Sunday, 7 October 2012

Breaking the (ecological) law: why illusory sounds don't make for safer cars

Electric cars are great for the environment, but they come with a problem - they're too quiet, and this makes them dangerous to pedestrians used to the loud noises of the internal combustion engine. One idea is to add noise to the cars, and Mark Changizi recently wondered whether perceptual psychologists could help design a better sound, using an illusion. I don't think it would work, and the reason is a nice example of why it's important to understand the relationship between the world and perceptual information about the world.

Events in the world create information. When a car moves towards you, for example, there are sounds coming from the engine, and the way these sounds change over time is information about the heading of the car. If it's getting louder over time, it's coming towards you; but if the pitch is decreasing at the same time, then the car is going to pass you, not hit you. Increasing volume and constant pitch, however, specifies a collision and the particular rate of change of these variables tells you about the time-to-contact. These patterns of change over time in the acoustic array are related to the way the car is moving via the laws of physics and are therefore informative about the details of the car's motion. Importantly, the relationship between these patterns is also governed by the laws of physics; increasing volume and constant pitch only go together when the car is really heading towards you, for example.

Every perceptual system has limits, though, and those limits are called thresholds. When a stimulus is outside a threshold (e.g. too quiet, or too high pitched) the perceptual system won't respond to it.  The problem with electric cars is that they are quiet at low speeds, and so the variation in pitch you need to detect a collision is small and possibly below threshold. You could make the cars louder, but this goes against one of the selling points of the cars - reduced noise pollution. Mark suggested creating an artificial sound, an illusion, in which the range of the variation in pitch created by the slowly moving car is amplified without just making the car louder. This is in principle possible, but it's a problem because it's breaking the law, and you end up with less information about the car than when you started.

Breaking the law
A commenter on Mark's blog quickly noted
I think we use a combination of increasing loudness and changes in pitch to figure the trajectory of the car. With your scissor-device, the pitch would tell us its approaching quickly, but the volume would tell us its approaching slowly. I’m not sure exactly what peoples brains would end up telling them in that case.

Mark replied
My thought is that once the market begins using some characteristic “illusory scissor” sound, pedestrians will begin to recognize that particular sound, and re-calibrate speed for that.
He's suggesting that people will learn about the new combination of sounds, and learn what it means. This is entirely possible; this is of course what happens with experience with the original, lawfully created collection of sounds in the first place. People experience patterns of increasing loudness, for example, and over time learn the relationship between 'rate of loudness change' and 'speed of car'. Learning this relationship is called calibration, and a calibrated auditory system that detects a given rate of loudness change can be said to have perceived the related speed. Calibration is an ongoing process; we regularly recalibrate our perceptual systems to handle local conditions, and so Mark is right - people should be able to learn the new relationship between pitch change and the speed and trajectory of the car.

However, there's a real problem here. If you make a car where the change in loudness and the change in pitch are created by two separate, unrelated mechanisms, then they are no longer related to the motion of the car or to each other in the lawful way they typically are. This means they that aren't telling you about the same thing, and could end up conflicting with one another in a way that is typically impossible.

Worse, only electric cars will have this new relationship between the sounds and the behaviour of the car; every other moving object the person encounters will have the original relationship. People will therefore have to learn a) to discriminate electric cars as being different from every other moving object, and b) learn the unique mapping between the behaviour of the electric car and the information it is generating.

People may be able to do this; given enough time and experience, there is a lot of flexibility in the way people carve up their perceptual space. However, there are limits on recalibration (you can't learn any old mapping and you can't make every possible discrimination) and the electric car is up against every other moving object the person experiences. In addition, what about a visitor to a city full of electric cars with no experience of this sound that breaks the basic laws of physics? This person would stand a real chance of getting run over, probably more of a chance than if interacting with a quiet car because of the confusion this illusory combination of sounds would create.

Physics: it's not just the law, it's a good idea
Mark's solution to the problem of sub-threshold variation in pitch was to artificially expand the range over which the pitch of the car's sound changed so that it was more easily detected. This is quite an interesting idea, but it comes with a cost - it breaks the lawful relationship between the actual motion of the car and the way the car's sounds change over time. This is a problem because what people actually need to know is the actual motion of the car, and our perceptual systems have evolved to depend on the typically law-based relationship between events in the world and the information about those events. Interestingly, all the existing systems follow this principle and add sounds to the car based on the car's actual behaviour.

If you want to understand (and maybe manipulate) the information a person uses to know about the world, you have to understand three things:
  1. the dynamics of the event in the world
  2. the kinematic patterns of information it creates, and 
  3. the process by which this information emerges from the dynamics of the event
Information is all the observer has - there is no 'peeking behind the curtain' at the event in the world to check to see if we're right. It requires careful science to do this peeking,and if you really want to create artificial sounds to enhance people's ability to hear that electric car coming, you need to know about all these elements. In the meantime, physics has our back and will continue to produce meaningful collections of changing patterns in sound and light for us to use, and we can dodge those cars by keeping our eyes and ears open.


  1. One of my favorite videos for demonstrating conditioning is relevant: the amazing power of a bicycle bell.

    Of course, the original author is correct, that we can take almost any arbitrary sound and get people to respond the way we want with enough experience.

    I think the more important point though, which you emphasize towards the end, is that the cost of errors during training is pretty high.... and we would expect training to be difficult because every other noisy object the person experiences will work differently! It would be much better to train visual vigilance. Didn't anyone tell these people to look before they cross the street?!?

    1. Didn't anyone tell these people to look before they cross the street?!?
      Seriously, although in fairness sounds are quite useful for pulling attention from your iPhone which is no small thing.

  2. I wonder if part of the problem or concern about 'artificial' car sounds might revolve around an over-simplification of the nature of acoustic 'information'. Pitch (which seems to refer only to sine tone frequency) and intensity are not the only aspects of sound that change with distance and changing position. For example, high frequencies are more quickly absorbed in the air than low frequencies and so more distant sounds have more energy in the lower end of the sound spectrum. Also, acoustic reflections (which would be more likely from a car in a city than the wide open country) are particular to the geometry of the listener in relation to the sounding object. My suggestion is that when a sound is sufficiently broadband, and more than just two basic acoustic forms of information are available, there may yet be some form of multi-auditory specification available (thinking of an intra-modality global array) in which physics can still trump the 'artificial-ness' of the sound. That said, the nature of the sound itself would probably require more thought than the usual map the pitch of a sine tone onto whatever variable you want to sonify.

    1. It sounds like you're suggesting a more sophisticated version of what I am - using real sounds that relate lawfully to the motion of the car, where 'real sounds' have a lot of meaningful dimensions along which to vary. I'm all for it; I just don't know a huge amount about the psychophysics of sound, so I stuck to the parameters Mark laid out.

  3. theirs always a saying that look before you cross the street..if you are thinking something while you where crossing the street it can really distract you and the noise will help you back in your senses