A.I. That Predicted Kentucky Derby Superfecta Reveals 2017 Bets

The betting world has spent the past several weeks obsessing over its predictions for this Saturday’s Kentucky Derby, but those who pay attention to artificial intelligence held back their money until Thursday, when a San Francisco A.I. firm released its predictions.

That’s because, in 2016, Unanimous A.I. used a technology called “swarm intelligence” to coordinate a group of racing fans to correctly predict the Kentucky Derby superfecta (the first four places, in order). The swarm beat 540-to-1 odds, along with the most-trusted handicappers in the world.

Now, the “swarming” A.I. technology that takes its inspiration from bees has revealed its prediction for the 2017 race. The predicted superfecta for the 2017 Kentucky Derby, aka The Most Exciting Two Minutes in Sports, aka The Run for the Roses, a race that includes twenty beautiful ponies with names like Irish War Cry, Always Dreaming, Classic Empire, and McCracken, is finally out.

Since correctly predicting last year’s race results, Unanimous A.I. has nailed the outcomes of the Super Bowl and the Oscars, foreseen Donald Trump’s unexpectedly low approval rating, and even predicted this year’s historic World Series win by the Chicago Cubs.

Now, the company that turned a $20 bet into more than $10,800 at the 2016 Kentucky Derby has weighed in with this year’s picks. History says that anyone betting on the race should at least take a look, but how does a tiny machine learning startup make these incredible picks? Founder Dr. Louis Rosenberg tells Inverse it begins with swarms of humble, homeless bees.

“A swarm of bees is an emergent intelligence that’s faced with a really challenging problem,” Rosenberg says. “Every spring, a full colony of bees splits in two. A group of about 10,000 bees now has to find a new home.”

These homeless swarms will send hundreds of bees to scout potential sites over about 30 square miles, and each assesses its candidate site along a range of competing criteria including size, ventilation, and proximity to sources of water and pollen.

Biologists were faced with a vexing question: how do bees pick the best of hundreds of competing, multi-variable options? More to the point, how is it that these little insects do actually seem to end up at the best sites?

Scientists eventually discovered how the bees were doing this, and it turns out their swarming approach to problem-solving looks a lot like the function of neurons in the brain. The scout bees gather together and perform a “waggle dance,” in which their movements express information about the site they’ve visited — what qualities make it a good candidate and, importantly, how good it is along each metric. The bees send out dance-signals that excite and suppress different aspects of the bee-dancing going on all around them. It creates an enormous, multi-variable tug-of-war that slowly eliminates minority waggles and converges toward fewer and fewer dominant options.

“This is very similar to how activation signals form in brains,” Rosenberg explains. “So, the bees are really forming a brain of brains.”

Unanimous A.I. is Rosenberg’s attempt to organize human beings into such a brain of brains. After all, he says, bees only have a few million neurons, compared to a hundred billion or so in humans, and yet they can achieve quite a bit when properly networked. Surely, he thought, the human intellect could achieve even more if it received the same swarm treatment.

The resulting platform is the Unu (pronounced: “Ooh, new!”). An Unu is a decision-making user interface, girded by A.I. programs in the background. Members of the swarm, be they professional sports analysts or friends trying to pick a restaurant, see their options arranged around the outside of a space, with a circular “puck” in the center. Each can apply a certain amount and direction of force to the puck to pull it toward their favored option. When the swarm successfully pulls the puck all the way to one option, the choice is made.

In the very first instant, with users pointing straight toward their favorite option, the Unu is not so different from a simple vote. Then that instant passes, and the puck begins to move.

The beauty of swarm intelligence is that the decision-making process is fluid over time, allowing participants to begin making complex strategic decisions in response to the ongoing movement of the puck. If at first a Kentucky Derby handicapper is pulling for one horse, but then sees the group slowly sliding toward a truly terrible option, that person might switch their vote to the second most popular option, which they can accept in spite of it not being their favorite. Since the angle of the applied force is independent of the location of the options, people can attenuate the force they apply toward more than one.

Of course, all this puts quite a bit of power in the hands of the person or algorithm that places the options. No two horses are any more coherently opposites of one another than any other two horses, and yet placing any pair on opposite sides of the Unu would result in support for one working as de-facto opposition to the other. This introduces a confounding bias to the seemingly simple interface, requiring the introduction of algorithms that explain why Unanimous A.I. is called Unanimous A.I.

A.I. algorithms monitor the progress of swarm decision-making, and modify the physics of the simple two-dimensional world to offset any bias introduced by the physical placement of options. According to Rosenberg, “If two popular options are near each other, thus biasing that side of the board, the physics of the environment adapt in real time, neutralizing that bias. It’s as if the geometry warps… to ensure parity between options.”

A wrinkle is that experts may not always be the best participants in a swarm, even if they are the most reliable analysts. With the insight of the swarm, ego-free enthusiasts might just beat stubborn professionals, since they will be more willing to embrace the fluid nature of its decision-making process. The best swarm participants are those who have accurate self-appraisals — it’s actually fine to have low accuracy, just so long as you don’t have any problem admitting it.

Swarm intelligence can’t produce insight that doesn’t already exist in the group of individuals who comprise the swarm. In the case of 2016’s Derby picks, the analysts knew what the superfecta results were going to be, they just didn’t know that they knew it. The idea is that throughout the tug-of-war of the Unu, aggregated across all those participants, all preferences will be heard and taken into account to the appropriate extent.

Swarm intelligence won’t work for every problem. No mass of bees or humans could collaborate to decide on the correct solution to a complex routing problem because none of the individuals making the decision has any insight into the answer. The designer of the Unu also couldn’t know that the right answer was among the options they were providing to the swarm. On topics where people can actually have some level of informed opinion, with a finite number of options available, the swarm seems to have power to pull the best aggregate answer out of the confusion and gridlock of pure democracy.

That’s why it’s so incredible that a swarm was able to pick the correct answers in the 2016 Kentucky Derby superfecta through all the uncertainty, doing something that no one of the swarm’s members could have, alone. Rosenberg noted that the same group that made the swarm predictions was also asked to make their picks through a simple poll, but taking their most common answers only predicted the correct placement of one horse out of four.

If swarm intelligence can repeat its superfecta win at 2017’s Derby, it will have proven that there is reliable depth to the insight that this A.I.-mediated digital Ouija board can pull out of human opinion. In so doing, it will provide evidence that there is a real advantage to making decisions as swarms.

A.I. often overshadows or even belittles human ability, but swarm intelligence is one of a growing number of examples of technology augmenting that ability, or perhaps maximizing it. Rather than dominating human decision making, A.I. may one-day help mankind by letting it make decisions the way bees pick their homes.

The A.I.’s 2017 Kentucky Derby superfect picks are:

1. Classic Empire

2. McCracken

3. Irish War Cry

4. Always Dreaming

Good luck, but you might not need it.