2. GEOLOGÍA REGIONAL
2.5. Evolución Tectónica Regional
As we have seen, th e six stages of induction can be split in to two groups: • E xternal stages entailing experim entation and observation
• Internal stages entailing prediction, evaluation, revision and selection
E xpenm entation involves the m anipulation of real w orld scenarios through the design and construction of controlled experiments in a suitable environm ent. Observation^ though somewhat m ore passive, is also involved in interacting w ith the external world and serves as an en try point for evidence into the system. Any interaction w ith real-world scenarios m u st adm it the inherent uncertainty (in a b road sense) th a t pervades experim entation and observation. Hon [42, 43], locates four sources of error: laying down the theoretical framework of the experim ent; constructing th e ap p aratus smd m aking it work; tsdcing observations or readings; and processing the recorded d a ta and interpreting them . We have already noted some of these in the previous chapter. Problem s w ith the theoretical framework involve background theory ra th e r th a n th e object dom ain theory. We can reduce these kinds of error from the broader classification o f uncertainty as experimental
error and observational error to more definite dimensions which we examine below.
6 .2 .1 R e lia b ility
Evidence m ust be observed (or perceived in some w ay), and it m ust be observed (or perceived) by a source. Reliability is a m easure of th e ability of a p articular source of evidence as th e in terp reter and recorder o f th a t evidence to provide correct observation statem ents. Any source of observations, be it a h um an observer or a m achine, is nec essarily im perfect, resulting in some degree of variation in th e reliability of th a t source. Reliability here is used in a very specific sense, referring only to the ability of a nam ed
source in a particu lar situation to provide accurate observation statem ents of th e phe nom enon under investigation. The reliability of a source is in question for m any reasons at different levels. A hum an observer, for exam ple, m ay have poor eyesight which could result in th e introduction of doubt into observations. In th e case of a machine observer, analogous dif&culties arise w ith considerations of th e equipm ent used, such as th e quality of lenses, and so on. The raw d a ta of observation m u st also be interpreted and recorded to produce comprehensible evidence, and this conversion firom sense perceptions to ob servation statem ents is another process susceptible to unreliability. Reliability is thus related to th e objective abilities of a particu lar source independent of the statu s of the observed evidence.
6 .2 .2 T r u stw o r th in e ss
Trustworthiness can be defined as a m easure of malicious in ten t on the p a rt of the in form ation source. A source of inform ation, even if it is absolutely reliable in term s of its ability to observe, m ight still provide uncertain evidence because of an intention to mislead. An independent agent acting as observer an d providing input to a system in the form of observation statem ents has its own gocds an d m otivations which do not nec essarily correspond to those of the reasoner. G arigliano et al. [26], for example, use the case of buying a second-hand motor-bike to illu strate this. A used-bike dealer has a definite financial advantage in selling a bike and will therefore present it as favourably as possible. He m ight try to sell it w ithout taking on any responsibilities b u t alternatively, if he is concerned for his reputation, he m ight give a guarantee. The evidence th a t he provides in attem p tin g to sell a bike thus comes into question given his aim of getting a good deal. Buyers would only accept his evidence if confident th a t the dealer does not intend to cheat them . The problem of judging the dealer’s trustworthiness is independent of problem s of his reliability, since one m ight believe th a t he is fully aware of th e true state of a p articular bike, b u t is not inclined to share th a t knowledge. This is related to and also provides a way of dealing w ith the notion of bluff [67] which is prevalent in domains of game-playing and m ilitary strategy where agents also intentionally mislead others. Trustw orthiness is thus an evaluation of th e m otives of an observing agent which are not necessarily th e same as one’s own.
6 .2 .3 A c c u r a c y
Accuracy is a m easure of the uncertainty arising th rou gh the evidence itself or in the m ethod of obtaining th a t evidence, independent of any problem s associated w ith an ob server. In addition to the difficulties arising from th e abilities and the n atu re of the observer, we can consider the phenom enon itself under investigation through experimen ta tio n in some form. Direct sense-perception in m any instances is inadequate, especially in scientific dom ains, and phenom ena m ust be observed an d quantities m easured using a variety o f instrum ents and devices of varying accuracy. Microscopes, telescopes, rulers and other such tools are commonplace and provide th e basis for observation, yet they all have lim its to their accuracy. A microscope for exam ple, has a finite degree of m ag nification beyond which things cannot be observed, and it also adm its the possibility of inaccuracy due to faults in th e lenses and so on. This kind of error occurs a t a different conceptual level to th a t described above to illustrate th e reliability of a m achine observer. Here, th e lens contributes to the accuracy of th e d a ta itself. As an alternative example, consider standing a t a bus-stop and looking a t the front of a bus as it approaches to see w hat num ber it is. W hile the observer m ay be perfectly reliable and trustw orthy, the num ber of th e bus m ay be partially obscured so th a t only the top p a rt of it is showing. It then becomes very difficult to distinguish betw een a num ber 19 or a num ber 18 or a num ber 1 0, for example. Here again, the evidence th a t is available is itself insufficient. Accuracy is the traditio nal dimension of uncertainty th a t is considered in evaluating evidence, and is m ost usually addressed by the provision of simple error tolerances.
6 .2 .4 C r e d ib ility
Credibility is a m easure of the uncertainty th a t arises th ro ug h conflict w ith established constraints and prior beliefs. A lthough th e observer m ay be reliable and trustw orthy, and the d a ta accurate, this does not deal w ith the case of exceptional circum stances when the evidence violates norm al constraints. These violations include problems caused by such things as hallucinations, m irages, optical illusions and so on. W hat is needed is a m easure of th e degree of credibility associated w ith such observations. For exam ple, the degree of credibility associated w ith the observation th a t Joh n is flying around his office is very low (since this is more likely to be an hallucination). A lternatively, when watching a m agician, th e observation th a t a lady has been sawn in h alf is also no t very credible. To
some extent, credibility is dependent on the expectations th a t are derived &om existing theories, and acts as a form of conservatism. The degree of credibility will also depend on the stren gth w ith which these theories are held. Strongly held theories which conflict w ith observations will produce low credibility, while weak conflicting theories (including the current theory) will not affect credibility as greatly. In this respect, it reflects concerns w ith th e theoretical firamework noted earlier. M easures of credibility thus reduce the value of evidence th a t is highly suspect through tricks of nature.
6 .2 .5 S u m m a r y
Simple num erical measures of accuracy alone provide no inform ation as to the circum- stcinces in which error has originated. Our concern is w ith th e acquisition of knowledge. As such, it is im po rtant to locate sources of error so th a t confidence in those sources in future situations m ay be adjusted in th e light of the results of th e current reasoning process, as well as coping w ith the uncertainty th a t currently exists. This contrasts w ith the use of simple error tolerances which assume a G aussian distribution of results, an assum ption which m ay not be justified, and which provides no useful inform ation about the uncertainty itself.
Because of the different kinds of error and un certainty th a t can arise, evidence cannot simply be accepted and used to refute or support a theory. An evaluation of observa tions received as evidence is necessary in order th a t th e character of the evidence can be assessed and accepted or rejected as the situation dem ands. The following section addresses the issue of how to use these four dimensions to determ ine w hether or not a theory or hypothesis is refuted by th e observations.