taylor@hplabsc.UUCP (08/01/86)
This article is from nike!harvard!wanginst!infinet!rhorn (Rob Horn) and was received on Fri Aug 1 01:23:05 1986 The recent story about a Macintosh ``testifying'' included considerable mis-understandings by the reporter. The other cases described involve similar misunderstandings. What was actually going on in court is an equally interesting subject. The courts are using their extablished approaches for handling ``experts'' to accomodate the introduction of computers. I believe that the jurors in that case were probably reminded repeatedly that they were listening to the testimony of the human expert. The Macintosh was testifying in the same way that a blackboard or flip chart testifies. It is used by the expert to present his testimony. It is evidence, but the credibility rests with the expert. I think the court would have no problem if it were revealed that everything shown had been made up by the expert rather than the result of the analysis programs. What matters is the credibility and skill of the expert. (Which would suffer greatly if it could be shown that the presentation was a phoney.) This is in fact part of a sophisticated approach to dealing with the fallibility of computers. The law has long had to deal with experts. Experts have various problems: - They may distort the evidence - They may present only partial views - They may have misunderstandings - They may overlook key information - They are selected with presentation and showmanship in mind The law has grown a variety of mechanisms to deal with these human failings, and they are trying to fit their existing mechanisms to computers. After all, human experts are at least as innovative as programmers in finding ways to make mistakes. If you can somehow make the computer like a human expert, then you can use all your existing approaches. I had to use a similar approach to satify the legal requirements for automatic processing of data from toxic waste analyses. At every point in the collection and processing of data, *paper* records had to be kept that would allow a human to independently verify any item in the database. Where this was impossible (for example with mass spectrometer processing) you had to come as close as possible, plus provide regular calibration records that were supervised by both operators and Q/C personnel. All through this there was constant use of dual person dated signatures, in ink, on paper. For example, instrument outputs were both recorded in the database and printed on a local strip printer so that they could be verified and signed on the spot. Finally about 2% of the samples were manually verified by Q/C including .5% re-verified from the original sample material. They still got most of the benefits of automation, since only a small percentage of the data was ever really processed by hand while retaining enough information that they could invoke human expertise whenever necessary in court. You learn quite a bit about non-erasable paper, bound notebooks, tracability requirements, etc. when working with hazardous wastes. Little things like just what it means to sign that slip of paper. You also get some truly superior quality paper and work notebooks. These notebooks will withstand both Coke and hot coffee without smearing ink or tearing. Any tampering with this material is both very difficult and leaves obvious marks. It will be a long time before computers reach anything close to this level of tracability. Rob Horn UUCP: ...{decvax, seismo!harvard}!wanginst!infinet!rhorn Snail: Infinet, 40 High St., North Andover, MA