Apart from this argument, which may be thought fantastic, there are reasons of detail for being more or less distrustful of memory. It is obvious that no _direct_ confirmation of a belief about a past occurrence is possible, because we cannot make the past recur. We can find confirmation of an indirect kind in the revelations of others and in contemporary records. The latter, as we have seen, involve some degree of memory, but they may involve very little, for instance when a shorthand report of a conversation or speech has been made at the time. But even then, we do not escape wholly from the need of memory extending over a longer stretch of time. Suppose a wholly imaginary conversation were produced for some criminal purpose, we should depend upon the memories of witnesses to establish its fictitious character in a law-court. And all memory which extends over a long period of time is very apt to be mistaken; this is shown by the errors invariably found in autobiographies. Any man who comes across letters which he wrote many years ago can verify the manner in which his memory has falsified past events. For these reasons, the fact that we cannot free ourselves from dependence upon memory in building up knowledge is, _prima facie_, a reason for regarding what passes for knowledge as not quite certain. The whole of this subject of memory will be considered more carefully in later chapters. Testimony raises even more awkward problems. What makes them so awkward is the fact that testimony is involved in building up our knowledge of physics, and that, conversely, physics is required in establishing the trustworthiness of testimony. Moreover, testimony raises all the problems connected with the relation of mind and matter. Some eminent philosophers, _e.g._ Leibniz, have constructed systems according to which there would be no such thing as testimony, and yet have accepted as true many things which cannot be known without it. I do not think philosophy has quite done justice to this problem, but a few words will, I think, show its gravity. For our purposes, we may define testimony as noises heard, or shapes seen, analogous to those which we should make if we wished to convey an assertion, and believed by the hearer or seer to be due to someone else’s desire to convey an assertion. Let us take a concrete instance: I ask a policeman the way, and he says, “Fourth turn to the right, third to the left.” That is to say, I hear these sounds, and perhaps I see what I interpret as his lips moving. I assume that he has a mind more or less like my own, and has uttered these sounds with the same intention as I should have had if I had uttered them, namely to convey information. In ordinary life, all this is not, in any proper sense, an inference; it is a belief which arises in us on the appropriate occasion. But if we are challenged, we have to substitute inference for spontaneous belief, and the more the inference is examined the more shaky it looks.