Robot AI beats world-class curling competitors (2020) | Hacker News Reader

Robot AI beats world-class curling competitors (2020)

scientificamerican.com

96 pointspmontra67 comments

This is nonsense. Sure the robot is better at making consistently the same strength throws. This ignores brushing and making throws that depend on brushing. Watch the men's Canadian Championships this weekend (Brier) or the womens' that finished last weekend (Scotties). Watch them curl a stone 8 feet around others with brushing. This is like saying they can make a robot that hits 100% of fairways in golf. That doesn't make it better at golf than the top players.

OK, this is not perfect/equal. Does it make it "nonsense"? If something is not equal or perfect from the get-go is "nonsense", then I am not sure how we innovate.

Let's come back here in a few years. Robots may brush. We may talk about how innovation iterates, regardless of nay-sayer.

I would agree with GP that having “beats” in the title when omitting that a key part of the game (sweeping) wasn’t included kinda makes it nonsense. It’s still neat though for sure.

Next up “robot beats F1 drivers” (on straightaway).

>It’s still neat though for sure.

Exactly. I've stopped getting riled up about a catchy misleading headline. Life's too short. The interesting part is afer that.

The video seems to show it "beating" the team without either using brooms, but both still playing a game. I.e. they all have stones in play at once and therefore there is theoretically strategy not just accuracy. Am I wrong? I don't know anything about curling. It seems fair to call this "beating" - just not fair to call it curling.

Technically sure, the robot won 3 out of 4 matches against world class curlers at something sort of like, but not really curling.

The implications from the title is that it beats them at curling. You certainly wouldn’t expect the article to be about a robot beating world class curlers at chess.

If sweeping wasn’t such an important part of the game (literally changing shots, which world class curlers would rely on as part of their throw technique) then the article would be fine.

But why is is not curling?

It was played on a curling ring, with a curling stone, using the curling rules against curling players.

The robot just chose to play without a sweep, which I'm sure is still legal.

The humans weren't allowed to sweep either, so it wasn't curling.

I agree that the title is very poor. "Robot AI beats world-class curling competitors at throwing accuracy" would be better.

It's still an impressive feat.

"Humans become better at curling thanks to robotic assistance"

They're not foes. They're tools baselined against human performance.

Up next

Jeopardy bot has faster fingers than human competitors

Following that

Wheel of Fortune bot, who knows the whole dictionary and was trained with the data of all past games, wins a car and waves from the passenger seat.

And spins the wheel to always land on the best option haha

Even saying the "AI" beats the curling competitors at throwing accuracy isn't correct. The AI has access to a highly precise mechanisms for controlling the power it throws with. Would the AI actually be any better than the players if the players had access to a similarly high precision devices for controlling throwing power that the AI is using?

Oh, and I suppose Pitchomat 5000 was just a modified howitzer?

The article states that the robot took on top ranked curling players and won 3 of 4 games.

So it looks like it won without the need for brushing.

The opponents also were not allowed to brush. So it's not curling, it's ice shuffleboard. Which is still impressive, but not curling.

Isn’t brushing mostly just to help correct slightly imperfect throws though? This thing can just put whatever speed/spin on it needs to achieve the same effect. I don’t think that’s great AI though, just very accurate lobbing.

The name of the sport implies that it's more than just correcting imperfect throws.

I think the brushing can curve the stone more than just adding some spin. Also, you can brush so other stones move further after collisions.

However, I don't see why a "perfect" thrower couldn't beat a world class athlete if they're accurate enough.

I'd encourage you to watch the finals of last years Brier. The brushing can do A LOT. Actually, brushing is so impactful that they have had to put limits on the brush composition in recent years because the materials have become so good that it barely mattered how close your shot was. People were moving the stone an incredible distance. Now the throw is much more important and sweeping can help but it can't move a stone 10 feet. [https://www.tsn.ca/broom-wars-a-thing-of-the-past-after-new-...]

Yes. Buried deep in the paper "The matches allow only the thrower and skip player without sweepers". So the video showing sweeping was misinformation.

Also, the "top" teams it was playing against were two women's teams and a handicap team, so they won't have been the actual top - probably just chosen so they can use the term "top-ranked".

If they'd called it a "wheelchair curling robot", it would have been accurate.

A positive point though is that South korea has nearly the top ranked women and wheelchair curling teams in the world. So they didn't just choose some shitty country's national players to call them "world-class".

Similar to the Deepmind boasts in Starcraft II... they beat highly ranked European players but not the top ranked Korean Players, and also they only played specific races (a key component in matchups).

Let me remove all the stuff humans are good at, and then boast how superior AI is to humans.

Deepmind ended up consistently beating everyone including the world champion in SCII with near human view and input limitations. The initial press release was early in the process before letting it play on the SCII ladder. However, it kept getting better.

Good to know. That's awesome they got that far!

This also irritates me.

Humans have built in limitations other than limited intelligence. For example, we have limited ability to remember stuff, our memory is not flawless, we can't perceive time or forces or distances with almost unlimited precision.

Claiming something is more intelligent than humans because it can use something else as leverage to accomplish the task is nonsense.

I am not better at calculating stuff in memory than you just because if I am allowed to use calculator and unlimited memory to beat you.

AI is not more intelligent at playing Starcraft because it can perfectly know what every single unit is doing at any point in time and can project exact resource goals into the future.

Its a neat concept, but this is shuffleboard, not curling. I was expecting a set of robots, a pusher and the ones sweeping. I've done curling before, it's a lot of fun and work. They need a robot that can do more then throw the stone, they need to show it sweeping into victory.

In curling, do the rules state that the team disqualified for not sweeping the path for the stone?

No, you don't have to sweep, but you're allowed to. Playing without sweeping is throwing out half the game, and a big part of what makes it hard.

The players weren’t allowed to sweep.

No but if you don’t let the other team sweep it isn’t quite the same game.

Hypothetically, could such a robot with much greater accuracy win a game without sweeping if its human opponents were allowed to sweep?

Yes! Why should the robot sweep exactly?

Seems easy enough to settle. If the robot can win against human teams that sweep, then we know you don't need to sweep (to beat humans). If the robot only plays against non-sweeping humans though, we don't know.

Also, I assume the purpose of this exercise is not to improve automatic curling, but to challenge an AI/robot with things. If the purpose is the challenge, or the accomplishment, then it seems to defeat the purpose to rule out the challenging part of the competition. This is like how I am a world class basketball player, if you discount scoring, I'm the equivalent of anyone in the NBA.

There are some shots that you just can't make without sweeping... particularly downweight hits where you're trying to exaggerate the curl by sweeping the outside after you squeak by a guard.

I assume the robot creators would have tried it off they thought it had a chance.

The task at hand seems to be a good fit for closed-loop control automation. Much like cruise control - observe output, correct input. Perhaps they don't have the time for enough input/output iterations to dial in the throw? It'd be nice if they mentioned that.

https://en.wikipedia.org/wiki/Automation#Open-loop_and_close...

New paper idea - deep reinforcement learning for constant speed cruise control. Because AI gets you grants.

They did mention it in the paper. They don't have time because every throw it does in the game is part of the game so it can't waste them searching for a good one. It does use previous throws but from other players as well as itself. You can't use other player's throws with a classic closed loop control system.

Curling without brooms isn’t curling. That’s like saying a robot beat Tom Brady at QB, but there were no defenders trying to sack it.

I'm not sure if this really captures what's difficult about interacting with the physical world. This feels like a task for which a machine is obviously better suited than humans, and is being implemented using specialized hardware for the task. Kind of cool, but it doesn't strike me as especially new ground.

All the really hard work here would seem to be in the software and sensor integration. I say that because the hopelessly specialized hardware kind of dulls the enthusiasm for me. I love Mark Rober but I had a similar reaction to his spring-loaded place-kicker machine (wouldn't really call it a robot). The fact that a human could actually even get close to that heavy clunky thing blows my mind.

It would be cool to try this with Atlas from Boston Dynamics. You could at least envision how it can get to and from the rink.

Isn’t this a bit like saying the phalanx system on a ship is better at hitting inbound missiles than a human? Oh wait, it’s AI!!

Yeah I'm with you - its like saying we built a machine that can go faster than usain bolt... a car...

No brooms were used so it wasn't a real curling competition.

"Men with Brooms" == curling (or women) https://www.imdb.com/title/tt0263734/

Is this really an AI victory? I mean, the ice isn’t changing that much.

It’s mostly just an aim-bot

You can position your rocks perfectly, yes. But the opponent team can displace them whenever it's their turn. It's a game of strategy. The only thing that counts is the final arrangement. The robot won over humans 3 times out of 4.

Yeah there’s strategy, but I think it’s a bit of a stretch to compare it to chess. Ultimately you just need your rocks in the middle.

If you’re a perfect shot and go last, you’ll always win.

A mechanical system having better consistency than a human is hardly much of an achievement in this day and age? And you don't need deep reinforcement learning to adapt to changing ice conditions. You could just apply a basic Kentucky windage style control algorithm instead.

Hmm. Maybe the problem really is too complex for that? "the external factors are intrinsically unpredictable and have strong nonstationarity because the friction of the ice essentially changes with every throw.". But it would have been more impressive if they actually compared to such a classical algorithm - made by someone else who was actually trying to be competitive. There don't seem to be any others though.

"Humans become even better at curling thanks to robotic assisted throwing"

Non-adversarial language would go a long way with AI headlines. These are tools, not foes.

The thing right now is that the majority of headlines these days with AI is showcasing how its beating humans at their own games. The media has put AI in an adversarial role on purpose.

I'm not sure what the point is unless used as a warning of how this type of technology could be negatively affecting humans lives in the short run?

"Machines better at exerting precise force on an object than humans using only their bodies"

As if the only purpose to play a game were to win it.

AI researchers use games because they're a tractable problem with clearly quantified metrics of success. It's not to win the game but to advance the field. Sometimes it might be just to publish a paper and seem cool, but every game that falls to AI feels like a step closer to AGI, or at least to better self-driving cars and other simpler kinds of AI.

Although curling is an Olympic event, it was introduced in 1924 and then resumed in 1988. It seems like a skill-oriented (rather than physical) sport. I wonder what it will take to get e-Sports into the Olympics.

"A sport or discipline is included in the Olympic program if the IOC determines it to be widely practiced around the world, that is, the popularity of a given sport or discipline is indicated by the number of countries that compete in it."

>LF2M tank dps olypmic gold medal round

As someone who enjoys (but is absolutely hopeless at) both an olympic sport and an esport, they shouldn't mix

One problem with eSports and the Olympics is that Olympic games tend to be relatively timeless. For example, wrestling or shotput seem like much more natural fits for the Olympics compared to Starcraft 2 or the latest League of Legends. It would be weird if patches, new releases, and private companies fundamentally changed the Olympics.

I also think games don't belong in the Olympics. I would differentiate games from sport by saying that games are purely mental exercises - that is, if I only knew which chess move to make I could play chess perfectly, whereas I may know exactly what to do in a sprint (run as fast as possible) but still couldn't sprint very well. Thus, chess is a game (purely mental/decision making) and sprinting is sport, which requires not only that you know what to do but that you physically can do what you know you need to. eSports are much closer to the "game" end of the spectrum, allowing for things like "Can you physically move your mouse with such precision and react in time".

Well the sport I do (badly) is fencing, and although it looks the same from afar it does change quite a lot

There was an interesting thread about e-sports in the Olympics eight months ago: https://news.ycombinator.com/item?id=23726597

IMO Olympic sports should have rules that everyone is allowed to use, study, modify, and share variants of, in other words they should give players the Free Software movement’s Four Freedoms¹. This way the rules are accessible and adaptable. Good candidates might be Quake or a freed NES Tetris or Super Smash Bros. Melee. The latter two have benefited from being reverse-engineered enough to be modified by their communities for better training and competitive play.

1: https://www.gnu.org/philosophy/free-sw.html

> It seems like a skill-oriented (rather than physical) sport.

Have you tried it? It's not nothing, physically speaking.

Golf and Archery are also considered skill-oriented sports.

Yeah, there's a level of physical exertion. Olympic archers only use 50lb bows however. Its a level of strength pretty much anyone (even non-athletes) can reach. Gone are the days of 100lb or 150lb longbows (weapons of war half-a-millennium ago).

Golf requires you to walk 18-holes while smacking the ball every now and then. I do get tired from swinging the clubs (I'm out of practice, and those weird back / stomach muscles go tired unless you practice). Still, its considered a skill-sport rather than a strength sport.

I suppose that's fair enough. The peak and average exertion for curling is well above either of those, but it's clearly not the physical strength/endurance that determines winners. Perhaps an odd entry point for e-sports though, given examples like you have given.

I was not expecting it when I started, but I recall coming home sore from curling practice.

Olympics added beach volleyball because they needed ratings and advertisers.

Curling was added because it’s ok winter sport that draw enough viewers and advertisers.

Possibly relevant: Chess might seem the least physically oriented of sports, not physical at all. But after about 40 years of age, it's hard to keep concentrating for hours - classical games often last 5 hours or more – and older players fade after the 3rd or 4th hour. One mistake and you're toast, often. See how none of the top players are over 40.[1] Two in the top 20 are over 40: Anand at 51 is world #17 – but in 2008 he was #1, and Topalov at 45 is world #19 – but was #1 in 2006. Experience doesn't seem to matter anything as much as youth! Also, the top players keep very physically fit, because they have to. It's no good winning for 5 hours if you then have a brain fade for 2 seconds. It's needs something like the stamina of a long-distance runner. It seems a top chess player can burn 6,000 calories – and lose 2 pounds – per day, playing in a tournament.[2]

[1] https://2700chess.com/

[2] https://www.cnbc.com/2019/09/20/chess-grandmaster-diet-exerc...

Reply on news.ycombinator.com