Page 29 - Deep Learning
P. 29

12                         Introduction

            any one earthquake. Similarly, meteorologists cannot now and have no hope
            of ever being able to predict the local weather with precision over long periods
            of time. The weather is not the kind of system that allows such predictions.
            Abrupt and unpredictable changes might characterize the weather in the long
            run – the climate – as well. 27
               We all know that we cannot rely on weather forecasts, but in the past we
            put this down to insufficient resources or perhaps to the competence of the
            forecasters. Among scientists, lack of predictability was blamed on insufficient
            information, the probabilistic character of the relevant system, the practical
            impossibility of collecting the relevant data, or, as a last stand, flaws in the
            theory used to generate the predictions. All material systems were assumed to
            be predictable in principle. But the complex systems revolution teaches us that
            lack of predictability is a real and central feature of the world. Evolutionary
            biology cannot predict which species will evolve next, or even how a given
            species will change in the next period of time. Evolutionary biology is nev-
            ertheless one of the most successful of sciences. The response to the lack of
            predictability is to give up predictability as a defining feature of science, not to
            give up the claim to be a science.
               The shift to acknowledging the complex nature of most material systems
            should not be taken to deny the successes of clockwork science. The lawful
            dampening of the swings of a pendulum is not illusory. clockwork science
            works  well  in  tight  contexts.  The  classical  strategy  of  clockwork  science  to
            define the system under study in such a way that it is nearly de-coupled from
            its  environment  and  hence  relatively  safe  from  externalities  is  perhaps  the
            main source of tight contexts. A chemical reaction vessel that is tightly sealed
            to prevent any impurities from entering the reaction is an iconic instance.
            tight contexts are local in space and last for a brief period of time, where these
            boundaries should be understood at the scale appropriate for the system under
            study. Allow externalities and extend the context in space and time and even
            the most well-behaved material system will become unpredictable: How does
            a pendulum swing when its rope begins to rot?
               complex systems have always attracted attention, but they were regarded
            by scientists as recalcitrant cases that would yield to a clockwork analysis even-
                 28
            tually.  In the last two decades of the 20th century, the complex system was
            promoted from fringe exception to central case.  The situation is the oppo-
                                                    29
            site of what the natural sciences taught for 300 years: The systems that fit the
            clockwork model are special cases, unusual and rare. Even in those cases, the
            clockwork model is an approximation that holds only under special condi-
            tions, epitomized by the frictionless, inclined planes of high school physics.
   24   25   26   27   28   29   30   31   32   33   34