Normal Accidents: Living with High Risk Technologies - Updated Edition
Charles Perrow
I'm struggling to write a fair review of Normal Accidents: Living with High-Risk Technologies. It's what I sometimes call a "I have a new hammer -- look at all these nails!" book. That is, someone has a new idea and writes a book explaining all the ways it applies. These can be very good books, Plagues and Peoples, Private Truths, Public Lies, and The Strategy of Conflict come to mind. Normal Accidents had the potential to be one of these, and if I'm fair, it probably IS one. The problem, though, is that, although it starts off very well, every successive chapter is weaker than the previous one, and the last three are dreadfully tedious. By the time I emerged from the end of the last, I was yelling to Charles Perrow in my mind (as Biden said to Trump), "Would you just shut up, man?" The temptation, given my state of annoyance with the author, is to nitpick the entire book to death, which would not be fair, because it really is based on a very good idea.
That good idea is "normal accident theory" (NAT). It started when Perrow was called to participate in the analysis of the accident at the nuclear power plant at Three Mile Island (TMI). Experts had previously argued that the chance of a serious nuclear reactor accident was very small. Perrow argued that, contrary to being improbable, it was almost inevitable. Not because of the specific combination of things that went wrong at TMI, but because the potential for such accidents was embedded in the nature of the system at a nuclear reactor. He said in 1984, when Normal Accidents was published, that we would have more TMIs. Since then, by my not-well-informed count, we have had worse accidents at Chernobyl and Fukushima, so two points for Perrow.
Chapter 1 is about TMI. Chapter 2 discusses nuclear power plants more generally. Chapter 3 is the heart of the theory the book presents. In it he describes the characteristics of systems that have "system accidents" or "normal accidents" -- complexity and coupling. The next five chapters deal with accidents in different types of systems, 4. petrochemical plants, 5. air travel, 6. marine shipping, 7. dams, mines, 8. space, weapons, and recombinant DNA. At first this was fascinating to a detail-oriented technical type like me. (YMMV, of course.) It was absorbing to read the technical details of how accidents in these different systems came about. They are all different, yet they are all the same, which is Perrow's point. But the examples become weaker as we go along, because we are venturing into systems to which NAT barely applies, or areas where Perrow doesn't know what he's talking about, either because the information is not available to him (nuclear weapons) or he is poorly informed. (Recombinant DNA, "The system appears to be complex in its interactions and tightly coupled, but I caution the reader that I know even less about it than I do about nuclear weapon systems" -- it shows).
Now, I am going to succumb to that temptation to nitpick that I complained of. Feel free to stop reading here. I want to be clear that I am not casting doubt on NAT, but only complaining about the way this book is written. In fact, I'm going (with superhuman restraint -- I hope you appreciate this) to restrict my criticisms to the last three chapters:
9. Living with High-Risk systems. This final chapter contains Perrow's recommendations. He begins by saying, "I have a most modest proposal, but even though modest and, I think, realistic, it is not likely to be followed." This is tantamount to saying he is not going to take his own recommendations seriously. The chapter then devolves into a long, "Why everyone who disagrees with me is wrong" screed. That never goes well.
Afterword. Here he explains that Normal Accidents went out of print, but that Princeton University Press offered him the opportunity to publish a new edition in 1999. The Afterword is mostly an update on normal accident theory. As written, it is a terrible missed opportunity. 1984 was almost 40 years ago, and as you read the first nine chapters you will constantly remark to yourself incidents that have happened since to which NAT might apply. Perrow barely discusses those. Instead, it is clear that he has become the Grand Old Man of an academic field, and he reviews the literature like an academic, not a person who is genuinely interested in understanding accidents. (That's a bit unfair, but it's how it felt to me as I read.)
Postscript: The Y2K problem. The updated edition was published in 1999, when there was a lot of worry about computer systems failing as the date ticked over to 2000. In this postscript, Perrow seems to have recovered his genuine interest in accidents. In fact, the entire chapter has the feel of him chortling in anticipation of the delicious new disasters that will become available to study. As the world now knows, Y2K was almost a nonevent -- there were minor disruptions only. Folks in information technology would like it to be known that this is not because there was no potential for a disaster, but because they worked their tails off to prevent one, successfully. Fair enough!
In summary, Normal Accidents is a good book about an important idea, but it could have been a better book. I'm going to make a recommendation I rarely make: if you read it, read through chapter 8 and skip the final three chapters.
Comments
Post a Comment
Add a comment!