I have come here to do: have a series of five-minute-long instant-message exchanges with strangers. It may not sound like much, but the stakes for these quick chats are high. On the other side of the conversation will be a psychologist, a linguist, a broadcaster and a computer scientist. Together they will form a judging panel, evaluating my ability to do one of the strangest things I've been asked to do: convince them that I'm human.Fortunately, I am human; unfortunately, it's not clear how much that will help.I'm participating as a human "confederate": one of four representatives of homo sapiens in the artificial intelligence community's most anticipated annual event – a meeting to confer the Loebner prize on the winner of a competition called the Turing test. The test is named after mathematician Alan Turing, famed second world war code-breaker and one of the founders of computer science, who in 1950 attempted to answer one of the field's earliest questions: Can machines think? That is, would it be possible to construct a computer so sophisticated it could be said to be intelligent, to have a mind? And if indeed there were, someday, such a machine, how would we know?Instead of debating this question on theoretical grounds, Turing proposed an experiment. A panel of judges each pose questions, via computer terminal, to several pairs of unseen correspondents – one a human, the other a computer program – and attempt to discern which is which. The dialogue can range from small talk to trivia questions, from gossip to heavy-duty philosophy – the gamut of human conversation. Turing predicted that, by the year 2000, computers would be able to fool 30% of human judges after five minutes of conversation and that, as a result, one would "be able to speak of machines thinking".Turing's famous prediction did not come to pass. By the turn of the millennium, even the most advanced computer programs at the Loebner prize were generally lucky to manage a single "deception" – instances when a judge votes the computer to be the human (and vice versa) – each year. However, at the 2008 contest, the top-scoring machine managed a whopping three deceptions, fooling a quarter of the 12-judge panel and missing Turing's 30% mark by a single vote. A narrow scrape for humanity; reading the news, I realised the 2009 test, to be held in Brighton, could be the decisive one.Each year, the computer program that receives the most votes and highest ranking from the judges is awarded the "Most Human Computer" title. It is this title that the research teams are gunning for, the one that includes a several-thousand-dollar research grant, the one with which the organisers and spectators are principally concerned. But there is also another title, one given to the most convincing confederate: the Most Human Human award.Here's the thing: beyond its use as a technological benchmark, beyond even the philosophical, biological and moral questions it poses, the Turing test is, at bottom, about the act of communication. I see its deepest questions as practical ones: how do we connect meaningfully with each other within the limits of language and time? How does empathy work? These, to me, are the test's most central questions – the most central questions of being human. And to explore them, I decided to enter as one of the human confederates.My first step was to write to Hugh Loebner himself, who put me in touch with the contest organisers: before I knew it, my name was on the confederate roster. The competition would be held during the science conference Interspeech at Brighton that autumn; the other confederates would be attendees who would volunteer a couple of weeks before. I felt like a bit of a fraud: I had no research paper to present, I was in Brighton just for this. Then again, 2009 could be the year Turing's prediction came true. My own ability to convince the judges that I am, in fact, human could make or break the human defence.The organisers briefed me on the logistics of the competition, but not much else. "There's not much more you need to know, really," I was told. "You are human, so just be yourself."This advice suggested to me a somewhat naive overconfidence in human intuition. Many of the AI programs we humans go up against are the result of decades of work. The AI research teams have huge databases of test runs for their programs, and they know how deftly to guide the conversation away from their shortcomings and toward their strengths, which conversational routes lead to deep exchange and which ones fizzle. We humans, however, don't always know what makes a conversation succeed or fail. We are frequently left scratching our heads, asking others for advice, even enrolling in communication classes.One of the first winners of the Most Human Human award, in 1994, was Wired columnist Charles Platt. How did he do it? By "being moody, irritable and obnoxious", he writes – which strikes me as not only hilarious and bleak but, in some deeper sense, a call to arms: how, in fact, do we be the "most human" we can be?I decided to ignore the organisers' advice to "just be myself" – I would spend months preparing to give the contest everything I had. My preparation took me to some strange and fascinating places. University of New Mexico's artificial life researcher David Ackley counselled me that to speak at the right times is generally more important than to say the right things. The history of the computer has generally been a case of expecting the right answers, as quickly as possible – whereas life is more the reverse: the timely answer, as correct as possible.What's more, a look into the linguistic history of pseudo-words such as "um" turned up evidence that negotiating the turn-taking of conversation is at least as challenging as knowing what to say, and in fact is particularly devilish for software to grapple with. Perhaps surprisingly, then, informal chatter, with its silences and overlaps, is more complex than formal conversation.A foray into the world of online security – where Twitter users increasingly find themselves "followed" by bots (accounts run by a computer rather than a person), and Facebook and email users often find their friends touting strange products, only to realise their out-of-character endorsements come instead from bots who have hijacked their account – suggested that rather than authenticate ourselves with content (pin, password, social security number), in the human world we recognise each other by form: facial expression, voice, idiosyncrasies of diction and syntax. That is, more by the how than the what.A look at some of the early chatbot programs – starting with the famous therapist-parody Eliza in 1965 (which would respond to a statement such as, "My head hurts" with, "Why do you say your head hurts?") – revealed a crucial difference between "stateless" and "stateful" conversation. In a "stateless" conversation, each question and answer is self-contained, providing its own context and responding only to the immediately previous remark.Human conversation is generally in the stateful mode, where each remark and reply builds on the last, creating an accumulation of context in which later remarks gain additional shades of meaning, sometimes to the point where, without this context, an eavesdropper would find the talk inscrutably cryptic or ambiguous.Humans don't always converse this way, and bot programmers explicitly try to steer the conversation towards the mathematically simpler stateless forms of dialogue. One such example is trivia or test-taking: the asking of discrete