Political middlemen and dart-throwing chimps

Martha Gill's "Irrational Animals" column.

Predicting the weather was once quite an interesting profession, needing skill in reading the instruments, intuition in deciphering the skies and years of experience in putting it all together. Now it’s the kind of job Nick Cage’s character would be given in a heavy-handed satire of the American dream, possibly also starring Michael Caine. We don’t need these skilled individuals any more – computers do all that. We just need an algorithm and a mouthpiece.

And so to Nate Silver – one of the biggest winners of the US presidential election. As the race neared its end, becoming “too close to call”, with money and opinions frantically changing hands, the New York Times blogger was calmly and correctly predicting voter outcome in every single state. He had what others didn’t – a formula to convert polling information into probabilities – and it turned out to be dead-on. He was not alone in getting it right but he was among the few. Many failed spectacularly.

Here’s Newt Gingrich on Fox News on 25 October: “I believe the minimum result will be 53-47 [per cent] Romney, over 300 electoral votes, and the Republicans will pick up the Senate. I base that . . . on just years and years of experience.” And here’s the GOP strategist Karl Rove in the Wall Street Journal on 31 October: “It comes down to numbers. And in the final days of this presidential race, from polling data to early voting, they favour Mitt Romney.”

These were not small errors. These people were standing in pre-hurricane wind and predicting sunshine. Are pundits more often wrong than not, or was it just this particular election that threw them? And how often do the statistics spewed out by experts hit the mark? One study found a statistic for it.

Algorithm blues

In the 1980s, a psychologist called Philip Tetlock took a group of journalists, foreign policy experts and economists – 284 of them – and spent the next two decades bombarding them with questions: would the dotcom bubble burst? Would George Bush be re-elected? How would apartheid end?

After analysing 82,361 predictions, Tetlock found that his experts performed worse than random chance. In short, they could have been beaten by dart-throwing chimps.

The reason was confidence. Tetlock found that the more often pundits appeared on TV, the more likely they were to be wrong. Their strong opinions were causing them to ignore dissenting facts or explain them away, leaving them trapped, he said, in the cage of their preconceptions.

Now, semi-expert middlemen are being squeezed out as the focus shifts to minute data analysis. Silver is one of the winners of this change but on the losing side is a whole industry of political forecasters. And it’s not just true of politics. Finance has been moving that way for a while. In UBS’s recent swath of job cuts, at least one trader, David Gallers, was replaced with an algorithm.

Difficult times for the old school, but what of the new? Silver expressed his concerns to the Wall Street Journal: “You don’t want to influence the system you are trying to forecast.” Only one problem with the new machines, then – accuracy. They’re so good that they might start controlling the weather.

Newt Gingrich opining away on Fox News. Photograph: Getty Images

Martha Gill writes the weekly Irrational Animals column. You can follow her on Twitter here: @Martha_Gill.

This article first appeared in the 19 November 2012 issue of the New Statesman, The plot against the BBC

Show Hide image

Should we protect artificial intelligence from sexual harassment?

Should anything be done to stop people sending sexually explicit messages to their AI personal assistants?

If you ask Apple’s artificially intelligent personal assistant “Siri” whether it is a virgin, it will waste no time in shooting you down. “We were talking about you, not me,” it replies in the clear, sharp tones of Susan Bennett, the woman chosen to voice the genderless computer program.

If you ask Apple’s artificially intelligent personal assistant “Siri” whether it is a virgin, you are probably not very weird. But a recent article in Quartz has detailed the extent to which AI systems – particularly personal assistant bots – are sexually harassed. Ilya Eckstein, CEO of Robin Labs, claims 5 percent of interactions in their database are sexually explicit, and that “some people try very hard to establish a relationship with the bot.”

Engineers have been aware of this problem for a while. Microsoft’s Cortana has been programmed to fend off sexual harassment, with Deborah Harrison, an editorial writer for the program, claiming: “If you say things that are particularly asshole-ish to Cortana, she will get mad.” But what about the other “female” AI out there? Amazon’s Alexa and Google Assistant, which is voiced by a woman, don’t currently seem to fend for themselves, so should we be fighting for them?

Probably not. Although developers should definitely program their “female” AI to shoot down anyone feeling frisky, as long as AI lacks sentience it’s hard to see these sexual interactions as a big enough problem to warrant further action. Yes, undoubtedly some lonely people have taken inspiration from Spike Jonze’s Her and fancy an AI girlfriend, and yes, a robust robot reply that teaches men to respect women can only be a good thing, but on the whole, most people that get saucy with Siri aren’t actually deranged perverts. They are just a boy, standing in front of a girl, asking them to say the world “willy”.

This is because despite what Quartz are claiming, the “sexual harassment” of bots is nothing new. It might, in fact, not even be gendered. Who among the MSN users of the Noughties didn’t ask the chatterbot SmarterChild whether he (most people, and media outlets, considered it a “he”) liked sex or had a penis? In fact, if you search Google Images for “Smarterchild”, pretty much all the screencapped chats are sexually explicit in some way.

Tumblr: The Dynamic Conversationalist

It’s hard to see someone sexting Siri as a problem, then, because it is part of a long tradition of humans being incredibly, incredibly dumb. Find me the man who doesn’t provoke every new chat bot on the market in the hopes of making them say something funny or rude, and you have found me a liar.

It is, of course, a big problem that AI personal assistants are so often female, as – in Laurie Penny’s words – it “says an uncomfortable amount about the way society understands both women and work.” But this, therefore, is the problem we should be tackling – instead of wasting our time debating the ethics and legality of coming on to Cortana.

I recently attended the UK launch of Amazon Echo, whose personal assistant is Alexa. Watching a room of old, balding, white, male journalists laugh heartily as the speaker on stage commanded Alexa to “Stop”, definitely troubled me. “If only I could get my…” began the speaker – as I desperately willed him not to say the word “wife” – “…children to do that,” he finished. Before we even begin to consider sexually explicit chatter, then, we should be confronting the underlying issue of gender bias in the AI industry.

Once we can set our personal assistants to have either male or female – or, even better, completely genderless – voices, we can get back to using them for what they were intended for. Asking them if they're virgins and then laughing at the response.

Amelia Tait is a technology and digital culture writer at the New Statesman.