> Or, there's the story about the guy who falls asleep in front of his > terminal with an ELIZA program running and his boss logs on and thinks he's > talking to him but is actually talking to the program, and gets pissed off. This may have actually happened. Joseph Weizenbaum (MIT professor, author of _Computer Power and Human Reason_) told the anecdote in a class, with himself as one of the actors. It went something like this -- some of this is doubtless my own memory inventing things. The dialogue is partially courtesy of GNU Emacs' Eliza program, and the rest is made up. Weizenbaum had recently written ELIZA on one of the MIT AI Lab's computers. In those days, computers were rather weak. The computer in question had a time-sharing system on it, yes, but it got rather sluggish when two people were using it at the same time. Weizenbaum left ELIZA running one evening and went home. That evening (around 4 a.m.), another AI Lab person was trying to get his program working for a demonstration to his funding agency the next day, and it wasn't working very well. He was using the computer Weizenbaum was logged on to, and decided that he needed the whole thing. He went to Weizenbaum's office, hoping that he could persuade Weizenbaum to log off. When he got there, Weizenbaum was nowhere to be found, and his terminal was on (and blank). The AIist thought that Weizenbaum was working from home, and had slaved his office terminal to his home one. So, he typed "Joe, please log off." "WHY DO YOU SAY JOE LOG OFF?" "I need the computer for an AI demo tomorrow, Joe" "WHAT MAKES YOU BELIEVE THAT?" "Joe, I'm serious. I've got a demo here tomorrow and I need the computer." "PERHAPS THE PEOPLE YOU HANG AROUND WITH COULD HAVE CAUSED THIS." After a few more exchanges like this, the AIist decided that Joe was being very obnoxious, and called him at home to scream at him. "Joe! You *******! Why are you doing this to me?" Recall that it was four in the morning, and that Weizenbaum had no idea that his creation was running amuck in the AI lab. He quite reasonably replied, "Why am I doing _what_ to you?" This sounded so much like what ELIZA had been saying that it was hard to convince the AIist that it hadn't been Weizenbaum on the terminal.
Although the computer roles are probably insignificant, the scope of the abuses in the savings and loan insolvencies (estimates are approaching $100 billion just in bail-out money) are such that upwards of 20% of the cases are alleged to involve fraud. The incentives seem rather simple -- set up an apparently legitimate S&L, make all sorts of loans to friends, let them all default, and then let the government pick up the pieces for the legitimate investors. Three of the nation's largest CPA firms -- Deloitte Haskins & Sells, Coopers & Lybrand, and Touche Ross & Co., plus smaller firms, have been sued for their roles in failing to detect fraud. Another large firm, Arthur Young, proclaimed Vernon S&L of Dallas clean shortly before federal regulators declared it insolvent -- because 90% of its loans were bad. Whatever the mixture of mismanagement, incompetence, fraud, and other factors turns out to be, the situation seems pervasive. Why were the auditors were out to lunch? Even if the era of decontrol were ended, it seems that a such widespread problem could not be aided by better computerization (knowing what we know about rigging computer systems, it might make fraud even easier!) -- except possibly in providing better on-line data for the auditors that might simplify their task of rectifying computer records with reality. Overall, enormous amounts of money seem to encourage fraud and creative mismanagement. Computer systems designed to withstand misuse by one user will no longer suffice. Separation of duties and the principle of least privilege help a little, but massive collusions may become the order of the day, in which case checks and balances -- even on the auditors -- become critical. Who checks the checkers? As far as who pays, I imagine that because of the S&L incorporation rules there will be no deep pockets other than the taxpayers and S&L customers. So the real culprits will probably go untapped. But recall the advice of Deep Throat: ``Follow the Money.''
In issue 12, Keane Arase details a story of a botched data collection manufacturing package at a large company which I assume to be Procter and Gamble. He reported on staff turnover, bad hiring, insufficient resources, bad design, and a host of other terrible problems. He points out that some of the trouble could be traced to bad management. It sounds to me like all of the trouble was due to bad management. Although large computing projects are often plagued by management problems, such difficulties are by no means unique to the computing business. For example, he points out that his department was made a profit center with profits measured quarterly even though the system wasn't expected to be profitable for two years. Normally under the profit center model, separate centers are supposed to deal with each other as though they were separate businesses, i.e. the client department should be making progress payments or the computer department should have some provision for treating the progressing project as a growing asset. Accounting of multi-year projects is hardly an unknown art, the construction business has been doing that at least since the time of the Pyramids. Finally, problems of under- or mis-specification aren't unique to the computer industry either. In New Haven CT there is (was? it may have been torn down by now) an extremely badly built pre-fab housing project called Oriental Gardens. It had no rain gutters, letting in the rain and snow to cause all sorts of damage. Why? The houses were partially constructed at a factory, then transported and assembled on-site. The factory expected the gutters to be added on-site, the on-site expected them already to be on the houses when they arrived. The message here is that project management is a real problem, but it isn't really a technological problem except where traditional project management techniques fail to handle unique aspects of computer systems. There is a lot of management knowledge to be had for those that want it.
What is Kerberos and why is it needed? In an open network computing environment a workstation cannot be trusted to identify its users correctly to network services. Software on the workstations may not be trustworthly, so being a privileged user on a workstation is not a meaningful test of authenticity. Source network addresses are so easily forged that they are are meaningful either. Passwords sent uncrypted on the network are vulnerable to wiretappers. Kerberos provides an alternative approach whereby a trusted third-party encryption-based authentication service is used to verify users' identities. Much more information is available with the documentation (see below). How to get it: The first public release of the Kerberos Authentication System is ready for retrieval. Initial distribution will be by anonymous FTP; eventually 9-track tapes will be available. To retrieve the distribution, ftp to ATHENA-DIST.MIT.EDU (184.108.40.206), login as anonymous (password whatever you like, usually your username@host), then cd to pub/kerberos. Retrieve README.ftp, it has directions on how to get to the rest of the software. Distribution is split compressed tar files (xxx.Z.aa, xxx.Z.ab, ...). If you would like to retrieve documents separately, you can get them from pub/kerberos/doc (documents) or pub/kerberos/man (manual pages). If you prefer hardcopy of the documentation, send your address and request to "email@example.com". If you would like to be put on the Kerberos e-mail list ("firstname.lastname@example.org"), send your request to "email@example.com". I would like to thank the following people for their assistance in getting Kerberos in shape for release: Andrew Borthwick-Leslie, Bill Bryant, Doug Church, Rob French, Dan Geer, Andrew Greene, Ken Raeburn, Jon Rochlis, Mike Shanzer, Bill Sommerfeld, Jennifer Steiner, Win Treese, Stan Zanarotti. FYI, the copyright notice: Copyright (C) 1989 by the Massachusetts Institute of Technology Export of this software from the United States of America is assumed to require a specific license from the United States Government. It is the responsibility of any person or organization contemplating export to obtain such a license before exporting. WITHIN THAT CONSTRAINT, permission to use, copy, modify, and distribute this software and its documentation for any purpose and without fee is hereby granted, provided that the above copyright notice appear in all copies and that both that copyright notice and this permission notice appear in supporting documentation, and that the name of M.I.T. not be used in advertising or publicity pertaining to distribution of the software without specific, written prior permission. M.I.T. makes no representations about the suitability of this software for any purpose. It is provided "as is" without express or implied warranty. John Kohl, MIT Project Athena/Kerberos Development Team
My friend, Brian Lloyd, and his dad, former California congressman Jim Lloyd, flew their single engine Piper Commanche across the Atlantic from Gander to Shannon to visit the Paris Air Show a few years ago. They firmly believe that small planes with single engines are more reliable than small twin-engine planes, and they decided to demonstrate it. Halfway across the pond, they're making one of their routine hourly position reports with a passing British Air 747. After the formalities, the following conversation ensues: BA pilot: What're ya flying down there, 448 Poppa? Brian: A Piper Commanche. BA pilot: That's a TWIN Commanche, right? Brian: Nope, single. (long pause) BA pilot: You're mad, you're absolutely mad, you know that! One engine?? I've got four! Brian's dad: Well, that's just three more things to go wrong! BA pilot: You've got me there, I've had to shut one down already! As you can see, they lived to tell the tale... --Phil
In the discussion about multi-engine aircraft failures, we've seen a lot of mathematical probability exercises that forget about analyzing the basis assumption about probability theory. That assumption is the *independence* of the events in question. Taking just the two engine example, everyone has been talking about the chance of a single engine failing as p. Thus the chance of an engine failing on a two engine plane is 2p (for small p, as has been pointed out). But then it has been assumed that the chance of the second engine failing is p. That would be true if the engine failures were independent. But this is not the case. A two engine plane flying on one engine is applying more stress and wear to the engine than normal (since it is probably running at close to full design capacity) Thus the chance of this remaining engine failing is more than p. How much more answers the question of whether a two or a three engine plane is safer. The second p is a function of all sorts of mechanical factors that would only be known through a careful study of the design af an individual airplane type and is probably different for every single plane marketted. (The airframe and other critical systems are similarly more likely to fail on a plane that is running without its full complement of engines). Engine failures are also not independent in another way. In a very recent crash, a pilot of a two engine plane got an indicator that one engine was on fire. He turned off an engine. Due to an unknown cause (pilot error, miswiring?) the wrong engine was turned off. On this flight two engines 'failed' even though one was in working order. From an engine designers standpoint, you might say that only one engine failed, but the plane still crashed. It could even be conceivable that a three engine plane after this occurence could get enough thrust from its remaining engine to allow a restart of the engine turned off in error. But the survivability of a three engine plane in this case is not my point. The point is that engine failures are not necessarily independent events when talking about engines on a multiple engine plane.
Please report problems with the web pages to the maintainer