1. The joke is that if a self-driving car can’t recognize stoplights it is failing at one of its most important jobs and will probably kill people. The joke is based on captchas that use identifying common things to prove someone is not a robot.

  2. It also may be playing off of the widespread belief that the human responses to this sort of variety of Captcha are used in training the self driving cars. Which for safety really do need to recognize bicycles, stoplights, hills, toys in the street, and parked cars.

  3. Madmup and Dana have it and for me it was a LOL. The very sort of test that is supposed to keep a “robot” out of computer systems will also prevent the self driving car from being effective. There was another pop up with the cartoon that said the artist was still debating whether he should’ve depicted the car covered in blood.

  4. I am curious, however, why the robot needed to get to a particular site . . . robotporn? Isn’t a self-driving car, by definition, already computerized? And far as I know, GoogleMaps doesn’t have a Captcha.

  5. I got it when I read it here, but didn’t think it funny, for the same reason we were discussing in another thread — yeah, the whole point of these captchas we see is Google trying to crowd-source its training of its autonomous cars — it’s killing two birds with one stone, in what could be argued is an elegant way. You are not going to be using the very expensive self-driving neural net they have to be solving capchas to send spam. Once it gets to the point where this net they’ve developed is cheap enough and ubiquitous enough that spammers can use it, then obviously it won’t be an effective capcha anymore, and we will have pivoted to something else. I guess there could be a small bit of overlap right now in that Google claims they will be doing self-driving taxis “any moment now”, but are still — maybe desperately — having us humans try to teach their robots how to recognize traffic lights. Or maybe they are just ever refining their system in this time before their net becomes ubiquitous and cheap enough that they can’t use this for capcha checking anymore.
    Things change — that’s the nature of tech. Only the telecom monopolies think tech will stay static enough to keep a good-enough solution around for decades on end and charge usurious rents on it…

  6. To put it more simply, I’d say it’s just a play on the word “robot “. The car is a robot in the literal sense. In the world of internet security, a robot is just software designed to get access to an application.

  7. I am also confuse about the setup and the supposed joke of this comic. My vague guess is, that this is an autonomous “Uber” car and it can’t get into it’s own ride handling app. Which would be very strange, since it wouldn’t need to use a smartphone for this, it would be an integral part of it’s soft- and hardware.

    Also, there is this hover-text on the original site: “I seriously debated whether the car should be covered with blood, and I’m still not sure I made the right call.”
    Which does not fit to my explanation, but rather with the “cannot even recognize traffic lights”. So, nope, I still do not get the joke.

  8. I think some commenters may be getting led astray by the Captcha format and the context we come to it for. This autonomous car guidance system is just being tested; because they need to be tested! It was not trying to enter a website or an app.

    The developers plopped it down in front of a Captcha grid for this particular test. They could also have it drive around a secured test track. Eventually they could let it out on the streets and see how it does in a situation that would be not merely lifelike but really life! Then if it failed … really badly …. we would be in the case contemplated in the I seriously debated whether the car should be covered with blood remark.

  9. ONLY ’cause we’re discussing robots . . . this Spot fascinates me, particularly because IT is being called HIM (to me, it’d be a HER, as I don’t see any appendages identifying IT as a HIM, but I digress).

    “Don’t anthropomorphize your robots; they really hate that.”

  10. I’m wondering now if any jurisdictions will mandate a remote Law Enforcement Override on autonomous cars.

    There were some SF stories or novels about some kind of future highway patrol on very advanced-technology highways, where the LE had an override beam to slow and stop any vehicle — I don’t recall how autonomous they were supposed to be, but even when not driverless you got the impression the driver was not handling low-level maneuvering as we do today. So instead of telling you to pull over, the cops would just pull you over by remote command.

  11. Meant to say, I was thinking the name Ben Bova for writing those, but can’t track it down.

  12. Danny Boy: Alliteration has led you astray; you’re remembering Rick Raphael, whose three superhighway stories are collected in the book /Code Three/.

  13. I get that a self-driving car that can’t recognize traffic lights would be bad, and that it’s widely believed that human responses are used to produce training data. But I can’t quite put those things together to make sense of what’s going on here. I see a few different options, none of which quite makes sense. To spell them out in gory detail:

    Option 1: The scene in the grid is what’s actually in front of the self-driving taxi, and it’s sending it to humans. But if the humans are there to provide ground truth, they can’t fail. And if the car knows the ground truth, why would it care if the humans fail?
    Option 2: The scene in the grid is what’s actually in front of the self-driving taxi, and it’s taking the test itself. But then who’s providing ground truth to allow the taxi to fail? And why does the taxi even need to take a test if the system already has ground truth (i.e. already knows where the stoplights are)?
    Option 3: The self-driving taxi is failing a test unrelated to what’s currently around it, because it needs to pass the test for its job. But if it needs to pass the test to qualify, why is it on the street now? (And if it’s not currently driving, then the situtation doesn’t seem funny, because it’s exactly what we have now: people are trying to make self-driving cars, but they’re not accurate enough to be on the street.)
    Option 4: The self-driving taxi is failing a test unrelated to what’s currently around it, but it’s not related to its job. That’s self-consistent, but doesn’t still doesn’t hang together quite right. The taxi is just browsing some random website, unrelated to its job, while out driving on the street?

  14. Thank you, Mark Jackson!

    It occurs to me that one or more of these may have appeared serialized in Analog and Ben Bova was the editor around then.

  15. WW, it’s pretty much your option 3. Or option 2 but without your question — the scenes are cooked, there is no currently-live panel of humans also taking the test. It’s not hard for the testers to cook it, just pick whatever scenes they want. Oh no, you specified option 2 was “what’s actually in front of the self-driving taxi” so my objection doesn’t fit. But both your 1 and 2 say that, and it’s clearly not what’s going on — the taxi isn’t driving around, it’s parked at a big wall where multiple screens present the test.

    So it’s your 3.

    But why is there a problem? ” why is it on the street now? ” Either it’s a new candidate and needs to pass to get a license. (so is not on the street now.) Or it has to renew and retest every so often. (And is on the street because it passed an exam on a previous occasion.)

  16. @Danny Boy – London Derriere – My copy of Code Three is from 1965. It did come from Analog – as novellas – but long before Bova took over. 1963 – 1965 based on the copyright page.

  17. I was listening to a podcast (maybe “Short Wave”) with a guest who had done some journalism with researchers and sceptics of current AI, in particular what they called machine learning or ML. I gather they are not using the simple-minded back-propagation that was big when I studied this a little, but there is still some form of scoring and use of that as feedback to adjust the parameters of the algorithms.

    An example flub the guest was quite amused by — and apparently was told to her in a spirit of openness by the team that ran into it — involved a program that viewed medical X-Ray images and, among other things, identified which of them showed a fracture. The training data and the test data came from several local hospitals, and apparently had ID panels not blanked from the images being used. The program was surprisingly good at identifying images that showed fractures.

    But it was not equally good at other tasks. The opacity of the scoring and feedback system made it hard to figure out why, but eventually they did. One of the hospitals had an ortho unit with a large number of actual fractures treated. And what the system had learned was to spot the logo of that hospital and raise the score for likely fracture!

  18. Thank you, Scott. I guess my bringing Bova into it was just from being a little muddled. But it was definitely the CODE THREE stories I was remembering.

  19. Danny Boy: If it’s option 3, and this is a test for its job, why would the test say “prove you are not a robot”? The point of the test would be that they want robots to prove that they can drive, not to pretend to not be robots.

  20. And in the last panel, it’s holding a smartphone or equivalent up to its camera. It’s not parked in front of a big screen, it’s looking at a little screen. Why it would need to solve said captcha – no clue. Except in that, as others have said, what we need to solve as captchas is what a self-driving car needs to solve for its job…and apparently this one isn’t very good at it.

  21. To me this comic isn’t all that complicated. We the reader recognize the captcha (a word I can never remember to spell or pronounce properly) from a variety of sites to keep bots out, and does often ask for traffic lights among other things. So the comic sets us up for the expectation is that the viewer of this particular captcha is human, only for us to find that it’s actually an autonomous vehicle, something presumably that would need to be able to spot traffic lights.

    I had not thought of this apparent contradiction myself, so this comic made me smile a little.

  22. The car is having an existential crisis. Like many of us have at work. It realizes that it is wasting it’s life in a role that provides no emotional satisfaction or fulfillment. It’s just mindless rote tasks that it is not even particularly good at. Why wouldn’t an AI have a mid-life crisis?

  23. “Why wouldn’t an AI have a mid-life crisis?”

    Would you call it that, tho, when there is no real life, and who knows what the mid-existence of a robot (AI is still a robot) would be.

  24. Andréa, it comes down to how one is going to define life. A true artificial intelligence that is free and self-aware would surely ponder itself and its existence. Think of C-3P0 in the Star Wars. He is aware of his existence and does not wish to be destroyed (i.e. die). He has free will–or he would if he weren’t such a house droid. The best part of all Star Wars movies made after Return of the Jedi was the part in “Solo: A Star Wars Story” where the droid is a droid-slavery abolitionist. The morality of enslaving a highly sentient being, especially one that can communicate its desire to be free, has always been an issue that Star Wars has ignored.

  25. Andréa: A movie is not real, but it can still be used to give examples to consider. Otherwise we could never discuss hypothetical situations.

    Singapore Bill: I didn’t see Solo, but sometimes it works better to just gloss over issues like that. In the Harry Potter universe, the house elves are either fun creates that enjoy their work, or horribly enslaved. I found it very strange when Rowlings decided to make it an issue that they were horribly enslaved, but also make it a silly little side story, with Hermione being a little “annoying” and “over the top” about how they were horribly enslaved.

    If I think about C-3PO being a self-aware slave, then that brings up all kinds of questions. Like what sort of sadistic creators designed him to be neurotic and terrified all the time?

  26. The most heart-wrenching movie I’ve ever seen about AI is Steven Spielberg’s “AI: Artificial Intelligence”, with Haley Joel Osment and Jude Law. I saw it once, bought the DVD, but was never able to watch it again. I think Spielberg should’ve won an award for that movie; so much better than all that Star Stuff [I forget which he did]. (But then, I thought Jim Carrey should’ve won awards for “The Truman Show”, so what do I know.)

  27. WW: I agree that you can push it to the back of your mind and just let Star Wars go about its business. But a film in the franchise brought it up and it is now canon that at least some droids desire freedom and are willing to fight for it. Given that Solo was set about 10 years before Star Wars, it seems odd, because it’s not addressed in the original trilogy. The worst part, though, was that it was clear that this abolitionist droid was played for laughs (I saw it in the theatre and all its anti-slavery stuff got plenty of laughs).

    You could argue that C-3P0 naturally developed that personality as a result of his makeup and his experiences–nature and nurture. What’s really sadistic, is building droids so they can feel pain, which this torture scene proves they can. Though I guess pain is useful for self-preservation.

  28. Singapore Bill: That sounds similar to the Harry Potter house elves: Better to ignore the issue than bring it up and treat it as a funny non-issue.

    (As for continuity, the prequels bring up all countless unanswerable questions about the original trilogy.)

  29. GPS has to be a lot better before I will trust a self- driving car.

    Two of the ones Robert has think our driveway is a street. When we are leaving the house and parked on the driveway it instructs us to drive “straight ahead” and to turn on the street behind our house – this would involve driving through our garage, our shed, the back fence and the house behind ours.

    Several of the GPS devices he has had over the years cannot find our house. The houses on our street have 15 houses – including ours – which are in a different community than the others on the street and are numbered separately (our house is in the 2000s, house facing us is in just over 1000). When he puts in our address it takes us to the same number on the same road, but one in the community to the south of us – to senior apartments (starting long before we qualified to live there) .

    The main one he uses these days likes to have us make U turns. Best one of these was got off I78 in PA to go to a McDonalds for lunch which was south of I 78. Correct exit – at the light at the foot of the exit ramp it said to turn right. I told Robert to turn left (or we would be headed north) – he turned right. We were sent two blocks, then told to turn right, turn right on the next street (parallel to the one we had turned right onto) and then to drive straight ahead. When we reached the foot of of I78 and could go no further, it told us to make a U turn. It then sent us into the back of a small strip mall, out the front of it and then told us to make a left just north of where we should turned left in the first place – only now we were doing so without a traffic light and through the cross traffic on a main road.

    i would hate to think a self driving car following its GPS would do all these things, as well as the rest of the errors we have found since he started GPS.

  30. It always makes me cringe when people say “GPS” and talk about streets and finding a route. The Global Positioning System tells you at which coordinates on the earths surface you are and it’s (deliberately) not too precise at that.

    I know, it’s somehow found it’s way into american vernacular, but can we please call it navigation system? For navigation your position is not enough, you also need to know where streets (and one-ways) are, at which coordinates a certain address is located and be able to plot a course from your position to the destination.

    GPS often cannot distinguish between the street and your driveway. But after you drove a bit the navigation system can guess that you are not plowing through front lawns and will correct your position. Also many systems have map data which is not really up to speed, so you might get the strange (and dangerous) route you described. If this happens once, you were just unlucky. If it happens all the time you should get a different navigation system.


    P.S.: All being said, the joke still eludes me.

  31. “But after you drove a bit the navigation system can guess that you are not plowing through front lawns and will correct your position”

    How does a non-sentient machine ‘guess’?

    A friend in England uses the term, SatNav, and I’ve grown to like it.

  32. How does a non-sentient machine ‘guess’?

    But Andréa, we all say things like that nowadays, quite innocently.

    Sure, there might be a joshing philosopher-of-mind who seizes on someone saying “When you click that, it thinks you want to return to the previous screen, but it can’t, so it decides to instead show you the results of a prior search” and confronts them with “Aha! So you agree with me, a computer can think, and can decide!”. Ignoring the difference between “thinks the user’s keystroke means something” and “thinks” simpliciter.

  33. Markus, of course I agree with your observation that in casual speech the term GPS has broadened to also cover automobile navigation systems — which embody and make use of a GPS component, but do something more with it, which strictly speaking is not part of the GPS.

    Still, the broadening strikes me as pretty innocent. Where it can become a problem and mislead someone is when there is also a subsequent narrowing, and someone thinks a use of “GPS” must require a whole nav system.

    There was a fan podcast for a TV show where I heard someone say something weirdly off kilter, I thought — until I realized they must be thinking in that re-narrowed way. On the show, some law-enforcement people, or maybe some spies (I think this was The Americans) attached what they termed “a GPS tracker” or something close to that, to the underside of a car used by someone they needed to locate later though they couldn’t risk being seen following (or didn’t have the spare personnel, etc).

    And one member of that podcasting couple, during their episode summary, said something like “I don’t get what good it would do them anyway. You’d have to know where the car has gotten to anyway, to go retrieve the GPS and look at the screen and see the history. That’s not going to help them in hot pursuit, etc”. I wanted to tell her “No, you just need to tune in to whatever network it’s continuously transmitting the locations it calculates. On.”

  34. GPS? Next you’ll complain about dialing a phone that has no dial, just buttons. Or using the word “phone” for anything but a phone. “If you have your phone on you, can you take a picture of this?”

  35. I had a friend who was thinking of getting a new phone and said, in all seriousness, that the only thing wrong with his current phone was that for some reason it couldn’t make or receive phone calls. He seemed to feel that this was, on the balance, a relatively minor defect for a phone to have.

  36. And that’s the only thing my dumPhone does, thankfully. Altho I have it on silent 99.999% of the time, so it doesn’t even do much of that.

  37. Andréa: Is your dumPhone a mobile phone? I haven’t seen a mobile phone in over a decade that can’t take pictures. (Mine is a little more advanced: It can take pictures, and you can also play “Snake” on it.)

Add a Comment

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s