what others say they have observed. In the very beginnings of science, it may have been possible sometimes to dispense with testimony; but very soon every scientific investigation began to be built upon previously ascertained results, and thus to depend upon what others had recorded. In fact, without the corroboration of testimony we should hardly have had much confidence in the existence of physical objects. Sometimes people suffer from hallucinations, that is to say, they think they perceive physical objects, but are not confirmed in this belief by the testimony of others. In such cases, we decide that they are mistaken. It is the similarity between the perceptions of different people in similar situations that makes us feel confident of the external causation of our perceptions; but for this, whatever naïve beliefs we might have had in physical objects would have been dissipated long ago. Thus memory and testimony are essential to science. Nevertheless, each of these is open to criticism by the sceptic. Even if we succeed, more or less, in meeting his criticism, we shall, if we are rational, be left with a less complete confidence in our original beliefs than we had before. Once more, we shall become less cocksure as we become more accurate.
Both memory and testimony lead us into the sphere of psychology. I shall not at this stage discuss either beyond the point at which it is clear that there are genuine philosophical problems to be solved. I shall begin with memory.
Memory is a word which has a variety of meanings. The kind that I am concerned with at the moment is the recollection of past occurrences. This is so notoriously fallible that every experimenter makes a record of the result of his experiment at the earliest possible moment: he considers the inference from written words to past events less likely to be mistaken than the direct beliefs which constitute memory. But some time, though perhaps only a few seconds, must elapse between the observation and the making of the record, unless the record is so fragmentary that memory is needed to interpret it. Thus we do not escape from the need of trusting memory to some degree.
