Abstract

Certain theories in the philosophy of science concerning how experimental results in science are confirmed as valid and are accepted by the scientific community have been proposed, but these are inadequate for much of today's "big science." One of these is the theory of Ian Hacking that we can justifiably believe in the existence of entities observed in experiments only if they can be interfered with or manipulated. A different approach is that experimental results are deemed reliable and are confirmed by inductive methodologies, such as repetition of the experiment, probabilistic induction, or Bayesian analysis. Furthermore, has been advocated by certain philosophers that scientific knowledge is gained using a foundationalist epistemology. It is assumed that experimental results are justified and validated by the use of known models which closely approximate nature, and various checks on the experiment such as calibration. ;The results of the first year of data from the COBE DMR instruments and the corresponding experimental methodology are used to analyze each of the above theories. It is argued that experiments can be done in astrophysics contrary to Hacking ; therefore, Hacking's criteria for a valid experimental result are not necessary conditions. It is furthermore demonstrated that the COBE DMR experiment will not be repeated as a check on the findings, and that in general scientists do not repeat experiments for confirmation unless the experiment is important enough and the results are in doubt. The COBE DMR results are also found to be incompatible with confirmation by repetitive, probabilistic, and Bayesian inductive approaches, but compatible with accounts that give robustness of data structures as a criterion for the validity of experimental results. Finally, it is shown that, as in the case of the COBE DMR experiment, when the models and calibration are uncertain, and each step is given as a probabilistic limit on errors, that a foundationalist approach which insists upon certainty of knowledge or traceability of justification from beliefs about experimental results to unproblematic base beliefs, or the ability to know that one knows at each step, fails to be adequate