Soft Interfaces Shouldnt Snarl Hardware Functions
"The hackers denounced it," narrated Steven Levy in his 1984 book "Hackers." Levy quotes Russell as later realizing that "weapons or tools that arent very trustworthy are held in very low esteem." And so, the space torpedoes were fixed with a few rewritten instructions.
I thought of that ancient episode when I saw Logitechs background material on the Agilent "MX Optical Engine" in Logitechs latest generation of optical mouse devices. Early optical mice were famous for indicating erratic, random movements when their tracking algorithms got confused: In fact, its daunting to realize how many things an optical mouse has to do correctly if its going to provide a response that feels right. And we certainly know, and care, when any kind of human-machine interaction feels wrong.
Pardon me if I feel slightly overwhelmed--even outraged--by the MX Engines consumption of what used to be scarce bit-processing resources. A Logitech MX mouse is processing 4.7 megapixels per second, as it tries to decide how the scene immediately underneath the mouse has changed from one moment to the next--and to deduce thereby how far the mouse has been moved, and in what direction. The digital signal processing sensor resolves movement at up to 800 dots per inch.
The result, Logitech claims, is accurate tracking at up to 1 meter per second with accelerations of up to 10 gravities--that is, reaching 1 meter per second in one hundredth of a second. I dont know if I can do it, but I know that Im not going to try.
I also know that the optics, signal processing and wireless link technology in the latest Logitech devices (providing 125 position updates per second) make these things feel right. Which is the whole point, isnt it? But I still suspect, on a "guilty until proven innocent" basis, anything that tries to put sensors and software (often called a "fly by wire" arrangement) into the loop between me and any device whose misbehavior could kill me--or even just make me uncomfortable.
I dont have to go far for an example of "fly by wire" that is not as good as the original. In my own ham radio setup, theres still a 1980s short-wave receiver whose tuning knob actually turns a variable capacitor to adjust the received signal frequency: The frequency readout is digital, but its just measuring the continuously variable frequency to the nearest kiloHertz. When I tune across a signal on the band, the tone varies smoothly--like a slide whistle. If I want to tune to the nearest Hertz, thats my option.
My newer radios are digitally tuned, with "optical encoder" tuning knobs that interrupt light beams to tell an internal counter to "go up" or "go down": The signals pitch changes in abrupt jumps that Ill never really accept as sounding "right." But if Im momentarily confused by a radio, thats not a big problem--unless Im driving at the time, and BMW is conducting an interesting experiment on how confused people can get without killing themselves in traffic.
The whole idea of BMWs "iDrive" controller, a push-pull-twist universal knob for controlling all manner of automobile systems, gives me the willies. Rental cars are bad enough when their headlight switches arent in the right place: A completely soft interface introduces completely new ways to mess up human-machine interaction.
My favorite example of getting it wrong is the F-16 fighter jet, whose original fly-by-wire software tried to protect the pilot from rolling the aircraft so quickly that the pilots neck would be broken: Unfortunately, a bug in the navigation software made the plane think that it was upside down whenever it crossed the equator, and the resulting autopilot input to roll the aircraft was on the back side of that buffer. They caught this one in simulation. We wont always be so lucky.
Its taken almost 40 years, beginning with Douglas Engelbarts original mouse, to produce a fly-by-wire (or perhaps I should say, "fly-without-a-wire") mouse that lets us forget that its there. My newest radios, likewise, finally improve on what came before with tuning steps that become more precise as I turn the knob more slowly. Improvement is possible, but it takes time.
Anyone who wants to introduce a fly-by-wire approach, or augment it with force-feedback "haptic" enhancements, had better be willing to meet users of what works now on their own terms. Or expect to be a target for their torpedoes.