The Auckland harbour bridge is an arched, 8-lane structure, whose inner 4 lanes are employed in a so-called "tidal" system to cope with changing traffic demands. For decades, control has been achieved by a simple system of lane signals above each lane, every 200m or so: a green arrow if the lane was open to traffic, a red cross if it was closed, and a diagonal arrow if the lane was closing ahead. Now some bright spark has obviously decided it's much simpler to indicate that a lane is open by having NO SIGNAL AT ALL above it. Shall we open a RISKs sweepstake on how soon it'll be before a power outage causes an accident? [Fortunately, head-on crashes are prevented by the use of a movable barrier.] Nickee Sanders, Software Engineer, Auckland, New Zealand
In the tradition of Smokey the Bear's campaign for fire safety, the new cartoon figure Dewie the Turtle is being promoted by the Federal Trade Commission to teach kids and their parents of the importance of computer and network security (http://www.ftc.gov/infosecurity). Dewie urges the selection of hard-to-guess passwords, the use of antivirus software and computer firewalls, and other security practices. Do as Dewie says or you'll be sorry. (*San Jose Mercury News*, 25 Sep 2002; NewsScan Daily, 26 September 2002) http://www.siliconvalley.com/mld/siliconvalley/4151919.htm [So, Do We Do as Dewie Says? OK, but that is nowhere nearly enough. But that's just what the recent draft of the President's Critical Infrastructure Protection Board (CIPB) said *each user* should do. Unfortunately, the CIPB's recommended 60 measures totally ignore the reality that most of the computer systems are so lame that those user measures are still seriously inadequate. Are you Dewie-eyed? Not me. The Dewie I'd root for would move faster than a turtle. PGN]
> ... study by the Gartner Group ... not referenced The non-publication of the Gartner/NWC study is a problem, I agree. At least, the audio recording of the conference discussing the outcomes afterwards is available: http://www3.gartner.com/2_events/audioconferences/dph/dph.html. But let's talk about what we know from open sources. The outcome of the study was, in my understanding, that the assumption "give me ten hackers, and I'll bring this nation to its knees" is plainly wrong. The U.S. military (including NSA) probably has more experience in offensive computer network attacks (CNA) than any other government body in the world. CNA have been part of the doctrine of "information operations" since 1998 (see Joint Pub. 3-13, Joint Doctrine for Information Operations, http://www.dtic.mil/doctrine/jel/new_pubs/jp3_13.pdf) and have been used in Kosovo and on other occasions. The "after action reviews" and people from these units I talked to all concluded that it turned out much more difficult than expected. It takes an immense effort in net intelligence (NETINT), technology and human expertise and manpower to really get some serious damage done. Therefore, the government cyber threat estimates in the last months (after some hysteria about "cyberterrorism" after 9/11) have been reduced to a more sober assessment. Though I normally am not in line with him, I totally agree with the conclusion Richard Clarke, the White House's cyber security czar, drew after the Gartner/NWC exercise: "There are terrorist groups that are interested. We now know that al Qaeda was interested. But the real major threat is from the information-warfare brigade or squadron of five or six countries." (quoted after Ariana Eunjung Cha / Jonathan Krim, "White House Officials Debating Rules for Cyberwarfare", Washington Post, 22 August 2002). If you look at the latest National Strategy to Secure Cyberspace, which was released on September 18 (http://www.whitehouse.gov/pcipb), in the chapter on threats and vulnerabilities there is one scenario that lists a number of cyber security/computer risks incidents that have already happened: "Consider the Following Scenario... A terrorist organization announces one morning that they will shut down the Pacific Northwest electrical grid for six hours starting at 4:00PM; they then do so. (...) Other threats follow, and are successfully executed, demonstrating the adversary's capability to attack our critical infrastructure. (...) Imagine the ensuing public panic and chaos." (p. 4) It clearly looks impressive, but: Many of these incidents have not occurred by purpose, but by plain technical failures. This is not really something any cyber attacker can rely on. And the main example for cyber vulnerabilities and risks in the Strategy are the Nimda and Code Red worms. These kinds of "weapons" can really not be used for any directed attack, and they to my knowledge are not at all capable of spreading to SCADA systems that do not rely on MS Outlook. ;-) I have just finished a review of the changes in the U.S. cyber threat discourse before and after 9/11, and one conclusion for me was: "The threat perception can change when the criteria for a threat are changed. The problem here is: There still are no clear criteria even within government organizations for deciding what is an attack and what is not, and some security agencies tend to overstate the real incidents. Until 1998 the Pentagon counted every attempt to establish a telnet connection (which can be compared with a knock on a closed door) as an electronic attack. Another example shows even better how arbitrary some estimates are. When asked by the Department of Justice about the number of computer security cases in 2000, the Air Force Office of Special Investigations (AFOSI) staff counted 14 for the whole Air Force. The Department of Defense overall count for all services, to the surprise of the AFOSI staff, later summed up to some 30 000. The explanation: The other services had counted non-dangerous events like unidentified pings as hacker attacks, while the AFOSI only had considered serious cases. On the vulnerability side of the problem as well, there are still no standard procedures for identifying and estimating the vulnerability of critical infrastructures. These are being developed since June 2000 in the Critical Infrastructure Protection Office's project "Matrix". Slowly, a discussion seems to emerge on the validity of statistics about the numbers, dangers and damages of computer insecurities. Even Richard Power of the Computer Security Institute that conducts the annual Computer Crime Survey for the FBI was quoted with some self-critical words on this problem." (I can send a copy of the full article to anyone interested. It will be published this fall as: The American Cyber-Angst and the Real World - Any Link?, in: in: Robert Latham (ed.): Bytes, Bombs, and Bandwidth, New York: New Press, 2002) Talking about "cyberterrorism": My problem with many of the publications and fears about it is the total focus on vulnerabilities. While you can see tons of quotes from "security professionals" or IT lobbyists on this, you never find any expert on real-world terrorism being asked about it by the media. If you try to think from this angle, the threat becomes much smaller: Terrorists are not used to hacking, and hackers and terrorists are totally different milieus and cultures. Terrorists don't need to hack, because low-tech approaches work perfectly well (I just say "boxcutters"). But even more important: Terrorism is a form of political communication. The terrorist act itself is not the goal, but the message transported by it and the psychological impacts. For this, computer attacks are just not "sexy" enough - you don't get these "great" TV pictures if you bring down a telephone network or a computer in a satellite control center. So, IMHO terrorist will use the nets more and more for organisational and communicational purposes, but not for attacks. So I guess, my main point is: Be aware of the risks related to computer networking, but do not participate in the fearmongering parts of the media and some interested parties on Capitol Hill are doing.
I'm glad that Stephen Fairfax in RISKS-22.23 considers as a "classic example" my rejection in RISKS-22.22 of his claim in RISKS-22,21 that a probabilistic risk assessment (PRA) finds "overwhelming evidence" that arming commercial pilots is an overall plus. I thank him for that characterisation. I myself didn't rate my note so highly. I hope to do better here. Fairfax doesn't buy my criticism of his reasoning by a long margin. It seems worth understanding the issue in detail, for two reasons. First, while the topic of arming commercial pilots is only marginally relevant to Risks (in that computerised control systems may be more vulnerable to bullets than hydromechanical systems), the subject of the appropriate application of PRAs is central. It was discussed in Risks eleven years ago inter alia by Hoffman (RISKS-12.16), Agre (RISKS-12.21, 12.24), Gardner, Seidel (RISKS-12.22), and Kerns (RISKS-21.24). Second, I have seen the type of invalid reasoning, exemplified by (**) below, more than once in discussions of PRAs for particular phenomena. It seems useful to put a refutation in the public record. To the argument. Fairfax correctly notes that I focus on just one assertion of his, namely that (B): there is "overwhelming evidence" that (A): arming commercial pilots would ameliorate hijacking situations. He wishes us to believe (A) with him on the basis of (B). Indeed, were (B) to be true, we would be irrational not to believe (A). How does he wish us to believe (B)? On the basis (C) of assessing the "probabilities of success and failure"; in short, a PRA. Let us look at the form of the argument. First, we have the indisputable premise that (A) follows from (B). Fairfax's argument then continues ostensibly with the form: (*) (C), therefore (B), therefore (A). But in fact it doesn't have this form, as his reply in RISKS-22.23 makes clear. His argument actually has the form: (**) If one were to perform (C), one would find (B). Therefore (A) That Fairfax hasn't actually performed a PRA (C) is made clear by his comments in RISKS-22.23 about how one would go about doing it. Not: how one actually did it; but, rather: how one would go about doing it were one to do so. It would be convenient were (**) to be valid under the supposition (=A7). For then we could achieve our desired results, not by actually doing things to achieve them, but simply by imagining the outcomes were we to do so. Making wine, bringing up children, winning the Olympics, and proving Fermat's Last Theorem would all be so much easier than we had thought. But unfortunately it is not so valid. Fairfax wishes us to believe (A). The reasoning he proposes is (*). He himself believes (A) on other grounds, though, for he does not have the components of (*); he has at most (**). So the grounds he actually has for believing (A) are not the grounds he is proposing that there are for believing (A). C.S. Peirce called this "sham reasoning" . I called it bogus. Reader's choice. So much for the general point. I also doubted that the chain of reasoning (C, therefore B) could be established, even were one to attempt it. I said Fairfax had no data. He disputes that. We have a different classification of data. I think that to perform any kind of probabilistic assessment of the consequences of arming commercial pilots, he needs at least some cases in which commercial pilots have been armed, and as far as we know there aren't any. He claims that all he needs are cases of attempted hijacking. OK, let's take that at face value and see what we get. He does note that the data are "sparse". Let me indicate how sparse. Aviation Safety Network lists just 16 occasions in the 50 years before September 11, 2001 on which aircraft have been lost to hijacking incidents . These are the most damaging hijacking incidents in larger numbers of lives were lost. Others were more or less successfully concluded. The list is not complete. It omits, for example, one hijacking-to-destruction of a US domestic flight (PSA, a BAe 146 near San Luis Obispo, CA on 12 December 1987 by a passenger with a gun). It also omits three suicide/murder incidents by pilots (one Air Maroc, whose date I do not recall, and two recent ones to a Silk Air Boeing 737 in Indonesia and an Egyptair Boeing 767 off the East Coast of the US. Note that the first is supposed, not proven, and the two latter are so considered by the NTSB but not necessarily by other parties to the investigation). So let's double the number to 30. Can these, *probabilistically*, tell us that arming US domestic pilots will help or hinder? Of course not. There are more potential confounding factors than there are incidents, so it is impossible to control for them, except in the one obvious case of the 4 incidents due to Al Qaeda operations. That virtually nothing probabilistically follows from these incidents does not mean that they cannot be analysed. One could go through on a case-by-case basis and propose counterfactuals: what do we think would have happened, had the cockpit crew been armed? Indeed, Fairfax proposes something like this. Additional incidents may become appropriate for such an analysis, say the Air Algerie incident which Fairfax notes. But this is not any kind of probabilistic evaluation, let alone a PRA, as proposed in (C). It is a counterfactual case analysis, the typical analysis used in accident investigation of all sorts, and does not have a role to play in an argument of form (*). Fairfax regrets that I didn't consider his "additional layer of safety" argument. OK, I'll bite. First of all, it is a metaphor. Second, I think it is an inappropriate metaphor to describe what is being proposed. The policy of the FAA and US domestic airlines up to now has been "clean aircraft". That is, no anti-personnel weaponry on board (with certain - unloaded - exceptions). The justification is that, if there is none on board, then none can be used. Arming pilots violates this policy. Far from adding an "additional layer of safety", it peels one off and replaces it with another. Besides, third, I don't think evaluating metaphors, mine above included, is an appropriate way to reason in safety cases. Fourth, what about cases in which pilots themselves are the problem (there have been three, at least, as above, fully ten per cent of what I take to be the total if one is impressed by such argument from tiny numbers)? Even the deployment of weaponry on board by trained enforcement agents has had problems which would not occur were the weaponry not to be present . Finally, readers please note that I have neither said nor implied what my considered position on (A) actually is. As I said above, I don't consider it a theme appropriate to the Risks Forum. Footnotes:  Peirce used the phrase to refer to reasoning to a conclusion to which the proponent is already committed for other reasons. See Haack, Manifesto of a Passionate Moderate, Chicago U.P., 1998, p8ff. I am using it here to characterise a situation in which the reasons one gives for a conclusion are not the reasons one really has, which is the same thing in other words. Haack was more concerned with the case in which a proponent was committed to a conclusion and would not give it up no matter what. I am not suggesting in any way that this is the case here.  http://aviation-safety.net/events/seh.shtml  See Bob Herbert's frightening NYT account of what happened to Dr. Bob Rajcoomar, a retired army major and physician, published in The International Herald Tribune on 24 September, 2002, at http://www.iht.com/articles/71537.htm [Also made *TheNYTimes*. PGN] Peter B. Ladkin, University of Bielefeld, http://www.rvs.uni-bielefeld.de
In analyzing the recent election failures in Florida, it is important to avoid jumping to erroneous conclusions about the role that machines can play in election systems of the highest quality. There are significant differences between information-based election systems and the simplistic electronic-based systems (often called Direct Recording Electronics or DREs) generally offered in the market today. Research on information-based voting systems has been conducted since the 1980's. Little, if any, of this research has been incorporated into the electronic voting systems widely used today. First of all, the vast majority of objections to electronic systems are not directed at fraud, which is actually the biggest weakness of simple DREs. Rather, objections are often directed at issues of reliability and performance. These issues are certainly important to the voting process; however, they can be resolved through proper certification, testing, and training. Such flaws are avoidable and are not problems uniquely associated with voting systems. Remember the butterfly ballot in Palm Beach County, Florida in the 2000 Presidential Election? This example clearly illustrates that even certified paper-based systems are subject to reliability and performance problems. Justifiable indignation, then, should be focused on the absurdly outdated and ineffective election standards and certification process. Ultimately, it is the job of an unbiased standards organization to enforce minimum reliability and performance policies for election systems. An unfortunate consequence of belaboring performance issues is that the thorniest election issues are not examined carefully enough. Those against electronic solutions have concluded, without appropriate supporting evidence, that election systems that use countable paper ballots are most trustworthy. The fallacy of this conclusion is demonstrated by both the facts that are often given to support the paper ballot solution, and by those that are conveniently omitted: 1) As most who witnessed the 2000 US Presidential Election agree, paper ballots created problems. Paper ballots, be they optical scan or punch card, still have to be counted by machines in an election of any reasonable size. This means that the opportunity for election fraud is not eliminated by the use of paper, but only shifted to a different point in the election process. 2) It is often suggested that electronic voting systems get retrofitted with some form of paper ballot output. I call this the $2500 #2 pencil solution. Doesn't an electronic machine retrofitted this way remain just as vulnerable to "catastrophic failure," "malfunction," and "usability problems"? 3) While most people intuitively understand how a collection of voted paper ballots could be supervised procedurally, in reality the process is always far from perfect. Even in what was arguably the most scrutinized election in history — the 2000 US Presidential Election — ballots were lost, damaged, and/or destroyed. We don't know, and never will know, the extent of the damage; nor will we know how much damage was due to accident and how much was due to malice. But it is clear that many voters were disenfranchised. The truth is that paper-based voting systems are "voter verifiable" in that they can help each voter check that his/her choices are recorded properly. But they are not "publicly verifiable" in that they cannot ensure that the final count is an accurate tally of all the voters' choices. Simple DRE electronic voting systems are neither voter verifiable nor publicly verifiable. Our goal should be to create a system that is both, and modern information technology gives us this opportunity. Another common objection raised is the use of "proprietary systems." I wholeheartedly support this objection. One of the basic tenants of a trustworthy election system is that nothing should be secret about the election process except the link between an individual voter and any one specific voted ballot. Actually, I support something stronger than "open source," namely "open protocol," which publishes the underlying voting technology in addition to the software source code. As Rebecca Mercuri recently said on this forum, "democracy is at stake." I agree. But I also fear the recommended paper-based solution. Doctors once prescribed leeches for deathly ill patients. Sometimes the patients got better; sometimes they died. In any case, the state of medical science was not well served by the common wisdom of the time. C. Andrew Neff, Ph.D., Chief Scientist, VoteHere, Inc. Copyright (c) C. Andrew Neff, 2002. All rights reserved.
Dr. Neff makes some interesting points but MISSES the point of the paper ballot solution. Here are the facts. DREs fail because of reliability, performance, and security issues, but these can NOT be resolved ENTIRELY through standards and testing. It is a fact of computer science that no manner of testing or code examination can assure software or system integrity. This was explained by Ken Thompson in his classic speech/paper "Reflections on Trusting Trust" (available in its entirety — at http://www.acm.org/classics/sep95 — it's a must read, especially if you believe Open Source is a viable solution to the voting problem). Neff appears to entirely misunderstand my paper ballot concept. First of all, I have NEVER said that people should go out and spend millions of dollars on expensive paper printers, rather, I have been recommending for years that communities buy simple optical scanning voting systems if they feel they must unload their coffers of the tax dollars they have collected. But the DREs (WITH PRINTERS) can do a better job in preparing the paper ballots, there's no need for blanks prepared in advance, and overvotes and undervotes can be flagged and brought to the attention of the voters. Where I see the computers being used with paper is to provide an ENHANCED voting system. For example, Dr. David Chaum has worked out an amazing system, using cryptography, where the voter can VISUALLY VERIFY that their ballot was cast, the ballot is produced in a form that can not reveal its contents (except through a verifiable process that does not identify the voter), AND the voter can anonymously verify AFTER the election that their ballot was indeed cast as intended. A human-readable physical ballot is ESSENTIAL to the process, not only in Chaum's system, but for any electronic ballot casting and tabulating device, because it is the ONLY WAY that the voter can be assured that their ballot is entered into the count correctly (no manner of recording of electronic data will suffice). But the "paper" (in Chaum's scheme, laminated plastic, but still a physical audit trail) is essential to the process. Once the vendors become willing to admit this is not possible without something the humans can actually SEE, they might finally start implementing viable systems that are truly auditable. BTW, you can read all about Chaum's and my theories in this week's issue of The Economist. Dr. Neff is wrong on two more counts. As it turns out, leeches ARE still used in medicine. They emit a type of substance that can be helpful in certain cases. And actually blood-letting (in modest degrees) also turns out to be an effective treatment for some ailments. (There were some articles on this a few years back, either in Science News or Smithsonian, I forget which, but well documented.) But I think the analogy he made is quite apropos to this discussion — it illustrates a mode of erroneous thinking where older technologies (like paper and leeches) are characterized as inherently bad, in favor of new- fangled (and occasionally widely off-base) solutions. This is consistent with other VoteHere technology choices — only a few years ago, their president, Jim Adler, was pushing Internet voting. At a debate sponsored by George Washington University in January 2001 — the GWU report (available at www.democracyonline.org) states that — Adler's team at Votehere.net "includes scientists who claim they have already solved many of the hardest problems associated with Internet voting, namely the security, privacy and auditing challenges. For example, addressing the question about audit trails, Mr. Adler said that Votehere.net has designed a system where votes are "burned onto a cd-rom"." Now that's real security for you. Thankfully, the NSF decided that Internet voting isn't a good idea, or the VoteHere scientists might have sold some of their secure systems to Florida. Even Bruce Schneier thinks Internet voting is implausible, and he does know a thing or two about crypto. I could go on further, but my thoughts are embodied in papers on my website (at www.notablesoftware.com/evote.html). I commend Dr. Neff on his initiative in engaging in this debate. I hope that he might also re-examine the immutable facts of computer science and perhaps he can eventually convince his team of scientists to develop voting systems that are truly verifiable, auditable, and secure. In the meanwhile, I have a few leeches for sale. Rebecca T. Mercuri, Ph.D., Professor of Computer Science, Bryn Mawr College
Please report problems with the web pages to the maintainer