One of the central principles upon which intelligent devices seem to operate is that of maintaining internal models of their external environments. How difficult this is, depends upon both the complexity of the model and its method of representation. In particular, it is usually easy when the problem is posed in the classical heuristic search paradigm, and the data structures used to represent static configurations of the puzzle are relatively unproblematic (arrays, lists, and so on). The lack of side-effects reflects the simplicity of the physics which such models embody. This limitation to elementary forms of interaction is not, of course, intrinsic to the heuristic search method; but when more complex models are constructed it becomes less trivial to pursue the consequences of performing an action. This approach is more general than the heuristic search method (but the latter -- when it has sufficient expressive power -- wins at present by its computational advantage). Assertions mentioning several different situations can then be used to describe dynamical laws which move us from one situation to another. But in some ways the resulting sharp separations between states of affairs are an embarrassment. For if we distinguish two situations s1 and s2, then from the fact, if such it be, that a predicate p is true of Si, nothing whatever follows concerning s2. And this is true even when s2 is directly associated with sl. Say s2 results from s1 by the performance of some action: s2 do (a, si) then no matter how remote -- speaking intuitively -- the connection between the property p and the action a, it still does not follow that p is true of s2. If we want it to so follow we must state this explicitly. Now, unfortunately, there are innumerable facts which might remain unchanged when actions are performed. So instead of writing a law of motion' in the form A(s) B(do(a, s)) where A and B are fairly short expressions, we are apparently obliged to list systematically all conceivable facts which are not changed. So that the law looks more like (Ci(s)& Ci(do(a, s))& & C„(do(a, s))&B(do(a, s)) for some very large n. This works for small problems (such as the familiar hungry anthropoid), but these are usually better formalized in the heuristic search paradigm anyway.

To every 3-dimensional scene there correspond as many 2-dimensional pictures as there are possible vantage points for the camera. It is, however, possible to construct pictures for which there is no corresponding scene containing physically -realizable objects. Pictures of such'impossible objects' can be useful in giving insight into the constraints or grammatical rules associated with the'language' of pictures, just as nonsense sentences can be useful in illustrating the rules of other languages. Impossible objects have been used by psychologists (Penrose and Penrose 1958) to create visual illusions which successfully challenge the ability of our perceptual systems to synthesize a 3-dimensional world from 2-dimensional information. The incompatibilities among the various portions of pictures of these objects are a novel way of testing our picture analysis procedures. The purpose of this paper is to demonstrate some possible decision procedures and to test them on pictures of both possible and impossible objects.

In this paper, we develop the algorithm, given in Plotkin (1970), for findingthe least generalization of two clauses, into a theory of inductive generalization.The types of hypothesis which can be formed are very simple. They allhave the form: (x)Px --> Qx.We have been guided by ideas from the philosophy of science, followingBuchanan (1966). There is no search for infallible methods of generatingtrue hypotheses. Instead we define (in terms of first-order predicate calculus)the notions of data and evidence for the data. Next, some formal criteria areset up for a sentence to be a descriptive hypothesis which is a good explanationof the data, given the evidence. We can then look for the best such hypothesis.Machine Intelligence 6

"In this paper we describe work on the recognition by computer of objects viewed by a TV camera. We have written a program which will recognize a range of objects including a cup, a wedge, a hammer, a pencil, and a pair of spectacles. A visual image, represented by a 64.× 64 array of light levels, is first partitioned into connected regions. These regions are chosen to have well-defined edges.Having chosen the regions, the program then computes properties of and relations between regions. Properties include shape as defined by Fourier analysis of the s–ψ equation of the bounding curve. A typical relation between regions is the degree of adjacency.Finally, the program matches the actual relational structure of the regions of the picture with ideal relational structures representing various objects, using a heuristic search procedure, and selects that object whose relational structure best matches the actual picture."In B.Meltzer and D.Michie (Eds.), Machine intelligence 6. New York: Elsevier, 377-396

New York: McGraw-Hill.

Amsterdam: North-Holland.

New York: American Elsevier.

Artificial Intelligence, Vol. 2, pp. 79-116, 1971.

Ph.D. dissertation "Bi-directional and heuristic search in path problems" (Stanford, Computer Science, 1970) summarized in this article in Machine Intelligence 6 (1971).In the uni-directional algorithms, the search proceeds from an initial nodeforward until the goal node is encountered. Problems for which the goal nodeis explicitly known can be searched backward from the goal node. Analgorithm combining both search directions is bi-directional.This method has not seen much use because book-keeping problems werethought to outweigh the possible search reduction. The use of hashingfunctions to partition the search space provides a solution to some of theseimplementation problems. However, a more serious difficulty is involved.To realize significant savings in bi-directional search, the forward andbackward search trees must meet in the 'middle' of the space. The potentialbenefits from this technique motivates this paper's examination of thetheoretical and practical problems in using bi-directional search.

Inference systems Τ and search strategies E for T are distinguished from proof procedures β = (T,E) The completeness of procedures is studied by studying separately the completeness of inference systems and of search strategies. Completeness proofs for resolution systems are obtained by the construction of semantic trees. These systems include minimal α-restricted binary resolution, minimal α-restricted M-clash resolution and maximal pseudo-clash resolution. Certain refinements of hyper-resolution systems with equality axioms are shown to be complete and equivalent to refinements of the pararmodulation method for dealing with equality. The completeness and efficiency of search strategies for theorem-proving problems is studied in sufficient generality to include the case of search strategies for path-search problems in graphs. The notion of theorem-proving problem is defined abstractly so as to be dual to that of and" or tree. Special attention is given to resolution problems and to search strategies which generate simpler before more complex proofs. For efficiency, a proof procedure (T,E) requires an efficient search strategy E as well as an inference system T which admits both simple proofs and relatively few redundant and irrelevant derivations. The theory of efficient proof procedures outlined here is applied to proving the increased efficiency of the usual method for deleting tautologies and subsumed clauses. Counter-examples are exhibited for both the completeness and efficiency of alternative methods for deleting subsumed clauses. The efficiency of resolution procedures is improved by replacing the single operation of resolving a clash by the two operations of generating factors of clauses and of resolving a clash of factors. Several factoring methods are investigated for completeness. Of these the m-factoring method is shown to be always more efficient than the Wos-Robinson method. The University of Edinburgh