A.I. wins the Superfecta at the Kentucky Derby, turns $20 bet into $11k
unu.aiWhen 'artificial intelligence' is the same thing as 'crowdsourcing and averaging'...
https://www.youtube.com/watch?v=JKNVwXU2rrI
This is misleading. It's just Wisdom of the Crowd stuff. (https://en.wikipedia.org/wiki/Wisdom_of_the_crowd)
Traditional "Wisdom of crowds" are polls that take averages, which end up being fairly impressive. Swarms are real-time dynamics systems that converge on local maxima. It takes a little thinking to appreciate the difference, but it's profound. Swarms blow away traditional methods because they foster a real-time emergent intelligence, while polls just take a snap shot of group opinion. They're very different.
It's intelligence.
This was a derby where the top horses finished in the exact order they were favored. Nyquist went off at 2-1, which by Derby standards is a huge favorite. Exaggerator finished second, and was 5-1. Gun Runner finished third at 10-1. Mohaymen finished 4th at 12-1.
This almost NEVER happens in the Kentucky Derby, which is known for being a wildly unpredictable race. They simply bet "the chalk" (gambling slang for favorites), which by definition is the crowdsourced results, since odds are determined by the relative bets placed by the public. They could have looked at the top 5 horses (there was one other horse at 12-1), bet them in order of odds for $2, and come up with the same result. The only reason the winning ticket was worth so much is because of the insane amount of money bet on this race.
Prediction: Next year's 'crowdsourced' prediction will closely match how the betting shakes out, and it will fail, because that's how horse racing works. Now, if they can come up with results that are profitable over time for a large amount of races, then I'll start to be impressed.
Well, keep in mind they ran the predictions twice, once two weeks before the race, and then again after the lineup was determined in the middle of the week, switching one horse. I don't know what the odds were at that time.
And, if anyone else has a winning Superfecta ticket, I haven't seen it. So, it's not like this was some easy pick.
They re-ran their analysis after post positions came out, but they didn't really have a big impact this year, since none of the top contenders had particularly unfavorable gates.
If it was so easy to win this bet, why did BING (with billions behind them) blow it. They got one horse right out of four, and they did that by processing all the available information. It's easy to say a pick was easy after the fact.
This is a completely irrelevant achievement. Nobody can reasonably expect this to be repeatable, and if enough people make predictions somebody will get it right.
It's so blatantly irrelevant, I assume this company only cares about selling and not at all about their technology being able to do anything better than chance.
It's a very easy claim to test though - just wait until next year's Derby and see if they can do it again.
It's also worth pointing out that they were challenged to make this pic by a reporter in advance of the Derby. They weren't just throwing out a random guess, they met the gauntlet that had been thrown down once already. This is a remarkable achievement in the face of skepticism.
If you read the CEO's statement in the original article, he wasn't claiming omniscience, but that horse-racing was certainly difficult to predict.
The better question is not, "will they hit the Preakness Superfecta?" but, how do UNU's picks do against the average guy at the track?
I'd bet they're pretty dang good.
See my above comment. Their picks are just a reflection of the odds set by the betting public. They got extremely lucky, since the Kentucky Derby almost never shakes out with the top 4 favorites finishing in order.
UNU has been making MANY impressive picks, and they do it in response to challenges from reporters, so there's proof.
NEWSWEEK challenged them to predict the Oscars and they outperformed over 90% of the experts, including 538.
Here is the NEWSWEEK article: http://www.newsweek.com/oscar-predictions-artificial-intelli...
Not strictly repeatable, but easily testable if they got lucky or not. There are thousands of horse races a year. Heck there are 2 more races left in the triple crown this year. It would be interesting to see if they could do the same for Preakness and Belmont. At some point they will either revert to the mean or show they have some serious predictive power in their methods.
The derby has 20 horses in the competition. Winning the superfecta requires picking the 1st, 2nd, 3rd and 4th places in correct order.
The possible ways to choose that ordering are 20 * 19 * 18 * 17 or 116,280 combinations. So, a 1 in 116,280 chance of winning (which is why it pays about 11,000x -- bookies get to keep the rest!). If this crowd-sourced horse pickin' works twice that would be about 1 in 13.521 billion odds. So, if it can do it twice there might be something to it. Doing it once isn't really that big of odds. If the favorite gets upset (Nyquist this time) then the crowdsource probably loses. Not sure how often that happens, but it probably isn't particularly rare.
Ah but as linkbait for a SaaS company? Priceless :-) As I recall there have been other efforts at this, generally using professional handicappers, and while crowds can silence otherwise compelling outliers they are certainly not infallible. I was reading where someone did something similar to the NCAA bracket, taking a source of many thousand user supplied brackets and trying to normalize them to a "golden" bracket, it was wrong in a couple of major ways.
Well, your math points out that doing it once is pretty big odds. Obviously not as big as doing it twice, but if it was easy them lots of people would have won the Superfecta. Only people I've heard of winning are UNU and those people who bet UNU's picks. As someone else pointed out, the swarm intelligence approach doesn't have to hit every time to be impressive. Just better than an average bettor. Look at their site. They've been successful with the Oscars, Super Bowl and College football.
Psychologically speaking no one gives a crap whether this is "really AI" or not. There are a lot of people who will hear about this and get very excited about learning computers. This is a big deal (regardless of what's actually in the box.)
No, this is not repeatable to nail the Superfecta of the Kentucky Derby (or pick a perfect March Madness bracket for that matter). The point is that this platform seems to outperform the experts. And every once in a while, they might nail it perfectly.
This isn't an AI. UNU.ai looks like a discussion platform or wisdom-of-the-crowds thing.
Is this really an AI? Or is it just harnessing the wisdom of the crowd?