AI Correctly Predicts TIME’s Person Of The Year. . . Again
D/SRUPTION talks with Unanimous AI’s CEO Louis Rosenberg about Swarm AI
For the past 90 years, TIME magazine has named a ‘Person of the Year’. The result is based on who has had the most influence on the news, leading to some controversial but entirely justified selections including Adolf Hitler and Josef Stalin. Last year, Donald Trump received the title over resounding favourite Narendra Modi. This year the Crown Prince of Saudi Arabia, Mohammed bin Salman, topped the reader’s poll with a clear majority. Despite this, history repeated itself yet again when the #MeToo Campaign was announced as TIME’s final choice. So, despite various online polls and predictions, it’s notoriously difficult to know who will be picked. But Unanimous AI, an Artificial Intelligence company based in San Francisco, did know. How? Through Swarm AI.
What is Swarm AI?
Swarm AI applies the collective intelligence of a group of randomly selected humans to solve complex problems. Taking inspiration from the combined intelligence of bees, fish, and other species, Unanimous AI provides the interfaces and algorithms to enable humans to converge online in a replica of natural decision making groups. According to Unanimous AI’s founder and CEO Dr Louis Rosenberg, “A Swarm AI system allows you to make better predictions based on the knowledge that you have. In a swarm, a population will negotiate, pushing and pulling and converging on the answer that is the best mix of their knowledge, insight, wisdom, and intuition.”
Unanimous AI’s approach used Swarm AI to ask 60 members of the public who they thought was least likely to win in each round. The same methodology was used to predict the result of the Kentucky Derby Superfecta, and to advise Jeff Bezos on how to donate his fortune. The system eventually whittled down TIME’s 33 candidates to five most likely choices – the #MeToo campaign, Vladimir Putin, Donald Trump, Kim Jong-un and Robert Mueller. On Wednesday, Unanimous AI’s final prediction was shown to be absolutely right. The reader’s poll, on the other hand, taken from tens of thousands of participants, was not. The question is, why did the TIME reader’s poll return results that were so different to those given by Swarm AI?
“It’s something we see a lot,” says Louis. “Polls tend to drive people to answers that are driven by personal biases, whereas swarms drive people to converge on the most likely answers. What we like to say is that polls, particularly political polls, are polarising – meaning that they amplify people’s biases because they give a single choice and are not encouraged to be flexible in finding an answer that the group can best agree upon.”
Perhaps Swarm AI could provide a better alternative to polls, encouraging flexibility. According to Louis, the technology also improves over time.
“Our predictions demonstrate that we do a much better job at amplifying the intelligence of a population than a poll,” he says. “We’ve done long term studies with a researcher in Oxford that predicted 50 English premier league football games over a period of five weeks. We compared individuals’ predictions to the swarm. The individuals were 55 per cent accurate, but when we had those same individuals predict together as a swarm, they jumped up to 72 per cent accuracy.”
Predictions, not prophecies
Even though Unanimous AI were confident in their prediction – confident enough to bet on it, in fact – it’s not taken for granted that the result will be correct.
“We always expect that some percent of predictions will be incorrect – if the TIME prediction had been incorrect, we would go back, look at the data, and try to understand why. Did our system miss something, or did something happen that was just not predictable? On our blog we predict sporting events and every week we are normally very accurate, but if we miss a couple of games we go back and we look at what could have been done better.”
In many respects, it was hardly surprising to see Putin and Trump at the top of the list. The #MeToo campaign, however, was something of a wild card.
“The facts on the ground can change so fast now,” says Louis. “Two weeks ago we ran the prediction and we came up with a top five, and then on Friday we reran the prediction because a lot of things had changed, especially in the US, with high profile news anchors losing their jobs over the #MeToo campaign.”
Various factors could have changed these odds, including the people behind the decision and the actions of potential winners. Last month, for instance, TIME was sold to the Meredith Corporation for $2.8bn. TIME’s buyers are backed by the controversial, conservative Koch brothers, leading to debate over the future direction of the publication. Trump also made public comments about the process, stating that TIME told him he would ‘probably’ win if he agreed to an interview and photo shoot. TIME dismissed the claims.
What if future gazing becomes future telling?
It’s in our nature to want to know what the future will bring. But if we can predict the results of things, what happens to matters of chance? Are we moving towards a world where anything can be predicted to a high level of certainty? There is also the recurrent issue of trust to consider. It’s possible to trust AI to be factually correct, but can we trust it to be morally right? Vladimir Putin and Donald Trump were arguably not the most moral choices for swarm AI to make, but the system transcended emotional considerations. From a business perspective, using collective wisdom to tell what will happen within a market or industry could be the difference between success or failure. When Jeff Bezos used Swarm AI to help him decide which charitable cause to donate to, he wasn’t necessarily thinking about the future of Amazon – but no doubt certain strategic decisions could be guided by the system. Take, for example, the eternal questions of where a market will go, what products will endure, and how consumer behaviour will change. Through Swarm AI, businesses could potentially find out what consumers really want. Nonetheless, they should not fall into the trap of taking these predictions as gospel. Louis is clear that the system is not completely accurate.
“We predict sporting events all the time, and while we can do very well across a period of 50 games, for any single game there will be randomness. Swarm AI is about getting the most out of the knowledge in people’s heads, their instincts and their intuition, as well as their gut feel. We can combine all of that in an optimal way, and make a good prediction, but it will never be perfect.”
To an extent, some things will always be left to chance. There are so many variable factors in any given situation that it would be foolish to claim that technology can successfully predict the future. What Unanimous AI has done is prove that by combining human decision making, it’s possible to come up with likely results. But as technology improves, we will have to ask whether we really can know what will happen in the future. As it so often does, the advancement of innovative tech will lead to philosophical arguments – but Swarm AI is about making the most of insights and intuition, and doesn’t claim to be infallible.
“What we find is that if you take a group of people and you combine their knowledge as a swarm, they will make a far more accurate prediction about the future. But it’s never going to be perfect, because there is always randomness,” says Louis. Despite this randomness, Unanimous AI has now correctly predicted the winners of one of the most controversial and changeable awards for two years running. Next year, can they make it a hat trick?
Dr Louis Rosenberg spoke at our Disruption Summit Europe event earlier this year – take a look here:
Does Swarm AI represent the future of human intelligence and decision making? Will trust and moral considerations affect the adoption of Swarm AI? Are there certain things that should not be subject to AI predictions? Share your thoughts and opinions.