Science and journalism

Today in science:

Driverless cars will be given MORALS: Scientists develop an ethical formula which will enable them to make life or death decisions

Three years ago in journalism:

When Should Your Driverless Car From Google Be Allowed To Kill You?

It’s not and never has been about whether we can code an ethical formula, it’s what in buggery should that formula be?

41 comments on “Science and journalism

  1. Issac Azimov tackled this problem decades ago. His 3 laws of robotics are a good place to start.

  2. @Kevin Lohse – this is more about when a death is unavoidable, which of the humans should it choose to kill…

    i.e. there is car coming towards to you at 70 mph and cars either side… does your car take the hit and let the others either side live, do you swerve into one of the other cars, pushing them out the way so you survive but they may die, etc…

    these scenarios will happen (brake failure, mechanical failure, wind blowing debris around)…

  3. What would happen if two different systems come to two different answers and so both end up taking worst case scenarios… It’s a fun game to play in theory but not something i would want to code in real life!!

  4. Hacking?

    The car might not try to kill you new from the factory but nothing inspires any confidence that some teenage hacker won’t decide to take control and kill me for laughs.

  5. Car has to choose between running into a tree (killing you) or a pushchair (killing baby).
    Extra points if you can’t see if there’s a baby in the pushchair…

  6. Tim Newman is entirely correct.

    Such bullshit as that article is like Sinclair C-5 owners talking about what they are going to spend their motor-racing winnings on.

  7. The driverless car will never go so fast that it would kill you if it hit a tree. So the higher the max speed that you want, the more padding (rubber bumpers, airbags etc) you need and the heavier the car is so the more protection you need for pedestrians, so again the heavier the car is and the bigger electrical engine and battery you need.
    The energy/speed curve ends up somewhere between a parabola and exponential.

  8. Pretty simple really, the driverless car (which we won’t get for many years after it is technically possible) will act to preserve the life of its passengers, exactly the way that a human driver would. An empty one would sacrifice itself. My prediction remains that we will have driverless electric buses enabling every city to have a free 24 hour instant tram service. If we want more point to point then a rental car service will send us a driverless car to our door and we will ‘drive’ with huge driver assist. Then we will send the car back under its own steam. Equally Uber drivers will exist but the electric car will do most of the work.

  9. It’s the trolley problem in drag. No one’s solved it, they just have opinions.

  10. Surely market forces would result in a car that protects the occupier. Do you buy Car A that kills you in preference to outsiders, or Car B that kills outsiders in preference to you? Oooo, let me spend my money on something that’s actively designed to kill me, instead of a fungible product that is actively designed not to, please, please!!!

  11. The altruistic driverless car should be designed to leave flowers, a framed photograph of the driver and an optional teddy bear when its final journey culminates in extreme prejudice.

  12. I’m with Tim Newman on this one.

    If we do ever get driverless cars, they’ll only be allowed on motorways and limited-access dual carriageways, and only on days when the traffic is light and the weather is clear. In other words, never. If they nevertheless do become popular, expect calls for more motorways to be built; obviously these will be tolled.

    The one type of driverless car we will see (and already do in a handful of places) is the slow pod; but these are public transport, not private cars.

  13. Personally I would like them to hunt down cyclists that jump traffic lighs and pass on the left.

  14. I would like them to hunt down cyclists that jump traffic lights”: there’s one junction near us where the cyclists do everyone a favour when they jump the lights and thereby get out of the way of drivers. You could argue that the system should be formalised at that junction but as a practical matter it works awfully well.

  15. dearime: as someone who has on several occasions been mowed down by cyclists crossing on red while I as a pedestrian have confined my crossing habits to the green man, I am somewhat biased.

  16. Brad Templeton has written extensively about this “problem” in his blog.

    His basic position is “nothing changes – the autonomous car will obey the law, and if it breaks it, then someone will pay”.

  17. All this Asimnov bollox was disproved in Dr Who “The Robots of Death” which includes the immortal line “Please do not throw hands at me !”

    The obvious answer is that if a robot car is about to be involved in an accident, it pours a lethal nerve gas into the car, killing the passengers and then pulls over to the hard shoulder and immolates itself, thus saving on funeral costs.
    They’d have to have decent air.con,though so as one doesn’t drive everywhere with the windows open.

  18. Asimov’s Three Laws are about as much use for designing safe Artificial Intelligences as a chocolate teapot. If you’re talking about Artificial General Intelligences (AGIs) then the sad fact is that at present no-one has a solution for AI safety. One of the chief problems is that we are unlikely to know the AGI’s utility function except possibly in terms of its eventual goals and so we can’t tell where it would regard as instrumental values things we would regard as intrinsic values (i.e. it would treat ends as means).

    AI safety is probably the hardest problem in computer science at the moment. Aligning the AGI’s goals and ours is something we don’t have a Scooby about, and it’s entirely possible that before long these things are going to be much, much smarter than us.

  19. I agree with comments above. The robot car’s prime duty is to protect its occupants.

    This includes not wasting processing cycles on deciding how least to injure others if it comes to that. Write one line of such code, and you become liable for any outcome.

  20. @mark T, July 5, 2017 at 11:05 am
    “Pretty simple really, the driverless car (which we won’t get for many years after it is technically possible) will act to preserve the life of its passengers…”

    Agree.

    However, there should be an option to enable SJW bleeding heart mode where occupants suffer instead.

  21. “However, there should be an option to enable SJW bleeding heart mode where occupants suffer instead.”

    Yeah, a “donor button.”

  22. Driverless cars wil come when insurance companies decide that the risks and liabilities of driverless cars are a damned site less than having idiots behind the wheel. At that point insurance for drivers will start to increase very quickly.

  23. “At that point insurance for drivers will start to increase very quickly.”

    I doubt it. The business model won’t change, and someone in the marketplace will price it competitively.

  24. BiCR has the measure of it. This isn’t a question of writing an algorithm so much as training an AI system to make the right decisions and getting that system to fit on a reasonably sized piece of silicon, bearing in mind that the computer that IBM built to win a simple quiz game 5 years ago filled a room and required an impractical amount of power to put it in a car.

  25. I don’t think we’re looking at this the right way. There will never be such a thing as ‘A’ driverless car. Doesn’t make sense. Why would one want to duplicate all that processing power making individual cars AI? What you will have is a transport network consisting of individual cars sharing their positional, routing requirements & sensor data with a coordinating centre. So there should be no instances of cars being on a collision course. And the incorporation in the car’s guidance of some simple flocking rules – already being incorporated in autonomous aerial drone swarms – would take care of any anomalous trajectory conflicts.
    So pretty well all of the “difficult choice” scenarios disappear. The only things left are the sort of situations where no human driver could improve on the traffic management system’s performance. The human not having a fraction of the system’s situational awareness.

  26. The changes will be incremental, just as ABS and stability control are now standard fit, lane following , automatic braking, speed control linked to GPS and rain sensors etc will become mandatory on new cars. Eventually you will not be allowed on motorways or city centres unless you have x safety systems enabled.

  27. I don’t understand all this talk of driverless cars when they should be making driverless trains first. They don’t even need to steer and there would be far fewer unions / strikes.

  28. History solved this problem more than a century ago. Now we have robots and robotics, we simply need to have a robot walking at 5mph holding a red flag in front of its client robot car. Even if there is a collision, no-one dies.

  29. If you have a travelcard and a few hours to spare, the front seat of a DLR is surprisingly entertaining.

  30. @john77, July 6, 2017 at 9:56 am

    @ Samuelbuca
    They *have* made driverless trains. The DLR uses them.

    As does the Victoria line since it opened, but unions insisted a paid driver still had to sit in cab.

  31. the front seat of a DLR is surprisingly entertaining.
    Agreed. I was sitting there once when a DLR uniformed type got on and requested (politely) that I move. He proceeded to unlock the ‘box’ beneath the front window, which opened up to reveal a set of controls – and drove the train for the next few stops.

    My guess is that they are required to be able to take over a train in the event of an automation failure and need to maintain a degree of ‘currency’.

  32. ‘So there should be no instances of cars being on a collision course.’

    Unless Gamecock edges over within inches of them to see how they react.

    Or teenagers drive erratically around them. Surely that could never happen.

Leave a Reply

Name and email are required. Your email address will not be published.