Liability for malfunction of medical expert systems---why an expert system is like a power saw (abstract only)

摘要

With the pervasive presence of computers has come a variety of legal questions arising from the use of this technology. Although a number of these have been addressed both by commentators and the courts, one area has so far received little attention: liability for misdiagnosis by a computerized expert system. We adopt the view that because of the nature of expert medical systems, the use to which they are put, and the relative positions of the developer, physician, and patient, strict tort liability should adhere to the developer.

An expert system provides for more than the mere storage and retrieval of data, significant though that function has become. Expert systems make "judgments" based on the information supplied to them and according to certain rules that are logic- and probability-based. One thing that distinguishes an expert system from most other software is that the outcome is not necessarily 100% predictable. Because of this and because of the complexity, it is impossible, in some sense, sufficiently to test the program.

In formulating a theory to handle the legal and technical complexities of the subject of computer misdiagnosis, we review existing areas of the law for analogies---product liability for less sophisticated medical equipment or medical and legal malpractice, for example. On the other hand, the qualitatively different role of the expert system calls for a new characterization of the theory and application of liability issues. What we present is a public policy, legal, and technical analysis of the field.

The role played by the expert system may be a fact question that leads to different assessments of liability depending on the degree of reliance. The contribution to the ultimate decision-making in several systems such as MYCIN, INTERNIST, PUFF, and HELP is analyzed with respect to this issue; whether the expert system is used in diagnosis or design of treatment is also considered.

A complementary question to how liability for malfunction in computer diagnosis may be assessed is whether or not the failure to use a computer may result in medical malpractice claims. Whether the physician who fails to use a program to assist in diagnosis is like a tugboat captain of past years who did not have a radio to warn him of an uncoming storm is clearly a question turning on the state of the art.

What happens when less-sophisticated medical equipment malfunctions---the respective responsibilities of manufacturer, dealer, hospital, physician and technicians---is fairly well understood. But when there is an error in the program in an expert system, who bears the responsibility? The general issue of liability for computer malfunctions is still unsettled, with courts applying quite different standards and reaching, unsurprisingly, quite different conclusions.

In general, discussion of liability for computer malfunction in other areas has focused on economic loss, not personal injury, so that standards that have been applied provide little guidance. The problem of balancing protection of patients with the need to encourage research in the determination of liability is compared with similar dilemmas in other areas of medical research.

One central issue has been whether software is a product and thus subject to warranty and, in some cases, strict tort liability or a service and thus exempt. We show that the evolution of expert systems leads to the conclusion that those in general use must be considered to be products. The standard enunciated in a key products liability case, Greenman v. Yuba Power Products, Inc., is: "A manufacturer is strictly liable in tort when an article he places on the market, knowing that it is to be used without inspection for defects, proves to have a defect that causes injury to a human being." Because neither the physician nor the patient is in a position to discover defects in the design of an expert system, the Yuba Power Products test must apply.