Review of “Artificial Intelligence: A General Survey” (1993)

http://www.chilton-computing.org.uk/inf/literature/reports/l...

> 1. Much work in AI has the ``look ma, no hands'' disease. Someone programs a computer to do something no computer has done before and writes a paper pointing out that the computer did it. The paper is not directed to the identification and study of intellectual mechanisms and often contains no coherent account of how the program works at all.

(It seems this was written in about 1973, published online in 1993. I infer this because the article being reviewed appeared in 1973, and this one speaks of 1978 as being in the future).

It split AI research into the categories of; "Advanced Automation", (category A), for any work whose goal was to produce useful behaviour which would be considered intelligent if seen performed by a human; "Computer-based [Central Nervous System] research", (category C), for any work that wanted to further the understanding of the natural central nervous system through study upon models of artificial neurons. A third category, "Building Robots",(category B) was placed theoretically between the other two categories. Category B contained all work which aimed to make progress towards the goal of the creation of an artefact that exhibited behaviour from category A through methods developed by work in category C, it was therefore also referred to as the "Bridge category".

The tone of the Lighthill report was more cautious than much related literature, often referring directly to admissions of naivety and subsequent disappointment from distinguished researchers. Categories A and C were therefore hailed as promising endeavours within the contexts that individual projects in category A remained within a tightly defined context, and work in category C remained tightly aligned to merely computationally supporting work that would have otherwise been performed in psychology and neurobiology. The tone was largely critical, though not unfair, of category B which was unable to unite categories A and C into a single research field by citing successful projects that drew heavily on both categories.

The central critique of category B was that no projects existed that were both, (i) sufficiently related to the other two categories and, (ii) successful. This critique can be interpreted as an example of the "No True Scotsman" informal fallacy in that any successful project suggested as a counterexample could be immediately claimed as being not sufficiently in either category A or C, and any project that drew significantly from both categories A and C when suggested as a counterexample could be claimed to be insufficiently successful.

This can be seen in the report as any projects that are inarguably successful, are inevitably described as using "general computational theory" which, given that the roots of computational theory lay in Turing's work in decidability and intelligence is not obviously distinct from category A research. Similarly, Winograd's SHRDLU project which is admitted to draw deeply on work from both other categories was claimed to be essentially not successful enough as "suggestions for possible developments in other areas that can properly be inferred from the studies are rather discouraging". Lighthill also noted that "one swallow does not make a summer", though this was a "banal" observation by his own admission.

The Lighthill report had a significantly negative effect on AI research funding, though it was not entirely unfair. It was broadly compatible with the view of Dreyfus and potentially saved the reputation of AI from a more disastrous period which may have occurred in the future had overconfidence and overpromising been allowed to continue unchecked.

Inspecting page source confirms it.