July08-to-trigraphic-final.qxp
11/17/2008
1:45 PM
Page 99
OPINION
COMMENTS REGARDING "ON
THE
O
n the Nature of Science [1] presents many valid
points regarding the epistemology and practice
of science in the 21st century, but there are several important incongruities.
umented observations of other parties are admissible.
Should less reliable observations (not repeatable but documented by others) have the same epistemological status
as repeatable experiments?
Three axioms presupposed by the scientific method are
realism (the existence of objective reality), the existence
of observable natural laws, and the constancy of observable natural law [2]. Rather than depend on provability of
these axioms, science depends on the fact that they have
not been objectively falsified.
The author describes the three-legged stool of scientific
error correction as experimental care, reproducible experiment, and peer review. Like Occam's razor, peer review
is a temporal expedient for judging scientific work at a
given time. Peer review serves the temporal interests of
publishers, funding agencies, and employers. The ultimate arbiter of scientific validity is repeatable experiment
alone.
Occam's razor and related appeals to simplicity are epistemological preferences, not general principles of science.
The general principle of science is that theories (or models) of natural law must be consistent with repeatable
experimental observations. This principle rests upon the
unproven axioms mentioned above. Occam's razor supports, but does not prove, these axioms.
In addition, Occam's razor fails to acknowledge that if
multiple models of natural law make exactly the same
testable predictions, they are equivalent and there is no
need for parsimony to choose one that is preferred. For
example, Newtonian, Hamiltonian, and Lagrangian classical mechanics are equivalent. Which one of the three
would be preferred by Occam's razor? Is this a justification for saying the other two are wrong? Likewise, how
would advocates of simplicity principles arbitrate between
wave and matrix formulations of quantum mechanics?
Defining science broadly as internally consistent and logical models supported by observation removes the requirement of repeatable experiment and broadens the realm of
applicability from questions of natural law to questions of
history and nearly every other question the human mind
can devise. More importantly, it contains the implicit presupposition that observations that are limited in scope,
audience, and duration are equally reliable as experimental observations that are repeatable a large number of
times by any audience that bothers to exercise sufficient
experimental care. One also wonders to what degree doc-
OF
SCIENCE"
The three-legged stool of historical inquiry includes physical evidence (and scientific analysis thereof), documentary evidence, and eyewitness testimony [3]. When considering historical facts of legal significance, one is hard
pressed to find many criminal convictions based only on
scientific evidence without corroborating documentary or
eyewitness testimony. The legal system recognizes that
the availability and quality of evidence degrades with
time. Increasing entropy guarantees that historical questions cannot be repeatedly tested as many times and by as
many independent parties as questions of natural law.
Because models of natural law can be repeatedly tested by
any audience that exercises sufficient experimental care,
the reliability (or even the identity) of sources is less
important in questions of natural law than in historical
matters. Few scientists would object to acceptance of a
new theory or even an anonymous experimental result if it
was repeatable by independent sources. In contrast, the
reliability of sources is critical in the believability of historical theories. For example, if those collecting evidence
at a crime scene are found to be careless or dishonest in
their handling of evidence, the prosecution stands very little chance of gaining a conviction based on the evidence
collected, and the procedure simply cannot be repeated in
a manner that counteracts the unreliability of the original
sources.
OPINION
There are many examples where Occam's razor would
have picked the wrong theory given the available data.
Simplicity principles are useful philosophical preferences
for choosing a more likely theory from among several possibilities that are each consistent with available data.
However, anyone invoking Occam's razor to support a scientific preference should be aware that future experiments
may well falsify the model currently favored by Occam's
razor. One accurate observation of a white crow falsifies
the theory that "all crows are black." Likewise, a single
instance of Occam's razor picking a wrong theory falsifies
the razor as a general principle.
NATURE
The author shifts from a relatively narrow definition of
science as reproducible experiment to "anything that can
be observed" in order to make the case that supernatural
Amy Courtney, PhD
Department of Physics,
United States Military Academy,
West Point, NY 10996
Michael Courtney, PhD
Ballistics Testing Group,
P.O. Box 24, West Point NY 10996
[email protected]
[email protected]
LA PHYSIQUE AU CANADA / Vol. 64, No. 3 ( juillet. à septembre (été) 2008 ) C 99
July08-to-trigraphic-final.qxp
11/17/2008
1:45 PM
Page 100
OPINION
claims can be investigated scientifically. However, this ignores the
presupposition of constancy of natural law that underpins the
demand of experimental repeatability as the ultimate arbiter in science. Since science presupposes the constancy of natural law in the
demand of experimental repeatability, supernatural claims cannot be
falsified scientifically without creating a circular argument. (In
addition, most supernatural claims concern specific historical
events rather than repeatable phenomena.) The objectivity of science rests in experimental repeatability.
Defining science as "observationally constrained model building" is
barely more specific than defining science as "what scientists do."
How far is this from defining sound science as "what scientists say"
(with appropriate homage to peer review)? At this point, is science
really a powerful, objective epistemology for exploring natural law,
or have we merely replaced one set of authorities (the Catholic
Church of the Middle Ages) with another (the scientists of the 21st
century)?
We must not replace experimental repeatability with peer-reviewed
observations as the ultimate arbiter of scientific validity. Only
repeatable experimental results qualify as scientific observations.
Observations of physical and documentary evidence of historical
events do not warrant equal status with repeatable experiments.
Evolution as a model of natural history is widely accepted because
many expectations have been fulfilled by later discoveries of preexisting evidence. (Such later discoveries are better described as
fulfilled expectations than specific predictions akin to Einstein's
prediction of the bending of light by gravitation.) As a scientific theory of ongoing biological processes, evolution would be strengthened by more direct future observations of speciation in higher
organisms via the expected mechanisms of natural selection.
"Predictive power" as an arbiter of science means the ability to predict the future course of systems under study, not merely the ability
to predict future discoveries of pre-existing historical evidence.
Valid models of natural law do not merely predict future discoveries of pre-existing evidence; they predict observations of future
events.
Does failing to make an epistemological distinction between natural history and natural law really serve the interests of science?
Does replacing the demand for predicting results of repeatable
experiment with the much broader "observable support" better prepare today's student to become tomorrow's scientist?
Science needs objective criteria to rank the value of predictions and
observations without the appeals to authority inherent in peer
review or "scientific consensus." Observations that are experimentally repeatable should rank higher than historical observations
whose repeatability is limited by increasing entropy. Specific predictions regarding future events should rank higher than expectations of future discoveries of pre-existing evidence. Thus, the science of natural law is inherently more objective than scientific
descriptions of natural history.
What is the benefit of pretending that science provides the same
high levels of certainty in historical theories of origins (species, universe, solar system) as the more objectively and repeatably testable
quantum electrodynamics and classical mechanics (within their
well-established areas of applicability)?
In conclusion, parsimony and failure to distinguish history from science can conspire to produce unwarranted levels of certainty. For
example, consider an innocent man whose blood is found at a murder scene. In the absence of observable exculpatory evidence, the
Jennings definition of science would seem to demand his conviction, regardless of plausible alternate explanations that might be
offered, since alternate explanations are likely more complex than
the theory that the defendant is guilty.
References:
1.
2.
3.
BK Jennings, “On the Nature of Science”, Physics in Canada,
63(1) 2007.
One might attempt to reformulate a scientific method without presupposing natural law, but the presupposition of natural law is a
common feature of the scientific method in epistemologies from
Bacon to Feynman. Novum Organum, F Bacon, 1620. The
Character of Physical Law, R Feynman, 1965. The Feynman
Lectures on Physics, Vol. I, R Feynman, R Leighton, M Sands,
1963.
M Courtney, A Courtney, Epistemological Distinctions Between
Science and History, http://arxiv.org/ftp/arxiv/papers/0803/
0803.4245.pdf
REPLY TO COMMENTS REGARDING "ON
SCIENCE"
THE
T
he article "On the Nature of Science" presented a simple
but subtle description of the scientific method with wide
applicability. It was meant to be descriptive, not prescriptive or axiomatic 1. It was based on the idea that science is
model building not recreating reality. The idea of model building
eliminates the assumption of realism. The scientific method, as outlined in the article, neither assumes nor implies realism. Model
building, as the fundamental concept behind science, does not rely
on the existence of "natural laws' either but only on the existence of
correlations in the observations which science then attempts to
model.
100 C PHYSICS
IN
CANADA / VOL. 64, NO. 3 ( July-Sept. (Summer) 2008 )
NATURE
OF
B.K. Jennings,
TRIUMF,
4004 Wesbrook Mall,
Vancouver, BC V6T 2A3
July08-to-trigraphic-final.qxp
11/17/2008
1:45 PM
Page 101
OPINION
The scientific method is the building of models that are tested
against observations. The article repeatedly makes the point that in
science it is the observations that are the ultimate authority not peer
review or other possible authorities. The constraining observations
can be used in two different ways. They can either be used to influence the construction of the model or they can be used to test the
model after it has been constructed. This is the difference between
curve fitting (postdiction) and making real predictions. This distinction is crucial. It is relatively easy to describe observations that are
already known. It is much more difficult to predict observations that
are not yet been made or whose results are not yet known to the
model builder. Predictions are the real test of a model.
Epistemologically, it does not matter if the observation refers to a
process that happens in the future or a process that occurred in the
past. The only important distinction is between predictions and
postdictions.
Repeatability is a key aspect of the scientific method, but it is not
without its pitfalls and subtleties. It is easy to repeat an experiment
and, due to systemic errors, make the same mistake over and over
again. Repeatability does not, of itself, guarantee correctness.
Repeatability is subtle in another way. Consider the location of
Jupiter in the night sky. Repeated observations will give different
locations due the daily, yearly and other real or apparent motions of
the planet. A simple interpretation of repeatability fails in this case;
however, since the models predict the variation, the motion can be
studied using model construction and testing against observations.
Repeatability must be interpreted within the context of a model.
Repeatability is just a special case of the more general idea of testing predictions against observations. It is the predictions that must
be able to be repeatedly confirmed not the individual observations
themselves. That observations are repeatable, if indeed they are, is
a prediction of many if not most of the models that we work with in
science. As the example of Jupiter illustrates, it is not absolute.
Within the context of model building and testing against observation, the distinction between "natural law" and "natural history" as
defined in the Comment is neither useful nor significant. In both
cases one constructs models, makes predictions, and tests these predictions against observations, most critically, against observations
that have not influenced the formation of the model. While models
involving history may have less possibility for testing, this is a matter of practice and not principle. Even tests of "natural law" may in
some cases not be possible. For example, if quantum gravitational
effects only have observational consequences near the Planck scale,
then tests of quantum gravity will be practically excluded.
Occam's razor is used to distinguish between equivalent models, i.e.
models that make the same predictions for all observables. This is
the only principle that would eliminate the Omphalas Hypothesis
and its cousin Last Thursdayism. In this regard its application is
ubiquitous; additional assumptions are automatically discarded, frequently without conscious thought. Occam's razor can and does lead
to errors when applied to models that can in principle be distinguished observationally but not with the currently available data.
Even in this case, it can be useful in refining tentative models for
further testing.
Peer review is an important part of the scientific process; however,
it is not an incantation that magically guarantees a result is correct.
Contrary to the implication of the Comment, peer review is not the
setting up of a new high priesthood but rather replaces it. In science
there is no human who serves as the ultimate authority thus when
decisions have to be made some process is required. The decisions
are referred to the peers of the person involved, as these people are
the ones most likely to have the ability to judge correctly. Is this
process perfect? No. Is there a better process? Perhaps, but no one
seems to know what it is.
Peer review plays another very important role unrelated to "the temporal interests of publishers, funding agencies, and employers".
This is as a check on the validity of one's work. Errors are commonplace. Running one's work past experts is simply prudent. They are
in the best position to spot errors and mistakes. Perhaps the authors
of the Comment have never made a mistake but I find this peer
review exceptionally useful. The article "On the Nature of Science"
had much peer review both before and after it was submitted for
publication. Even this reply has been subject to review by my peers
before I sent it to the editor.
The concept of the supernatural is not well defined. The Vikings or
ancient Greeks would have regarded lightening as supernatural,
resulting from the actions of the Gods. What is frequently regarded as supernatural, for example the healing of sickness by anointing,
if it actually occurs would be repeatable and could be studied by
even the restrictive scientific method presented in the Comment. All
observations can be treated on an equal footing with no distinction
made between natural and supernatural. The distinction, if one
wishes to make it, comes purely through the models used to
describe the observations. Is mental illness due to daemons (supernatural) or chemical imbalance (natural)?
"Observationally constrained model building" is much more than
saying science is what scientists do. As previously pointed out, the
idea of model building decouples science from the assumption of
realism. The scientific method outlined in the paper could be used
by a "mind in a vat" and is even consistent with solipsism. In the
both cases it might be sterile, but it could be done. By saying
"observationally constrained" we are putting the emphasis on observation as the final arbiter and not on any human agency.
The Comment says: "One also wonders to what degree documented
observations of other parties are admissible." As a theorist I rely
entirely on "documented observations of other parties". My experimental colleagues would gladly tell you that if I did an experiment,
that would, indeed, not be science. But seriously, modern science
would grind to a halt if "documented observations of other parties"
were excluded or not used.
In conclusion, there are two main differences of opinion between
myself and the authors of the Comment. First they regard the scientific method as of limited applicability whereas I regard it as applicable to most if not all areas of human knowledge. Second they are
committed to realism whereas my view of science is based firmly
on instrumentalism.
1. For a more axiomatic treatment see B.K. Jennings, The Scientific Method,
arXiv:0707.1719v1 [physics.hist-ph]
LA PHYSIQUE AU CANADA / Vol. 64, No. 3 ( juillet. à septembre (été) 2008 ) C 101