1. Hamilton, Canada legalizes street hockey once again.
2. When will Amazon ship to the moon?
3. New Zealand offers free holiday to people who agree to do job interviews there.
4. $200 AI device to judge whether your tennis balls are in or out.
5. My podcast with The Art of Manliness. And my interview with Harvard Business Review.
6. Spanish Uber for priests. And John Cochrane on what economics really should be for.
4 will be interesting as a measure of McEnroe’s bullying. He seems to have done it deliberately to intimidate the umpires. But bullying an AI would be harder.
Ahh, for the days when people called their opponent’s shots.
Perhaps this AI could be shipped to Canada to arbitrate street hockey? That could be interesting. Especially as Farid Zakaria and CNN are arguing that the American dream lies in Canada these days. Does that mean he is considering a move? What are the plagiarism laws like there? Street hockey certainly sounds attractive but perhaps not to the hipsters planning to flee there.
Catholics are often accused of shopping around for a good confessor. Will Uber-for-priests allow comments of the “I slept with my sister-in-law and I got away with three Hail Marys!” sort? I can’t think of any other use for it.
4. – The headline is wrong – this is not an AI device, as can be noted from the article itself. ‘On close calls, we rushed over to watch a video replay on the In/Out screen. At hour’s end, Gentil whipped out a tablet and connected to the In/Out app, which showed where all our shots had landed and provided some other stats.’ And the device’s margin of error is 2 cm – that is essentially 4/5 of an inch. Which just might explain the need for non-AI intelligence to review the playback.
The article very clearly describes an AI system (which also allows video review in addition to that).
Could you give me the quote? Not this one, though – ‘AI also helps the device track the ball’s flight, pace, and spin. “This would not have been possible five years ago,” Gentil says.’ because that is pretty much absurd. Particularly the ‘this wouldn’t have been possible five years ago’ aspect, though I’m more than willing to grant not at such a cheap price point. All sorts of equipment have such broad capabilities these days (think targetting or navigation), without their manufacturers generally claiming to use AI (however it might be defined apart from the term being used for marketing). A broad overview can be found here – https://en.wikipedia.org/wiki/Fire-control_system#Modern_fire_control_systems – and unsurprisingly, such systems incorporate a lot more information to arrive at a valid solution concerning a hit or a miss.
To be fair, this hinges on what ‘AI’ means, apart from marketing. Basically, a tennis court is already a defined boundary, and to say that an ‘AI’ is used to determine those lines each time the cameras are attached is either less than accurate, or just an extremely broad definition of AI. The ball tracking itself does not require an AI either, since it again is tracking a fairly distinctive object, much like a fire control system.
More concretely, Mercedes cars in the last several years have sensors that use the standardized shapes of street signs, and the standardized way information is presented on those street signs, to actually operate the car (think speed limits, for example, or a warning sign concerning an upcoming sharp curve to change vehicle speed without driver input). One can debate how much such a system adds to the cost a Mercedes, but I have never seen any marketing (German language marketing, admittedly, and in conversations with Germans that work for Mercedes) claiming that a Mercedes car uses an AI to determine whether it is within the lines marking lanes, or how it is returned automatically to its proper lane (in all fairness, Germans broadly seem a lot less enthused by ‘AI’ as a term than Americans). One can quibble about the five years part – the time span between testing, initial introduction in extremely high end models followed by eventual mass adoption in all Mercedes vehicles is significant in this example – but truly, the considerably more complex real time capabilities of a Mercedes car in determining whether it is in or out of bounds does not reach the definition of ‘AI’ apparently.
I think most people would consider both the tennis device and Mercedes technology to be examples of AI. Both involve decision making based on processing fuzzy and complex input, to perform tasks which not so long ago required a human.
I’m not sure you really grasp what is required to check whether a ball went out based on raw video data. This is not a trivial problem by any means. Though the technology to perform such tasks is now getting quite mature.
P.s. the fact that Mercedes chose not to advertise their technology as AI is hardly decisive. That’s a marketing decision.
‘Both involve decision making based on processing fuzzy and complex input, to perform tasks which not so long ago required a human.’
And I would disagree – which is fine, because in part there is no definitive answer. Nonetheless, that part about ball spin, for example, sounds like marketing intended to convince a tennis player to pay for the system, with its AI advantage.
‘I’m not sure you really grasp what is required to check whether a ball went out based on raw video data.’
Actually, it should be fairly simple, as long as we discount the immense advances in terms of processing power now available to bring the cost way down. The lines are marked on the court (thus creating a boundary, and even after smudging, the line as a boundary exists in memory), it is fairly trivial to compare motion of a specific defined object frame by frame against a background (again, discounting how cheap and fast processing has become, not to mention the cost/quality of the camera), and the fact that the system is inaccurate to almost an inch suggests the limits of this approach. Though only in a cynical sense, I wonder if the system merely attempts to match the line as a defined point with the ball, as a defined point, then deciding based on such simplistic criteria whether it is in or out – and that is really quite simple with enough image resolution and processing speed/power. I will also admit to not being so familiar with the rules of tennis that there might be other definitions of out which this system would help with, as compared to where the ball hits the ground (I do not mean that the ball is clearly out of bounds when it lands, but possibly more is considered than the line when calling out in such situations).
And to combine two into one – ‘the fact that Mercedes chose not to advertise their technology as AI is hardly decisive. That’s a marketing decision.’
Of course it is – Mercedes has a general reputation to protect for not overstating what its products offer. At least in Germany, that is, where Mercedes is considered a staid company which one can rely on to deliver what it promises. And where the industrial robots used to make a Mercedes probably come closer to something approaching AI than either the car or this tennis system. AI is a legally undefined term, which means those with something to sell will happily use it to make a sale.
“‘I’m not sure you really grasp what is required to check whether a ball went out based on raw video data.’”
Dan gets it. This is an amazingly hard problem. Only due to openCV is this even possible for a one man shop to attempt, and even then there is so much code to determine the tracklets).
Prior shows that he is completely out of his depth. As always.
‘Dan gets it. This is an amazingly hard problem.’
No, it isn’t, since the parameters are fairly easily defined – the boundary of the court is presumably marked with an identifiable line and remains fixed, as does the camera(s), and the ball has a fairly well defined shape. Of course, there may be aspects different than compare frame to frame, track motion of defined size object against background in each frame, compare to non moving boundary – but to a major extent, that is what it comes down to. Or not, as the case may be – I certainly do not want to say that the man does not have a much better way of doing it than using current processing technology to brute force an answer, even if I personally doubt he does.
But do notice in the following text, which does demonstrate an impressive range of features, not a single word concerning AI – ‘The library has more than 2500 optimized algorithms, which includes a comprehensive set of both classic and state-of-the-art computer vision and machine learning algorithms. These algorithms can be used to detect and recognize faces, identify objects, classify human actions in videos, track camera movements, track moving objects, extract 3D models of objects, produce 3D point clouds from stereo cameras, stitch images together to produce a high resolution image of an entire scene, find similar images from an image database, remove red eyes from images taken using flash, follow eye movements, recognize scenery and establish markers to overlay it with augmented reality, etc. OpenCV has more than 47 thousand people of user community and estimated number of downloads exceeding 7 million. The library is used extensively in companies, research groups and by governmental bodies.’ http://opencv.org/about.html
Color me skeptical about calling that collection of algorithms AI, particularly considering how the actual boundary lines do not shift, and the ball is not an irregular object which could be difficult to keep track of. (But as noted below, machine vision/pattern recognition is a blurry subject in AI terms.)
‘Only due to openCV is this even possible for a one man shop to attempt’
Reusing libraries is the point of something like the LGPL – or BSD, though of course he is now free to make his code his own, without sharing if he wishes.
Strangely, this web site, at least in the eyes of a disloyal reader, has so studiously avoided any discussion of the free software movement that it could almost make one wonder if there might be some reason for not exploring it, particularly in light of this web site’s attention to so many other, far more trivial, software themes.
Look, idiot. There will be no ‘picture’ with the ball right next to the line.
I even said the word tracklet. You latched onto the word openCV, now try tracklet. This is an amazingly hard problem. Luckily for this guy he is standing on the shoulders of giants, but still he either wrote or contracted out a lot of code.
‘There will be no ‘picture’ with the ball right next to the line.’
Well, apart from the video replay – you did read the article, right? See, the AI is not good enough in many cases, so the players get to look at a recording. One can safely assume that the video is the source of the raw data the system is evaluating, at least when talking about a $200 price point.
‘You latched onto the word openCV’
No, I latched on to the openCV software project, which makes no claims to being an AI project – you may have noted that another commenter, equally dismissive of this system representing AI and just below, has a link to http://simplecv.org/ Strangely, though, it seems his opinion is not retarded, though it is the same as mine in dismissing the claim that this is an AI system. But then who knows, maybe using his link, even someone with a day job might find a way to simplify their use of openCV enough to realize their software dream.
‘This is an amazingly hard problem.’
It is not a hard problem – but then, I guessing you aren’t all that familiar with the world of fire control systems. They have amazingly hard problems to solve, by the way – I’m fairly certain that air temperature, ball temperature, ball pressure, racket temperature, racket tension, air pressure, wind speed, humidity, and camera motion are not part of what the system measures and compensates for, at least not when ii is only accurate to 4/5 of an inch.
In this case, the camera is fixed, the lines are fixed, and the ball has an easily identified shape that can be easily marked. Which is not necessarily required, as the only thing the system truly needs to measure is where the ball hits the ground – essentially, the system only is required to compare when an existing defined shape (being essentially a sphere is really helpful in this regard) enters the area already defined by the fixed lines. Other measurements can certainly be added, but to the extent that the ball is a defined element, this is really not all that hard, one would assume. Regardless of whatever buzz words one likes to throw around, good programmers actually try to simplify, not complicate, what they are trying to achieve.
Interestingly, you do not discuss some of the ways to reduce the amount of processing power required – for example, the system could be only handling a small amount of the visual field, using predefined parameters concerning the motion of the ball. No need to consider it moving at 300mph, for example. Such a predictive function would at least come closer to the older idea of AI – but then, such a fairly banal predictive function would also have essentially nothing to do with computer vision, either.
You are amazing. You know nothing and yet you spew out so much. You are everything that is wrong with modern liberals.
Look, dork. There is no picture showing the ball is in or out. This isn’t a camera operating at thousands of fps. The problem is to:
1. Construct from the image the 3D coordinates of the playing field.
2. From the video feed determine the 3D position of the ball within each one (this will not be fully determined)
3. Construct tracklets from the the data from (2)
4. Determine which tracklets are consistent, and from this determine a 3D model of the flight path of the ball.
5. Determine where this path would intersect the playing field and the determine is the ball is in/out.
All of these stages are hard. All are beyond you.
I agree with prior_test2 on this. It seems that any software whatsoever is called AI for marketing purpose nowadays.
Computer vision is a field in AI. Full stop.
Given this do you want to retract your nonsense statement?
‘Computer vision is a field in AI.’
Strangely, the openCV people do not make that claim when discussing themselves – perhaps they are retarded? Or maybe they realize that their tools will be part of something that is recognizable in the future as part of an AI system, and not merely a collection of algorithms?
The definition of AI has shifted a lot in the past 20 years – and mainly in a direction showing just how debased the term has become, such as people claiming that nudity detection in color photographs as a sign of progress in AI.
AI researchers did foundational work on machine vision. They wrote the standard libraries. You would use those to build a “ball and line recognizer.”
http://simplecv.org
That said, “in or out” is pretty far from general intelligence. I would not call it AI. Real AI would be coaching you on your serve.
Yes, I left out the whole pattern matching/machine vision aspect of AI, because that becomes rapidly blurred in such conversations.
LOL prior, you’re retarded.
Anyway, this is a great device. It will sell, however it has a very short lifespan as there will be a phone app that does this. And since the phone has so much more compute it will be able to do so much more.
(I’ve been wanting to do this app, since the code that powers it is so cool, but it is hard to give up the lucrative day job).
Conundrum re The Complacent Class: if government plays a bigger role in the management of risk in everyday life, shouldn’t we see more geographic mobility in the US population?
Before the rise of government as risk manager, presumably people had stronger relationships in the community. These relationships required investment, and provided a return in the form of risk management. Breaking a relationship like this by moving to another town would have been costly. You’ve invested time, effort and money into your local relationships, and cannot easily get that support in a new town without substantial new investment. Gaining risk management services is less costly now with a large government playing that role. And yet we now see less geographic mobility, not more. Despite the government helping to emancipate us, we are staying put more and more. Perhaps our ‘complacency’ is worse than we thought.
There are so many moving gears. More business travel. Internet slowly destroys place. Hard to believe migration should be uniform in face of technology.
6. And if it wishes, that Spanish service can incorporate Greyball like features, to ensure only valid confessions, in keeping with the Catholic Church’s TOS. ‘Uber used a tool, code-named Greyball, in 2014 to identify Portland, Ore., officials who posed as regular customers to request rides in order to gather evidence that the company was operating illegally in the city, according to the Times report. But rather than procuring a driver for the “customer,” the service showed officials fake versions of the Uber app, complete with fake drivers. Any real ones who did respond to the requests for rides would quickly cancel, sometimes after direct intervention from Uber officials to drivers, allowing the service to avoid detection in a city where it was banned.
“This program denies ride requests to fraudulent users who are violating our terms of service, whether that’s people aiming to physically harm drivers, competitors looking to disrupt our operations, or opponents who collude with officials on secret ‘stings’ meant to entrap drivers,” Uber said Friday in a statement.’ https://www.washingtonpost.com/news/the-switch/wp/2017/03/03/the-latest-bad-news-for-uber-shows-just-how-far-its-willing-to-go-to-get-its-way/
#2 “It’s said that the new administration may be changing course by pushing for a faster and cheaper return to the Moon before the end of Trump’s term.”
We choose to go to the Moon in this term and do the other things, not because they are easy, but because they are hard, because that goal will serve to organize and measure the best of our energies and skills, because that challenge is one that we are willing to accept, one we are unwilling to postpone, and one which we intend to win, and the others, too.
Ahhh, but New Zealand will only let in 100 people on this generous offer, and only tech workers! I guess life’s good as a programmer.
#2 – The real question is whether Amazon can ship a Prime purchase there for free given two days’ time.
5b re: The Complacent Class
The book seemed to me to be just another “thinker” predicting the future by extrapolating from the present and cherry picking the past.
“It may sound like those people are not complacent, but if you compare them to earlier times in American history — the 1930s, or the Civil War period, or the 1960s — people’s willingness to just put up with things, to improve the quality of their leisure and then get on with life and not really agitate for very urgent change, that’s higher than before.”
“Higher than before” based on what measure?
I would have preferred that an economist would support his thesis with more economic principles.
@Mike W: Tyler cites many figures (“measures”) in the book on interstate migration–one measure of relative complacency. One is that 30% of African-Americans moved northward from 1920s through the 1960s. It is much lower now. I think if you look at the different generations in your own extended family you will see the same thing. My parents–now in their late 80s and early 90s–moved across country to California in 1952, as did many of their peers. I moved across country in the opposite direction 35 years ago, but my brothers have stayed put, as have their children. Interstate migration is way down, and that is a type of complacency.
“Interstate migration is way down, and that is a type of complacency.”
This is a problem with what Tyler says in interviews and I assume writes in the book: “That is a type of complacency” when some of what he argues is complacency really isn’t. For example, in this recent podcast he says that putting more people in prison is a type of complacency. How? The same for his criticizing wanting to make cities nicer as “complaceny”. Having the Cuyahoga River “catch fire” in 1969 sounds more like complacency to me.
Tyler great conversation at hbr.
A few choice quotes :
“We’re seeing a response in terms of protests and social movements where I think we get a bit of a redo of a lot of angles of the 1960s, but with social media to speed it all up. And that will be a very interesting, but also a somewhat scary, experiment.”
Agreed, the combination of 60’s progressivism and social media is terrifying. We could easily end up with the French Revolution on steroids.
“We know they can work, but in terms of the legal and regulatory framework, it will be decades, I think, before we pass enough laws that they can revolutionize our lives.”
Got to disagree. You are taking the line of many pundits, however you are an economist. You should crunch the numbers as to the benefit to society of driverless cars. There have been few innovations with this level of payout. While, of course, regulation can stop anything, the public backlash will be so great (again amplified by social media) that I’m guessing that regulators will back down. This is all driven by the immense social benefit.
Ps: note I use simple lane following in my car every day and it has decreased my stress level in traffic immensely, and that is just a taste of what is to come.
“If you look at the federal budget, we’ve moved from a situation where about 20% of it was decided in advance, say, in the early 1960s, to where about 80% of the budget is locked into entitlements, essentially”
This is a great point, and needs to be repeated often. Many who read hbr already knew this, but it still needs to be repeated.
Don’t worry about progressives and social media when they are demonstrably not the ones going nuts.
http://www.cjr.org/analysis/breitbart-media-trump-harvard-study.php
And re simple lane following
https://twitter.com/Jalopnik/status/837343883046551552
There is such think as premature adoption of technology. (And re “AI” discussion above, not all “AI” is AI.)
More of the french revolution within the US:
http://www.slate.com/blogs/the_slatest/2017/03/04/middlebury_professor_injured_as_protest_against_racist_author_charles_murray.html
And that is SLATE *SLATE* reporting on the event. Not like they condemn the actions, of course, but still they reported it.
AI cars are not being held back by regulation. They are being held back by not being as good as humans.
Wow. If the following proves true, it was a very bad day to worry about progressivism and social media.
https://www.bloomberg.com/politics/articles/2017-03-04/trump-calls-obama-sick-claims-trump-tower-was-wiretapped
It is called “litigation strategy” : it is typically not as simply passionate and adversarial as it seems. All litigation – without exception – diminishes to unread history after a fixed period: perhaps the period has been a little elongated in recent years, but probably not by much. Even Cicero is mostly unread these days. Don’t get overexcited.
Neither of the responses so far were really answers to the problem. They were redirection or dilution.
And so if there is a real problem, they don’t really serve.
If we have a corrupted media system, and that now extends to Whitehouse decision-making, we are ill prepared for actual crises.
What I said was not redirection, it was not dilution. Try again. The out-party said to the in-party, we will litigate against you. The in-party said to the out-party, for what? for things that were done by yourselves even more than we did? Are you being paid to not understand?
That being said, Rags, welcome to my world. I have worried about one crisis after another – and worried if I should tell a friend of a friend who had a friend who might be able to do something about it – since the year of our Lord !974. Well, one crisis will follow another. God amuses himself by helping us: but the word amusement is perhaps too subtle, He LOVES us as well: amor omnia vincit. 3 sunt magna: caritas super omnia. (a reference to Saint Paul: there are three: hope, faith, and charity: charity is the greatest of them all. The algorithm of algorithms, as Shakespeare might have said, if it would have fit into iambic pentameter. ) Welcome to my world, my friend.
Or if it makes you happy: redirection, dilution. If it makes you happy. I don’t at all care in the way you think I might. “””Ou; “reveillez-vous”. You have friends who you did not know about, and some of those you thought were friends would like to have been friends, but in their weaker moments decided to delude everyone they knew to the best of their ability. (skip the following French quote if you want:):: Dieu, qui reveles aux coeurs mieux qu’a l’intelligence … Ressere autouur de nous, faits de joie et de pleurs, Ces groupes retrecis ou de ta providence Dans la chaleur du sang nous sentons les chaleurs …O soirs, o douces veilees, dont les images mouillees Flootent dans l’eau de nos yeux,,, Amor amnia vincit.”””
In the podcast Cowen said that the U.S. has _barely_ had 1 percent prodctivity growth in 3 of the past 4 decades. That isn’t correct and an exaggeration. (1947-1956) 2.8% (1957-1966) 3.3%, (1967-1976) 2.3%, (1977-1986) 1.4%, (1987-1996) 1.5%, (1997-2006) 2.8%, (2007-2016) 1.2%.
Looking at the most recent four decades, only one out of the for “barely managed beyond much past 1 percent” and that is the one that includes mostly the years following the worst recession since the 1930s.Great Depression. Notice how the higher than average 2.8% productivity growth of 1997-2006 balances out the previous lower than average 1.5% growth of 1987-1996.
This statement was also misleading as he made the productivity statement right after saying te U.S. used to have 4% growth a year rather than be consistent and use productivity for both or at least GDP per capita growth.
I also don’t see why Cowen doesn’t just say he wishes we could get back to the late Clinton/ early Bush years instead of going all the way back to the 1960s.
P.S. What annoys me about Cowen’s productivity statements and other cherry picking is that a general audience won’t understand the history even if recent like the 1997-2006 productivity boom. (It wasn’t exactly a boom – I’m just trying to speak ‘Cowen’ here.)
Hey Tyler,
I enjoy reading this blog every day and your articles, but there is something very different about your accent on podcasts. It may just be that the hosts of the shows have great audio voices, and your’s is not bad at all, but is very different. I’ve never heard anyone with your accent, is it an intellectual northeast accent? Don’t get me wrong, I don’t hate your voice and it doesn’t give me a headache, it’s just very unusual. .
I never thought of his accent (we don’t have those in Wisconsin) but his speaking style is one where he likes to get to the point quickly, faster than most people, which I much prefer over the alternative that sometimes his interviewers have.
#2. Astrobotic is also in this market space. Former Google Lunar XPrize team.
https://www.astrobotic.com/
Your answer lifts the intgllieence of the debate.
Comments on this entry are closed.