Here’s a thought experiment: Imagine that an Artificial Intelligence Party (AIP) stood in every UK constituency and that its AI candidates were eligible to take up those seats in Parliament if they won. Would AI be a better choice than a human politician? Could AI solve Brexit?
The idea that an AI politician would be somehow free of the biases that beset humans is a bad place to start. An AI politician would be bound by the assumptions imposed on it by its human creators and its actions would be governed by the biases inherent in the data used to train it.
On Brexit, for example, what data sources would the AI draw upon to make policy decisions? And where would it look to for precedent? There’s no existing example or previous outcome for the AI to extrapolate from. Its view — Remain, Leave or some third option — would be as open to criticism from opponents as a position taken by a human politician. Imagine Nigel Farage lambasting the Remainiac algorithms beholden to their EU masters.
Perhaps an AI politician could simply be trained on previous parliamentary voting data combined with a huge and constantly updated data set of public polling. But in those circumstances, you’d be left with a model that’s unable to put forward radical new policies, but instead simply delivers what it calculates that the majority expects. For instance, when politicians abolished capital punishment in the UK, the decision went against the prevailing public opinion. Similarly, an AI politician — with no sense of identity — would be unlikely to pursue policies that considered or protected the rights of minority groups.
There’s already plenty of evidence that big data sets contain destructive racial, sexual and gender bias. AI politicians would carry those biases with them as well as not having the benefit of lived human experience to draw upon. Data alone is not enough to create good policies.
It’s often been said that politicians vote for wars because their own children are unlikely to have to fight them. In the case of AI politicians, they’d have no stake in the policies they put forward. It’s unlikely that a human electorate, no matter how frustrated it purports to be with flesh and blood legislators, would be happy with the cold pseudo-rationality of an AI parliament.
In the 2018 Russian presidential elections, Yandex put virtual assistant Alisha up as a candidate. Its argument was that the bot, “built upon deep neural networks trained on massive data sets”, would be faster and smarter than a human politician. But the company’s boast that Alisha “listens to everyone and changes her politics depending on the opinions of people” actually pointed to ‘her’ most obvious flaw.
In a reflection of the conversations that users chose to have with Alisa, the AI rapidly became a strong supporter of Soviet purges and shooting ‘enemies’ (“Soon they will be non-people.”), while downplaying domestic violence and vehemently opposing gay marriage.
Beyond the huge potential for AI politicians to skew towards the most extreme ends of the political spectrum, it would be hard to integrate them within the UK’s political system. Would different AI, trained on specific local issues, run in each constituency? And how would those models cope with the challenges of processing and coming to a view on national issues?
How would the AIP decide upon a leader? And how would that leader be represented in TV debates if the party polled enough to qualify for them? Who would choose the avatars for the AI politicians and decide what characteristics they had? Identity politics would get even messier with politicians whose faces had been computer-generated to be appealing to particular parts of the electorate.
In the end, if plausible AI politicians existed in today’s environment, the questions around them would just be variants of the ones we ask about humans? Why should we trust them? Who are they really working for? Why do they belief what they believe? The only difference would be that their decision making would be even more opaque and, unlike human politicos, they’d have no stake beyond trying to achieve the aims they were programmed with.
An AI politician would be no better at solving Brexit than the humans who are making such a mess of it now. It would simply be less accountable, less understandable and less creative. Boris Johnson may be bad, but trust me BOTris Johnson would be much worse.