15.01.2013 Views

Causal risk models of air transport - NLR-ATSI

Causal risk models of air transport - NLR-ATSI

Causal risk models of air transport - NLR-ATSI

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Notice also that P(A|C) does not necessarily imply that A is the effect and C is the cause.<br />

Take the example A = car does not start, C = empty fuel tank. Both P(A|C) and P(C|A)<br />

exist. In order to be able to differentiate between cause and effect there must be additional<br />

information on the direction <strong>of</strong> the relation between C and A. In a causal model this<br />

direction is indicated by a directed edge (see section 3.7).<br />

3.3. Causation to predict the future<br />

One <strong>of</strong> the main reasons why we are interested in causation is because it allows us to<br />

predict system behaviour if we assume that the past and present determine the future.<br />

Among the first scientists to seriously address this issue were the German philosopher<br />

Immanuel Kant (1724-1804) and the French mathematician Pierre Simon Laplace (1749-<br />

1827). Kant stated that every effect has a cause which is a prior event and that this cause<br />

effect relation is ‘fixed’. Therefore, if we have observed, in the past, that certain causes<br />

have certain effects we can assume that the same causes will have the same effects in the<br />

future [Kant 1781]. Laplace went even further when he published his theory <strong>of</strong> scientific<br />

determinism. He wrote [Laplace 1814]:<br />

"We may regard the present state <strong>of</strong> the universe as the effect <strong>of</strong> its past and the cause<br />

<strong>of</strong> its future. An intellect which at any given moment knew all <strong>of</strong> the forces that animate<br />

nature and the mutual positions <strong>of</strong> the beings that compose it, if this intellect were vast<br />

enough to submit the data to analysis, could condense into a single formula the<br />

movement <strong>of</strong> the greatest bodies <strong>of</strong> the universe and that <strong>of</strong> the lightest atom; for such<br />

an intellect nothing could be uncertain and the future just like the past would be<br />

present before its eyes."<br />

According to Laplace’s conception <strong>of</strong> nature, all laws in nature are deterministic, and<br />

randomness surfaces merely due to our ignorance <strong>of</strong> the underlying boundary conditions<br />

[Pearl 2000b]. Werner Heisenberg‘s (1901-1976) uncertainty principle shows that Laplace's<br />

vision, <strong>of</strong> a complete prediction <strong>of</strong> the future, cannot be realised (although Albert Einstein<br />

(1879-1955) did not agree with Heisenberg, because, according to Einstein, “God does not<br />

play dice with the universe”). According to Heisenberg’s principle, effects do not follow<br />

causes in a rigorous chain <strong>of</strong> events [Feynman 1965]. Nevertheless, we assume that to a<br />

very good approximation the laws <strong>of</strong> science behave deterministically. The problem then<br />

becomes one <strong>of</strong> complexity, there are simply too many variables (or in Laplace’s words,<br />

too many forces and beings that compose nature) to encompass. Because <strong>of</strong> this, all theories<br />

and <strong>models</strong> are approximations and we use probabilities to express the resulting<br />

uncertainty 12 . Apparent stochastic behaviour is then introduced in our deterministic view <strong>of</strong><br />

the world as a result <strong>of</strong> our approximations. Each approximation introduces an error term<br />

with respect to the ‘true’ value. The error propagates when numerical operations with other<br />

uncertain quantities are being conducted. Under some conditions the error term may grow<br />

disproportionately fast. Prigogine [1977] introduced the concept <strong>of</strong> bifurcation points.<br />

According to this theory a system which has bifurcations will imply both deterministic and<br />

probabilistic elements. In between two bifurcation points the system behaves<br />

deterministically, but in the neighbourhood <strong>of</strong> the bifurcation points fluctuations play an<br />

essential role and determine the ‘branch’ that the system will follow. The outcome <strong>of</strong> the<br />

model will then be increasingly uncertain when more cause effect relations are called upon,<br />

typically when predictions further into the future are made. Rapid error accumulation may<br />

12<br />

De Finetti (1906-1985) stated that probability is an expression <strong>of</strong> the observer’s view <strong>of</strong><br />

the world and has no existence <strong>of</strong> its own – probability does not exist (quoted by Lindley<br />

[1986]).<br />

27

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!