On 31 August Johnny Carson did an impression of Abe Lincoln as a budding young standup comic, wearing a two-foot top-hat. The end of the skit involved triggering a radio-controlled device that blew the hat upwards off of his head. A long delay arose during the taping of the show when the hat kept blowing off Johnny's head backstage before his entrance. His technicians figured out that RF interference from nearby kids with remote-control robot cars kept triggering the hat. The obvious solution was to get the kids to stop transmitting during the skit. Honest, Abe had some good lines, but Johnny's hair-raising gimmick top-hatted them all.
There is a very nice article by Jonathan Jacky entitled "Programmed for Disaster, Software Errors That Imperial Lives", in THE SCIENCES, the bimonthly of The New York Academy of Sciences, September/October 1989. Jon gives an in-depth analysis of the Therac 25 case. (The credit line notes that Jon is currently designing softwre for a computer-controlled radiation-therapy machine.)
By George C. Wilson Washington Post Service The USS Vincennes, which shot down a civilian airliner in the Perian Gulf on July 3, 1988, killing 290 people, had gained a reputation for being an overly aggressive "robo-cuiser" and probably provoked the sea battle with Iranian gun- boats which preceded the shootdown, according to a commander of a ship on the scene at the time. Writing in the September issue of the US Naval Institute magazine _Proceedings_, Commander David Carlson said the radar on his ship, the USS Sides, a patrol and escort frigate, showed that the airliner was climbing in a non-threatening way inside the designated commercial airline corridor. He said he would not have ordered his own crew to fire missiles at the plane, as did Captain Will Rogers III, skipper of the Vincennes. In another part of his remarkably frank account, which is likely to refuel con- troversy generated by the shootdown, Carlson wrote that the Vincennes started the sea battle in response to what may have been nothing more than warning shots from Iranian gunboats feeling threatened by the cruiser's helicopter. Having watched the performance of the Vincennes for a month before the incident," Carlson wrote, "my impression was clearly that an atmosphere of restraint was not her long suit. Her actions appeared to be consistently aggressive and had become a topic of wardroom conversation. "Who's driving the problem in Vincennes?" was a question asked on numerous occasions prior to 3 July. "'Robo-cruiser' was the unamusing nickname that someone jokingly came up with her, and it stuck,' Carlson said. My guess was that the crew of the Vincennes felt a need to prove the viability of Aegis (the highly sophisticated anti- aircraft system on the cruiser) in the Persian Gulf, and that they hankered for an opportunity to show their stuff..." The Vincennes crew misidentified the civilian airliner as an Iranian F-14 fighter plane, which does not carry anti-ship weapons, and shot it down with missiles, to the horror of Carlson watching the same plane on radar. "We locked up and illuminated" the approaching Iranian airliner "with our missile fire control radar," Carlson wrote. "The aircraft continued climbing on a southwesterly course that would take it directly over the Vincennes position. Based on closest point of approach to the Sides (range and altitude), lack of any significant known F-14 antisurface warfare capability... lack of detected radar emissions and precedent, I evaluated the track as a non- threat... "The Vincennes announced her intentions to take TN 4131 (Track NUmber 4131, the Iranian airliner) with missiles at 20 miles. I wondered aloud in disbelief, but I did not do the one thing that might have helped. I did not think to push for a re-evaluation of IFF (identification of friend or foe)..."
Yesterday the University of Colorado at Boulder newspaper carried an article entitled "Officials say CU keeps records confidential." The article was in reaction to reports that a murderer obtained his former girlfriend's course schedule by giving a University of Washington clerk the victim's Social Security number and name. The point of the article was to assure students that "that couldn't happen here." In interviews CU officials reviewed employee training on the Family Education Rights and Privacy Law and said that reminders about the law are consistentlly sent out. A course schedule would not be released to anyone other than the student and only if the student had ID. Every one of administrators interviewed was confident that no one else could obtain a student's course schedule. The RISK involved is that no one seems to have thought about our on-line course registration system. This system allows a student to review his or her schedule by phone. And what do you need to access that information? The student's Social Security number and a secret PIN. Unfortunately, the PIN is assigned as the student's birthdate...
>From: David A Honig <honig@BONNIE.ICS.UCI.EDU> [writing in RISKS] >It seems to me that if the specifications are complete and correct then >there should be no problem. It seems to me that much of the grief in our industry, and many of the risks, stem from the assumption that every system has an "a priori" specification, if only you can find the right person to ask. In my experience, specifications change as people understand the implications of what they have asked for. Specifications also change because new ideas come along, or the users change, or the constraints change. Most of all, specifications change because every useful system mirrors the real world in some aspect, and the real world changes. So specifications (in the sense of *real, current needs* ) change throughout the development of any useful system and throughout its service life, too. This is why "lowest compliant, fixed-price bid" is a poor way of purchasing systems.
How about "most trusted bidder with a reasonable price"? The problem with many lowest bidder schemes (particularly those often legislated by well-meaning government bodies) is that they do not provide an adequate mechanism to separate competent bidders from incompetent bidders. What good are detailed specifications if you don't honestly believe that the contractor will be able to meet them, or if you believe you're going to have to pay enormous legal fees to force the contractor to meet them? There's plenty of opportunity for abuse if you let public officials hire their "favorite" contractor regardless of price, but a pure lowest bidder scheme is not the solution, either.
A somewhat jaundiced view of the process from the viewpoint of an employee of several large and small contractors (who shall remain nameless) is that first of all, the going practice seems to be to ask the developers what it will cost to develop a system to the spec provided, then say "no, try again" to realistic estimates until the bid comes down to less than the program office believes the competition will bid. The same, of course, applies to schedule. In short, it's a low-ball contest with the award going to whoever makes the most audacious bid. Next comes the engineering challenge of devising an implementation which can be done in the time and budget which was bid, and which can somehow be construed as meeting the requirement. A specification calling for on-line error diagnostics may be met by having the terminal beep, or some such travesty. Of course, there is the problem of divining what the spec does ask for. I have been at a design review where our customer was a procurement agency who was acquiring a system on behalf of the end-user agency, and there were anguished cries from the end users who didn't even recognize the system the bean counters had specified, let alone what we were proposing to provide. The solution to all the above comes about in engineering change proposals in which the necessary features are finally added at a price which brings the total cost about up to where it should have been bid in the first place. Bob Hirsch, Sterling Software, NASA Ames Research Center, Mail Stop 233-10, Moffett Field, CA 94035 415 694-4807
I think David Honig has pointed out a kind of naivety about the process of quality software that allows government and management to persist procurement through lowest bidder. > It seems to me that if the specifications are complete and correct then > there should be no problem. The problem is that specifications are never complete. The human users and implementors are always finding changes that need to be made. The procurement process must take into account that the spec will evolve over time. The procurement is not only for the product specified, but for the team that builds it, and revises it as it goes along. > And I cannot think of another decent way to buy systems (the > next-to-lowest bidder? the highest bidder?) that will solve the problems. Indeed, this is a hard problem. Here's a suggestion for a procurement process that will do a little more to keep vendors honest: 1. Sealed Bids (same as now) 2. Discard the highest and lowest bids. (Since all bidders are bidding on the same job, if the bids are wildly out of range there's either a problem with the spec or the vendor.) 3. Have those who produced the spec visit the remaining vendors and judge whether the vendor can do what they say they can. 4. Have management choose either the lowest bidder, or the highest quality based on the recommendation made in the previous step. Any procurement process that reduces to a simple mechanical process engenders RISKS. Although there is the possibility for human error in the process I have suggested, it is designed to address the faults of the existing procurement process of "lowest bidder", while minimizing the risk of doing worse than "lowest bidder". -wdc
If, on the one hand, pilots in modern automated planes risk falling asleep from boredom, and on the other hand, they never seem to get enough simulator training in really oddball failures and problems, maybe future cockpits should be designed with simulation software built in, to allow them to practice while they fly. They couldn't be as good, naturally. Real cockpit simulators simulate everything, including the motions of the simulated airplane, but perhaps going a few rounds with something more modest, along the lines of a video game, would be useful nonetheless. And it would certainly keep the pilot awake! You just need to make sure the simulation is never confused with the real thing... Dan Franklin
>However, the article does not objectively cite balancing situations in which >automated systems handled sets of conditions that were too complex and changing >too fast for humans ever to respond satisfactorily. That's probably because such situations don't tend to exist. Automation in airliner cockpits serves to reduce workload, such as when entering congested terminal areas. To insinuate that the pilots *could never* handle such sit- uations is to ignore that single-pilot general aviation IFR operations do it all the time. There may be two exceptions to this generalization: (1) Extending routine automation into critical weather situations, namely 0/0 (Category III) landing. Airliners can now land in weather that most of us wouldn't be caught dead driving in. Think about it...:-) (2) A secondary layer, usually transparent to the pilot, when his airplane itself is so aerodynamically unstable that he could not hope to control it in a consistent manner over time. In this case, the flight-control system "sim- ulates," in-flight, either traditional (as in the case of the 747, which tries to provide the "feel" of a 707, through control-force dampers or amplifiers) or untraditional (A320/330/340) flying characteristics. <> Still, while simulator training could help for some kinds of emergencies, >One wonders, "Why not?" Can the great minds of the aircraft industry not >postulate virtually any kind of emergency? Writing simulators is extremely difficult. The computational ability may not be available to accomodate higher-fidelity simulations, even if a satis- factory method existed to write the billions of lines of code required to rep- resent *every* fault situation. The most cost-effective method, and the current trend, has been to identify those factors which are most likely to occur in a flight (emergency and otherwise), and represent them as realistically as possible. In certain cases, "classic crashes" (such as windshear incidents) may be simulated and pilots run through them. Rolfe's text, _Flight Simulation_, is a nice overviews of the problems and promise of flight simulators, both military and civil. >For years, the US Air Force has been >flying high performance tactical jets using significant amounts of automation. >If the automated system fails, the game is over. The automation is an essential component of the flight control system for these types of aircraft. And if it fails, like you say, the game is over: EJECT. That option does not exist in civil situations. >The entire article is a good example of the warped treatment afforded a >complex, scientific topic by the popular press. The article was largely on-base. If anyone would like to see what other "so- called experts" have to say, I recommend Nagel and Wiener's _Human Factors in Aviation_ (Academic Press: 1988, $65 or so, hardcover).
Please report problems with the web pages to the maintainer