Passing the test: The Most Human Human, by Brian Christian
By Adam Sell
The Bot Book Review is an occasional series of short reviews of robot-themed books as written by HUMAN team members.
When we went through the exercise last year of finding the most appropriate moniker with which to reanoint ourselves, we test-drove a bunch of options to see how they felt before we dug too deeply into whether they’d work on a practical basis. We tried some versions of Ops, referencing the name we were leaving behind; we played with different versions of the word ‘human’, which is where we ended up. But we also kicked around some versions of the name and concept of Turing, which reflects the very nature of what it is we do: we find bots on the internet pretending to be humans and identify them as what they are.
The Turing test, though, isn’t solely a thought experiment in finding the inflection point at which artificial intelligence crosses some boundary of humanness. It’s also a real-life exercise wherein researchers in artificial intelligence put their creations to the test. Every year until 2019, the Loebner Prize was awarded to two different entities - the developers of the computer/robot that convinced the greatest number of judges of its humanity, and to the human control subject—the confederate, in the terms of the competition—who convinces the greatest number of judges of their humanity. (In 2019, the structure of the competition shifted away from individual judges and toward a broader internet vote, absent the human confederate control subjects.)
This brings me to the subject of the book for this month’s review. The Most Human Human is the firsthand account of Brian Christian, selected as one of the confederates for the Loebner Prize competition in 2009. Christian took his role as a representative of humanity in a perceived war of existence very seriously, with the overwhelming majority of the book dedicated to exploring just what approaches to proving one’s humanness would be most effective.
It’s worth noting that the Wikipedia link above explaining the Loebner Prize competition also makes note of the competition’s critics, who decry the endeavor as a publicity stunt, or who suggest that the parameters on the contest put too many restrictions on how a judge might determine bot or not.
Christian answers these questions in the course of his narrative, with the comment that the longer a conversation continues, the likelier an artificial intelligence conversation bot will slip up in some way. Christian recounts some text snippets that would have been apparently human if the conversation had stopped a line or two sooner, whereupon the bot answered a question in a bizarre and telling fashion.
The book went far deeper into philosophy and navel-gazing as a subject than I expected for a book about whether you could prove you were a human better than a bot could. I believe at one point I even tweeted a quote from the book that I found to be especially “okay, shut up, nerd”. Never before have I needed to google so many phrases and terms around epistemology and other meta-knowledge topics simply to gain a basic understanding of an author’s point.
What I resonated far better with were the explorations of how certain tactics might work or backfire in a Loebner Prize context. Is it better to dive into one topic and plumb the depths of detail and nuance, or make like the most controversial Jeopardy champions and hop around at will? And if humans playing humans know this, then the humans writing robots playing humans know this, and you’ve got yourself a prisoner’s dilemma.
It made me think of the ways our detection team uses technical intelligence to spot bots requesting ads or trying to log into web apps in a fraction of a second: what’s the calculus that the fraudster uses to figure out their plan of attack, and how does that differ from the calculus our detection team uses to find them? And since they’re both playing this odd cat-and-mouse game, how do you know when to double down and how do you know when to back off?
Brief spoiler alert, the bots did not win in 2009. Christian himself wins the title which he shares with the book (that of the Most Human Human), and we haven’t yet reached either the singularity or the robot apocalypse. And even with a reduced threshold (30% of human judges convinced), no AI has yet beaten the Turing test. We’re 13 years removed from Christian’s episode with the Loebner Prize and we’re either no closer to true artificial intelligence, or real human intelligence has kept pace.
All that is to say, my money’s on humans.
The Most Human Human
By Brian Christian
Bot Book Review Rating: 4/5 stars