> (Regarding a posting on the Therac-25 radiation therapy accidents, Don > Alvarez writes) ... The problem of testing and validating advanced > hardware is not any way unique to computers. (Then he gives examples > of accidents that might arise from people abusing non-computerized > equipment: trimming hedges with electric lawn mowers and putting portable > radios in the shower). I work in radiation therapy and just finished a lot of research on the Therac accidents, and there are two points I would like to make: First, the Therac accidents were *not* examples of people abusing equipment contrary to instructions, as in the examples Don gives. The accidents happened because the machine included faults in software and, many would argue, additional design errors in the hardware which provided insufficient protection against software faults. It is arguable that the clinics do bear some responsibility also, because they continued to use the machines after they had some evidence that there were problems with the machine --- but faults in the machine were the source of the problem. Second, are computer-controlled devices a *special problem*? Overall, I agree with Don that the problems of testing and validating machinery are broadly similar whether the machinery includes a computer or not. However, we currently have a special problem with computer-controlled devices because industry practices in software development are often much worse than for other kinds of technology. The Therac is a glaring example of this; the physical design of the radiation-producing apparatus was considered superb; the control system, and in particular the software (it is now clear) were very poor, relative to the safety requirements of this application. Therefore, I do not think that articles in the press (or RISKS postings) devoted to this problem are in any way analogous to "blinders"; rather, they are well-deserved attention to a problem that ought to be fixed. In particular, it is very important to understand that people are not picking on the Therac-25 just because the faults involved a computer. This machine was more dangerous than machines with similar functionality that were not computer controlled, even the ones built by the same manufacturer. The particular hazard manifested in the Therac accidents has been well-understood since a similar series of accidents with one of the first (non-computerized) accelerators in 1966. Evers since, this hazard has been adequately handled in most machines by non-programmable hardwired interlocks. It is reasonable to expect that successive product generations that introduce new technologies should represent progress overall. When a new product turns out to be *less* safe than its predecessors, that is newsworthy. - Jonathan Jacky, University of Washington
[Gerald McBoeing-Boeing and the Near-Sighted McCrew?] Path: sq!geac!yunexus!utzoo!utgpu!watmath!clyde!att!pacbell!ames!pasteur! agate!ucbvax!hplabs!hpda!hpcuhb!hpcilzb!bills From: bills@hpcilzb.HP.COM (Bill Standerfer) Newsgroups: rec.aviation [with one typo fixed] Subject: Boeing Sense of Humor? Date: 10 Jan 89 16:37:33 GMT Organization: HP Design Tech Center - Santa Clara, CA I was paging through a recently acquired 727 manual and came across this little gem of wisdom. (GPWS is the ground proximity warning system. It tells the crew when the ground is getting too close for what they're doing.) "Note: the GPWS will not provide a warning if an airplane is flying directly towards a vertical cliff." Gee, thanks. I'll keep that in mind. :-} Bill Standerfer, KG6FQ — hplabs!hpdtc!bills — firstname.lastname@example.org Hewlett Packard Design Technology Center 5301 Stevens Creek Blvd., Santa Clara, CA 95052 — 408-553-3139 Restoration crew chief - B-29A and KC-97L - Castle Air Museum
In reply to Steve Philipson's remarks about aircraft, a friend once pointed out to me that fighter aircraft are designed to a lower safety standard than civilian aircraft, "because if 1 in 1000 crashes due to mechanical problems, that's far less than are lost due to combat" — as a matter of policy, some safety is sacrificed for improved performance. Mr. Philipson also wisely points out that people involved in technology should point out to the public the risks associated with that technology, so that intelligent policy debate can be carried out. Unfortunately, new technology is often sold as "risk-free", when it isn't. Even more unfortunately, new technology often won't be allowed by the public unless a (false) appearance of no risk is maintained — people reject new technologies on the basis of risks, even if larger risks are already accepted in old technologies. (A bizarre case is AIDS in the United States — the number of people who have ever died of AIDS in the U.S. is less than the number who die yearly of motor vehicle accidents, but we don't convene national commissions on motor vehicle accidents!) Dale Worley, Compass, Inc. email@example.com
By now we've seen several cases in which computer-based systems failed because they did not implement features which had been implicit in the physical systems they replaced. Thus, for example, physical mechanisms do a great deal of implicit sanity-checking, inasmuch as ten and ten thousand look much more different when coded as angular velocities than when coded in binary. Computational abstraction is attractive because it is less cumbersome than physical realization, but cumbersomeness is very often a virtue in itself since it assures that important parts of the world will tend to move at manageable speeds. Drawing up balance sheets of risks and benefits of various uses of computer technology is a good and necessary thing. The problem is that we've always benefitted from the implicit virtues of physical objects without ever having to articulate them. The time to make a good, thorough list of these virtues is now, before we've lost them for good.
A few random thoughts: Yes, the time is here when we can no longer believe photographs we see published. This even goes for the bastion of reliability: The National Geographic. At least two of their covers have been digitaly retouched. One was of two pyramids and a camel in the sunset. One of the pyramids was moved over to fit the space requirements of the cover. Another cover was a photo of an old man somewhere in the mid-east, I believe. They liked his face, but also liked the headdress another man was wearing, so they put the other headdress on his head. It looked real. Painters have done this for years. The Mona Lisa was a composite. What is new is the technology for doing it in a supposedly "trusted" medium. But this information is catching on. A friend was in an auto accident. No one was hurt, but damage was done to the car. One of the parties took a Polaroid photo of the scene. The attendant police officer asked to see it. He signed and dated it on the back. Otherwise he said it would be inadmissable as evidence. His signature was there to state that yes, that's the way things looked. As to evidence of computer crime, I believe U.S. Federal rules regard whatever the computer prints out as "best evidence". Scary. The intelligent gun brought to mind a friend who's an Electrical Engineer. An appliance manufacturer came to him to design an intelligent toaster. It has a knob on the front and an LED readout of the "brownness" setting. Unfortunately, all it is is a timer circuit that times how brown the toast should be. The old way of doing things (a bimetal strip) had feedback from the active site. Not so the new. The intelligent toaster with an open heating element will proudly pop up raw bread after 90 seconds. Worse yet, flaming toast could keep being heated until it's supposedly brown enough. On the computerization of hospitals, a friend of mine (who shall obviously remain nameless) was working on software for a hospital. One project was the scheduling of IVs. A typical regimen would be to administer a unit of saline mixed with some drug every six hours. i.e. noon, six p.m., midnight, six a.m., etc. Daylight saving time then happened. Being a good UNIX system, it carried right on: noon, six p.m., midnight (time change) seven a.m., 1 p.m., etc. The hospital was up in arms. They claimed the IVs were an hour late. My friend had to give in. So now between the midnight and six a.m. doses, there may be five or seven hours depending on the time change. The administration wasn't particularly worried about over or under medicating the patients. Doses around 2 a.m. tend to get skipped. Moral: don't leave your money in the bank around the year 2000, and don't check into a hospital around daylight savings time changeover. brent laminack
>Are we approaching the point (or have we reached it already?) where >truth is, for all practical purposes, whatever the computer says it is? >Where what is accepted as truth is easily manipulated by those who are >privileged to have access to the digital keepers of truth? Recently, in an archeological excavation in the middle east, a large stone tablet was unearthed. Scholars determined that it was an ancient audit report, complaining about the use of papyrus scrolls by the scribes. It was clear that such scrolls lacked the evidential integrity of stone and clay tablets. As recently as when I got into data processing, auditors were complaining that punched cards lacked the integrity of ledger cards. I had to work very hard to convince the auditors that the new batch controls were equal to the transaction-by-transaction controls to which they were accustomed. There is a cruel irony to the fact that I am still here to hear them complain about the passing of batch controls and the return to transaction controls. The more things change, the more they stay the same. What goes around, comes around. Those who fail to heed the lessons of history, are doomed to repeat them. The same computers that enable us to manipulate records, also enable us to make so many copies that no one person can alter them all. The same computers that enable us to digitize an analog record (e.g. a photograph), manipulate it, and return it to analog, also enable us to create digital signatures to make any such tampering obvious and the absence of such tampering equally obvious. In the nineteenth century wills and contracts were expected to be hand written. When the typewriter came along, they continued to be hand written for some time for reasons of admissability as evidence. Today, a hand written will is suspicious. Even though digitally signed wills and contracts are orders of magnitude more difficult to forge than typewritten ones, type written documents will like survive, even be preferred, for two more decades. There was a time when the testimony from memory of the elders was preferred to written records. In this context, it is interesting to note that a vanishingly small number of transactions are disowned. Almost none are litigated. A single forgery hardly ever carries the day. Hardly ever is the record of the contract at issue; it is almost always the intent. Written on the list of heresies and other words I try to live by, it says "there is no truth, there are only hypothesies and evidence." In the short run, while we rethink our ideas of evidence yet again, the forgers may have a field day. I am not much worried for the long run. William Hugh Murray, Fellow, Information System Security, Ernst & Whinney 2000 National City Center Cleveland, Ohio 44114 21 Locust Avenue, Suite 2D, New Canaan, Connecticut 06840
In RISKS volume 8 issue 5, Dave Robbins writes: > We have no practical means of verifying the integrity of > such large volumes of information, and are thus left with no choice but > to trust that the electronic records are accurate. On the contrary. Our other choice is to REFUSE to trust the accuracy of the records. If there is a computer record of at $100,000 withdrawal from my savings account, the bank does not have to trust the record. The computer record is circumstantial evidence: it might provide useful insight for further investigation, but it is not to be trusted as conclusive proof. > It is wholly > impractical, for example, for the Social Security Administration's > entire data base (how many hundreds of millions of individual records?) > to be manually audited to verify its accuracy. It would be no less impractical if all that information were on 3x5" cards. When dealing with volumes of information like this, you accept a certain RISK of fraud and error as the norm, and investigate (manually audit) the most egregious cases. You can't blame computers for causing this situation, and I think you'll have to give them credit for helping ameliorate it. Opinions expressed above do not necessarily — Allan Pratt, Atari Corp. reflect those of Atari Corp. or anyone else. ...ames!atari!apratt
The Hacker's Conference episode is just one of many. Readers of USENET last month closely followed attempts by the press to shut down my own moderated newsgroup. As in the CBS case, where you were "guaranteed" that the story would put you in a good light, the reporter who interviewed me acted in a very sympathetic manner. Ha. With most reporters I have encountered in this area, the fact is this: If the reporter decides in advance that you're a wrongdoer, then just about anything is ethical to get the story. In particular, they will pretend to agree with you and indicate that they are writing a favourable story. After all, it's not unethical to lie to criminals to get them to expose themselves, is it? This is general advice, but we must be particularly careful when it comes to public exposure of modern technology. People are predisposed to fear it. People are now predisposed to link hacker with criminal. People are predisposed to link "computer network" with "underground." Watch out for this. If you suspect the slightest bit of prejudice, clam up. Don't trust a word they say — their motives are not yours. The image of technology is very important to RISKS. It controls what technologies people will trust, and how they will trust them.
An anonymous contributor in RISKS-7.82 notes the dangers of computer course registration procedures using touchtone phones. Our university also has the same system and also uses the easily accessible SSN and birthdate as id's and passwords. Those risks are bad enough but I'm more fascinated by the risks produced by the unexpected interaction among new computer technologies. Our university is much more concerned presently with getting computer registration to work right than about security of the system. Last semester, the system's first run, many more students than anticipated had incomplete schedules because the computer, not knowing any better, actually enforced prerequisites that had long been ignored, blocked out the entire three hours scheduled for a lab that everyone knew really only lasted one hours, etc. This meant an astounding number ("astounding" means about 30 times more than the system was designed to handle) of students had to complete their schedules in a two-day period using touchphones and a few scattered terminals to drop and add courses. Of course, most students trying to call the computer got busy signals. Now here's the interaction: not long ago the university also installed a fancy local switch that gave all campus phones, including one in every dorm room, all sorts of fancy features. Not only was automatic redialing available, but also a cute feature that calls you back when the busy line you are calling becomes free. No telling how much of the switch's resources are required for that little goodie. The obvious outcome was that both the computer registration system AND the campus phone network were brought to their knees. Smart students then figured out they were better off calling from off campus even without the auto-redial features, but then the whole community phone system became sluggish. Gary McClelland
A few weeks ago I mentioned the problems that the Irish Department of Health had with the MCAUTO IRELAND installed system. I would be very grateful if anyone reading this with experience of said system in the US would take the trouble to email me their observations.
Please report problems with the web pages to the maintainer