The 23 November 60-Minutes tore apart Audi, Inc. It seems that the Audi 5000 model (automatic transmission) has a terrible habit of accelerating when moved from PARK to either FORWARD or REVERSE. The problem has been denied by Audi. They blame driver error ("They step on the gas instead of the brake"). Of course this appears to only be a problem with drivers of the 5000 model, none of the other models has such poor drivers. The "alleged" defect is blamed for about 250 known accidents, and at least one death. [See more below.] The alleged causes of the alleged problem (I should have been an alleged lawyer) include 1) excessive pressure build-up in the transmission, 2) a faulty "vacuum" (not sure of the exact words ??) unit (which Audi has voluntarily notified its customers needs replacement, or will result in "performance" problems), and 3) a faulty on-board computer. Although Audi insists that they cannot find a problem with the car, an "independent" expert hired by a group of people that have all experienced problems with the car (there are enough victims out there that a self-help group was formed) actually demonstrated the gas pedal *visibly* moving downward when the car was put into gear causing the alleged surging. Even some valet parking garages have posted signs that they will not accept the Audi 5000 with automatic transmissions. Audi is so convinced that the problem is driver error that they have issued a recall notice to install a safety switch so that the driver could not change the gear unless pressure was on the brake. Footnote: Three accidents have occurred similiar to the stated alleged problem that had the brake safety switch installed. Audi said that 2 of the cars had the switch improperly installed, and the third was unexplained. [Clive Dawson recalled "driver error" being cited for the third case.] This whole incident is reminiscent of the Ford Pinto fiasco. The Federal Transportation Safety Board (??) is investigating. Audi (The Art of Engineering) came out looking very bad. (But what else would you expect from 60 Minutes?) Corporate responsibility appears very low. I would not be surprised if they came out with a corporate apology within a week (in time for the next broadcast) to try to save face. Howard Israel [It is unusual for RISKS to get four different reviews of the same TV program! Excerpts from the others follow, with moderator's effort to minimize duplication and achieve accuracy. PGN] Excerpts-From: Brint Cooper <abc@BRL.ARPA> Even while the driver (quite literally) stands on the brake pedal, the car roars ahead. One young woman ran over and killed her own three-year-old son. The "idle stabilizer" was said to be responsible for keeping a minimum flow of fuel to the engine during idle when the brakes are applied. The idle stabilizer is either a part of a computer-controlled system or is controlled by an on-board computer; it wasn't clear which. Audi denies that anything is wrong. Two Audi representatives appeared on camera to assert that they could find nothing wrong with the car. They even claimed that the motorists are stepping on the wrong pedal. Brint Excerpts-From: firstname.lastname@example.org <Charlie Hurd> The cars have accelerated with enough force to punch through walls. Many of the cars have been totalled. Audi has checked the cars in question and failed to find any defects. They claim that the drivers became confused and pressed on the gas pedal instead of the brake. The drivers (one of them a police officer trained to drive under extreme conditions) maintain that they were trying to put the brake pedal through the floor, without effect. It seems to me that this is good response to Peter Stokes's question (RISKS-4.5) about the risks of buying/driving a car with a computer- controlled engine. The only question I have is why the brakes did not stop the car. Some of the victims said that they had to turn off the car to stop it. Do Audi 5000s have anti-skid braking? Could this have allowed the cars to keep moving? Is this an example of many small malfunctions resulting in a *major* problem? Charlie Excerpts-From: Clive Dawson <AI.CLIVE@MCC.COM> This has resulted in at least one death. A young (6-year-old?) boy was let out of the car to open the garage door, after which the mother stepped on the brake and shifted to forward. The car hit the boy, pushed him completely through the garage door and pinned his already-crushed body against the rear wall of the garage. A heavy black skid mark was left which showed how even then the wheels continued to spin at a high rate of speed. The Audi people claim that all of these accidents are the result of driver error, in which the accelerator is mistaken for the brake. One of the more memorable quotes from Audi: "We're not saying we can't FIND anything wrong with the car; we're saying there ISN'T anything wrong with the car." Attention is focusing on a microprocessor-controlled mechanism which regulates the idle speed. Apparently Audi has sent letters to all owners of the vehicles involved stating that this part will be replaced by Audi for "performance reasons". The report didn't make it clear whether the microprocessor was an integral part of this part or not, so I don't know if this replacement will involve a change in the processor or its software. I don't know what the final verdict on this will be. But listening to that devastated mother tell how she witnessed the death of her son, and knowing the cause might eventually be tracked down to some software bug sent chills down my spine. Clive
I haven't seen it mentioned yet, but I believe that last week I saw a news story that purported to blame a crash of a small light plane in the Southern California area on a changeover of software in either a radar system or a general flight controller computer system, causing either the plane to be lost from the screens or directed into a hillside. Since my memory is vague, perhaps someone else can provide a better recollection of this RISK of computer software. Greg Earle, JPL [The delay in running this item was due to an unsuccessful attempt to get further information... PGN]
Disclaimer: I'm a grad student working in semantics of programming languages, and therefore qualified to pretend to know the answers to these questions. I haven't been studying semantics all that long, though. These are solely my opinions and bear no necessary resemblance to those of my advisor, my department, or my ceramic dragon. > From: dplatt@teknowledge-vaxc.ARPA (Dave Platt) (<a href="/Risks/4/16">RISKS 4.16</a>) > 1) Are existing programming languages constructed in a way that makes > valid proofs-of-correctness practical (or even possible)? I can > imagine that a thoroughly-specified language such as Ada [trademark > (tm) Department of Defense] might be better suited for proofs than > machine language; there's probably a whole spectrum in between. No, they are not. Actually, there are a few existing programming languages (Euclid, for one) which are, but most popular ones are not. A precisely-specified language is easier to prove things about than an imprecisely-specified one, of course. I haven't seen anything approaching a precise mathematical semantics for Ada; if the research in semantics of distributed semantics goes very well we might be able to give you one in ten or fifteen years if we're lucky. The best languages for proving things about are functional languages (FP, Hope, Lucid, ISWIM). I have yet to hear of a "real program" written in any of these. > 2) Is the state of the art well enough advanced to permit proofs of > correctness of programs running in a highly asynchronous, real-time > environment? No. Not even remotely. We can't cope with slightly-asynchronous, non-real-time environments in any general way. > 3) Will the compilers have to be proved mathematically correct also? or > might something like the Ada compiler/toolkit validation be adequate? The compiler will have to be proved too, if the idea of proving programs correct is to make any sense. > 4) The report seems to imply that once a system is proven correct/safe, > it can be assumed to remain so (for the [limited] lifetime of its > License to Operate) so long as maintenance is performed by a > certified software engineer. Is this reasonable? [...] It is reasonable if you re-prove the patched system. I can't imagine it being reasonable otherwise. Note: you can probably patch the proof also, if it is arranged in a nicely modular form. > 5) Many of the program "failures" I've encountered in "stable" software > have been due to unexpected inputs or unplanned conditions, rather than > to any identifiable error in the program itself. Can any proof-of- > correctness guard against this sort of situation? Not really. All the proof guarantees is that the software does what the specification does. That's a big help, since you don't usually have even that. But you have to get the specification right. (I can't even pretend to answer questions about legal aspects.) > 8) Military systems such as the SDI control software would appear to belong > to the "disaster-level" classification... will they be subject to this > level of verification and legal responsibility, or will they be exempted > under national-security laws? [Of course, if an SDI system fails, > I don't suppose that filing lawsuits against the programmer(s) is going > to be at the top of anybody's priority list...] That's a terrifying thought: don't verify Star Wars, it's too secret to have the code so exposed! -- Bard Bloom
Today's Washington Post, Sunday, November 23, 1986, pg K1 [Business Section] contains an interesting article on program trading and the Finance theory behind it all. (The following is partly based on the article and partly from my own knowledge.) The basic idea is to view all of the different financial instruments as interrelated, even though the instruments may be traded on different markets across the country (or around the globe). The people that make it all work are called "Quants" (standing for Quantitative analyst). The "Quants" create models based on the markets and their interelationships and known financial theory. When an "inefficiency" occurs (i.e., the price differential of an underlying security in two or more markets occurs that is big enough to cover the cost of the transactions involved), the computers that monitor the information issue simultanous buy and sell orders in the appropriate places. The net effect is the *total* elimination of risks once the initial set of transactions are complete (the winding up). (This is a simplification of it all. But the previous assertion in today's RISKS entry concerning the "last trader" losing money is not accurate.) The profit is "locked in" when the first set of trades are completed, but will not be actually known until the positions are closed out (winding down) at the end of the finanical instruments life. The "published" profit margin is said to be in the 7% to 9% (annualized) range. (Anything above the current T-bill rate is considered good.) However, only each trader really knows what he is making. (A personal friend on "the street" claims that the profits are really much, much higher because the "invested money" stays in the market a very short time. I am not convinced of this based upon my knowledge of the trading --and "margin"-- necessary.) The "Quants" differ from "Qualitative" traders, in that, Qualitative traders base their trades on the perceived quality of the companies (traditional recommendations of buy company ABC and sell company XYZ). A nice analogy is made in the article to the gambling world. The "Quants" are the bookies, while the "Qualitative" traders are the River Boat Gamblers that bet on instinct. Has anybody thought of the implications (since the computers, based on its programmed models and incoming data) of an error ? Not only is big money involved (it is estimated that one needs a *minimum* of $50 million to play the game), but so are bigger reputations (not just the brokerage houses, but insurance companies, too). Is it "bad" for the market ? I think not. When the computer generated trades are executed they force market correction. The article makes a point that new financial instruments are emerging that will "play the game", much like a mutual fund does now for the small investor. These instruments will limit the "downside" loss, while maintaining unlimited "upside" gain. Then these new instruments can be used in conjuction with the already existing ones to create even more instruments, the end result, potentially being, that in time anyone can bet the market in any way. ---H
In response to Roger Mann: (I mentioned this about a year ago in our last discussion of computerized stock markets.) Instantaneous and non-instantaneous negative feedback to not produce the same results. In this case, the fact that thousands of computers can respond to the possibility of profit before the effects of the responses get back (through whatever feedback loop) to any of them, opens the door for disaster. Eric Nickell
The problem of decreasing system stability as time constants change is not a new one. Take the steam engine: "Watt's use of the flyball governor can be taken as the starting point for the development of automatic control as a science. The early Watt governors worked satisfactorily, no doubt largely due to the considerable amounts of friction present in their mechanism, and the device was therefore widely adopted. ... However, during the middle of the 19th century, as engine designs changed and manufacturing techniques improved, an increasing tendency for such systems to hunt became apparent; that is, for the engine speed to vary cyclically with time. ... This problem of the hunting of governed engines became a very serious one (75,000 engines, large numbers of them hunting!) and so attracted the attention of a number of outstandingly able engineers and physicists. It was solved by classic investigations made by Maxwell, who founded the theory of automatic control systems with his paper "On Governors," and by the Russian engineer Vyschnegradsky, who published his results in terms of a design rule, relating the engineering parameters of the system to its stability. Vyschnegradsky's analysis showed that the engine design changes which had been taking place since Watt's time - a decrease in friction due o improved manufacturing techniques, a decreased moment of inertia arising from the use of smaller flywheels, and an increased mass of flyball weights to cope with larger steam valves - were all destabilizing..." "The Development of Frequency-Response Methods in Automatic Control", Alistair G. J. MacFarlane, IEEE Trans. Automat. Contr., pp. 250 - 265, Apr. 1979
Those interested in the recent item on the science of decision-making [see Jim Horning, "Framing of Life-and-Death Situations", Risks 4.13] might find this reference a bit more accessible: "Decisions, Decisions", by Kevin McKean. DISCOVER Magazine, June, 1985. This article is a well written account of the work done by a number of researchers, notably Daniel Kahneman and Amos Tversky, and has several very nice examples of how the framing of a question affects the decision making process. Anybody who had trouble locating CMU's "1986 Accent on Research Magazine" would have better luck with Discover Magazine.
Please report problems with the web pages to the maintainer