NRLgate -
Plagiarism by Peer Reviewers
Sections 10 thru 11
This page is part of the NRLgate Web site presenting evidence of
plagiarism among scientific peer reviewers involving 9 different peer review
documents of 4 different journal and conference papers in the fields of
evolutionary computation and machine learning.
This page contains section 10 entitled "Public policy and scientific
policy issues" and section 11 entitled "References."
Go to top of NRlgate Home Page
Go to Abbreviated Table of Contents
Go to Detailed Table of Contents
Go to Complaint Letter
to the Evolutionary Computation journal
Go to top of Previous Page
10. Public policy and scientific policy issues involving the scientific
peer review process and the Naval Research Laboratory
The events discussed herein raise several public policy issues as well as
issues for scientific journals and conferences.
For background information, see chapter
1 and particularly section
1.6, section 1.7, and
section 1.8 on the Naval
Research Laboratory.
10.1. Fragility of the scientific peer review process
Governmental and educational research institutions expend large amounts
of money on scientific research. Evaluation of scientific research usually
requires technical expertise that is not possessed by high-level managers
and policy makers. These managers and policy makers must therefore rely
heavily on measurements of the merits of scientific research that is made
by other scientists.
The peer review process dispenses rewards and punishments to individuals
and research groups in the scientific community that are perceived (correctly)
as being extremely important to their survival, prosperity, and standing
in their fields. Every researcher knows that the outcome of the scientific
peer review process translates into direct benefits in the form of retention
and advancement in his employment; tenure in academic careers; funding for
his own or his group's projects; recognition; and increased ability to hire
assistants.
I assume that no one thinks that the people from the governmental, commercial,
and educational, institutions involved in peer reviewing for the fields
of evolutionary computation and machine learning are fundamentally different
from all other human beings in all other fields of human activity in that
wrongdoing is impossible and inconceivable.
An unsupervised, unaccountable, and secretive process involving perceived
significant temptations, built-in conflicts of interest, operating in a
lawless environment is guaranteed to produce ethical violations.
The question is not "if," but "when," and "where"
and "who."
The scientific peer review process is generally secretive. There is usually
no meaningful supervision. There is generally no mechanism for settling
complaints. Instead, complaints are often handled on a "circle the
wagons" type of "code of silence" that makes police officers
look talkative when it comes to exposing misconduct by fellow officers.
While the police at least rely on the overriding "greater good"
of public safety to rationalize their corruption, it is not clear what overriding
"greater good" motivates scientific researchers to try to "circle
the wagons" and go into as deep state of denial whenever the possiblity
of scientific misconduct is raised.
10.2. Need for established mechanisms for supervision, accountability,
and complaint resolution in the scientific peer review process
Wrongdoers do not usually voluntarily confess to their misconduct, particularly
if it is serious.
The scientific community believes it has high standards in its peer review
process. However, there is no mechanism for ongoing supervision to verify
that these standards are being maintained. More importantly, there is no
mechanism for reaching a decision on a complaint that these standards are
being violated.
Why then is there is no established mechanism for supervision?
Why then is there is no established mechanism for fact-finding and accountability?
Why should the victim of wrongdoing bear the burden of proposing a one-off
dispute resolution procedure as an additional burden in the process of complaining?
Why should the wrongdoers directly involved in the complaint participate
in a voting process that determines whether or not anything is to be done
about their case?
What is needed is a compulsory complaint resolution procedure that does
not depend on the wrongdoer's generous willingness to be held accountable
for his own misconduct.
The procedure must be automatic. The parties in cases of scientific misconduct
will usually be politically active in the scientific hierarchy and totally
enmeshed in the circular backscratching system of rewards and punishments
dispensed by the peer review process.Indeed, such involvment is the precondition
for the ability to commit multiple instances of serious scientific misconduct.
This is simply a way of saying that the scientific community needs the "rule
of law," rather than the "rule of men." What we have now
is a lawless environment in which there is no compulsory jurisdiction over
wrongdoers.
One approach is for a federal law requiring that there be some compulsory
complaint resolution procedure for all federally funded research (both inside
and outside the government). Federal requirements of this type usually become
the model for handling such matters for all areas of the covered field,
even those areas that are not within their direct reach. This procedure
could be implemented as a requirement for arbitration (as exists, for example,
in the securities industry) or by providing access to an administrative
or judicial procedure that is both streamlined and respectful of the rights
of both accuser and accused. Absent such a federal law, scientific organizations,
such as journals and conferences, should adopt a compulsory complaint resolution
procedure that does not depend on the wrongdoer's generous willingness to
be held accountable for his own misconduct.
The compulsory complaint resolution procedure should not involve people
directly involved in the scientific research process. The experience with
scientists attempting to act as judges is consistently poor (as would, no
doubt, an attempt to get people trained in the law to do computer science).
The most common outcome of attempts at self-regulation and self-judgment
is that the well known "circle the wagons" "code of silence"
culture of the scientific community immediately takes control of the fact-finding
process. A less common, but equally bad, result is that there is a lynch
mob approach that tramples over people's rights.
10.3. Reform of the scientific peer review process at journals and conferences
First, the scientific community needs to acknowledge is that reality
is that an unsupervised, unaccountable, and secretive process involving
perceived significant temptations, built-in conflicts of interest, operating
in a lawless environment with a "circle the wagons" culture is
guaranteed to produce ethical violations. The question is not "if,"
but "when," and "where" and "who."
Such violations are, of course, regrettable, but they are a fact of life
whose their existence should not be treated as impossible or inconceivable.
The recognition that a problem exists is the precondition to doing anything
about the problem. Curiously, privately, there are few scientific reseachers
who do not privately acknowledge that there is a problem. Yet, even though
most scientific reseachers would characterize their work as a search for
the truth, they seem singularly incapable of uttering the truth concerning
this forbidden subject.
.
Second, the scientific community should not reward overparticipation
in the peer reviewing process. Everyone involved should recognize that the
secretive peer reviewing process has a special, differential appeal for
potential "control freaks." Often, these would-be Lysenko's and
"sci pols" are gross under-producers in terms of scientific research
who think that massive amounts of peer reviewing and other administrative
activities is a substitute for productive results. Peer reviewing should
be strictly an amateur, not professional, sport. The analogy should be to
jury service (where most states forbid frequent service and volunteering).
In peer reviewing, those who work hard should be generously thanked by the
community; those who work too hard should be sent packing.
Third, there is a simple, easily implementable thing that the scientific
community can easily implement that will greatly diminish the potential
for many of the worst problems in the peer review system: Rigorously and
untiringly enforce diversity in the reviewing process. There is no reason
for any group (whether a university laboratory, a commercial enterprise,
or a government agency) to have multiple people involved in the peer review
process in a circular back-scratching network of inter-locking peer reviewing
activity. The likelihood of problems in peer reviewing (and particularly
the worst ones) can be greatly diminished by the simple expedient of preventing
the concentration of peer reviewing activity in the first place. Rules against
overconcentration must be institutionalized because when this principle
of diversity is not preserved in an institutionalized way, it becomes extremely
easy for people to make exception after exception.
Fourth, the scientific community needs to recognize that it is important
to have "government of laws --- not of men." An important advantage
of established and institutionalized rules is that they are remembered.
Arrangements that are based on casual personal judgments rarely pass the
test of time. Each of the lessons must then be painfully relearned.There
is a widespread exchange of mutual pleasantries among so-called scientific
"colleagues" at periodic get-togetherssuch as conferences. However,
there really are very few scientific researchers who have any possible way
of knowing anything about the true character of their geographically dispersed
scientific "colleagues."
10.4. Reform at the Naval Research Laboratory
The high-level management of a scientific research institution cannot possibly
be conversant with every different scientific field and should recognize
the difficulty of the problem that it faces in evaluating internal performance
of its own research activities.
The high-level management of the Naval Research Laboratory can do several
simple things quickly to deal with the issues raised herein.
First, the management should recognize that its own departments do
not necessarily act in accordance with the best interests of the overall
mission of the Naval Research Laboratory. The management of the Naval Research
Laboratory owes it to the Laboratory, the taxpayers, and the country to
understand the fundamental dynamics by which individual departments in a
large organization adapt their behavior to survive and prosper in their
environment. It is all really very Darwinian. It is very much in the
interest of an individual department within the Naval Research Laboratory
to give the high-level management the impression (or misimpression) that
the particular department is at the cutting edge of research in a particular
field and, more importantly, that the commercial, education, or private
research arenas are all neglecting areas that are potentially important
to the mission of the Naval Research Laboratory. The successfully creation
of this image (in the minds of NRL management) is the "fitness measure"
that determines the survival and prosperity of a department and therefore
is the driving force that dictates the department's behavior.
Second, the management should recognize that scientific peer reviewing
is not a proper governmental function in the first place. It is not clear
that this activity serves any public purpose whatsoever, that it is an appropriate
activity for the government in the first place, and that it is an appropriate
use of taxpayer money. There is certainly no justification for massive governmental
participation in scientific peer reviewing in terms of information acquisition
since of the scientific papers involved herein are unclassified research
that is freely and readily available. There is certainly no scientific journal
or conference that will live or die if it doesn't have governmental reviewers.
Third, if there is any scientific peer reviewing at all by government
agencies, it should be strictly limited in volume. A government agency is
an essentially limitless resource in comparison to other entities in the
academic and scientific community. A good rule is that no scientist should
be reviewing more than about 3 to 5 times the number of papers that he submits
(on an annual basis). The principle is simply that this approximate rate
of reviewing balances the effort by other reviewers in reviewing that researcher's
work. Once a person starts reviewing more than this number of paper, he
moves out of the category of a scientist who is "paying is own way"
in the peer review process and starts to become a "professional"
reviewer. If there is any scientific peer reviewing at all by government
agencies, it should be limited (as should all reviewing) strictly to the
scientist chosen by the journal or conference involved. This appointment
should always be non-delegatable (especially to subordinate employees or
students who may be working in an agency). If there is any scientific peer
reviewing at all, there should never be more than one person from the same
agency involved with the same conference or journal.
Fourth, once a governmental department gets into the business
of inventing technology in-house and promoting its own creations (which
are, by themselves, reasonable activities), there is an additional strong
argument that the department should not be involved in the peer review process
at all.
Fifth, to the extent that governmental agencies are involved in funding
research (which NRL generally does not, while agencies such as ONR, NSF,
and DARPA do), it is particularly inappropriate for their personnel to be
taking the actions (i.e., accepting papers for publication) that provide
a key measurement in deciding the suitability of individual researchers
or research groups to be awarded grants and also deciding the outcome of
grant proposals that are heavily based on published papers. The potential
for abuse becomes magnified when the implied threat of withholding favorable
intra-government reviews for non-governmental scientists' funding proposals
can be leveraged into favorable reviewing decisions by those applicants
for the agency's own papers and unfavorable reviewing decisions by those
applicants for papers on technologies that compete with the agency's in-house
inventions.
Sixth, there is no legitimate governmental reason to support established
ongoing technical conferences. This is particularly true concerning conferences
that charge drastically below-market admissions fees and conferences that
routinely run generate large profits. I believe that there is a good argument
for governmental research agencies to occasionally support the initial start-up
of a new scientific conference in a field that the government deems to be
of some national importance. However, the fields of evolutionary computation
and machine learning are well past this stage The International Conference
in Genetic Algorithms will be in its 12th year in 1997; it produced a profit
of over $30,000 on revenues of about $80,000 at its last conference. Yet
this conference continues to receive Department of the Navy support (nominally
for student travel). However, when a "non-profit" corporation
runs at a large profit, the money is simply moving from government coffers
to the coffers of the private corporation that operates the conference.
Student travel has nothing to do with the situation. In the case of ICGA,
the corporation involved is a memberless and purposeless organization (ISGA)
that appears to be accumulating money that is acquired in the name of student
travel with the objective of deciding, one day, what to do with the money.
If ISGA ever comes up with an idea for a project that merits government
support, it should seek government funding for that clearly identified project
based on the merits of its specific project. The Machine Learning Conference
(MLC) is a successful ongoing conference that routinely draws over 200 attendees,
but charges dramatically below-market fees to its attendees. MLC reportedly
has recently broken-even or shown a small profit at its most recent two
conferences in Tahoe and Italy. After 13 years, there is no possible governmental
reason to subsidize this particular group of conference attendees --- particularly
when the conference charges below-market attendance fees. Certainly, conference
attendees in machine learning are no poorer than conferences attendees in
any other field. It is time to end the perceived dependence of these financially
viable technical conferences on government support.
Seventh, it's time for a performance review of Code 5510's and Code
5514's back-scratching circularity of publication acceptances at conferences
and journals on which people from Code 5514 are heavily represented (and
which are sometimes also subsidized by small grants from NRL or by larger
grants from ONR that are influenced by NRL). This performance audit should
be performed by scientists who have no connection with the NRL-influenced
funding stream.
Eighth, the potentially intimidating practice of using government
employees to directly contact scientific researchers to collect citations
to NRL authors should be stopped. If any agency wishes to compile citation
counts on its own authors (a dubious use of public money to begin with),
it should do by inspecting the readily available public literature in the
field.
Ninth, the scientific community needs to realize that it is in a
continuing involuntary and unseen competition for government
funds with in-house departments within governmentt agencies even though
a particular scientist (e.g. myself) may never have directly participated
in any explicit funding competition. Individual departments within the
Naval Research Laboratory need to give its high-level management the impression
(or misimpression) that they are at the cutting edge of research in a particular
field (an impression that is created by acceptance of research papers by
the outside scientific community). More importantly, individual departments
need to create impression that outside commercial, education, and private
research arenas are neglecting areas that are potentially important to the
mission of the Naval Research Laboratory because this provides the fundamental
reason that they should be internally funded. The absence of publications
in the scientific literature by other researchers and about competing technologies
helps create this impression. The operation of both of these imperatives
can be seen by browsing through the table of contents of the Machine
Learning journal (where the vast majority of genetic algorithms and
evolutionary computation papers are from authors associated with NRL), and
the Machine Learning Conferences (where genetic programming is noticably
absent, over a prolonged period of years, in spite of numerous submissions
by numerous different authors), and the Evolutionary Computation
journal (where all research that is remotely competitive with in-house NRL
work and most research by the people most active in the field has been permanently
frozen out).
11. References
DeJong, Kenneth A., and Spears, William. 1993. On the state of evolutionary
computation. In Forrest, Stephanie (editor). 1993. Proceedings of the
Fifth International Conference on Genetic Algorithms. San Mateo, CA:
Morgan Kaufmann Publishers Inc. Pages 618 - 623.
DeJong, Kenneth A. 1996. Evolutionary computation: Recent developments and
open issues. Proceedings of the First International Conference on Evolutionary
Computation and Its Applications (EvCA 96). Moscow: Russian Academy
of Sciences. Pages 7­p;17.
Grefenstette, John J. 1988a. Credit assignment in genetic learning systems.
In Proceedings of the Seventh National Conference on Artificial Intelligence.
Morgan Kaufmann. Pages 596­p;600.
Grefenstette, John J. 1988b. Credit assignment in rule discovery systems
based on genetic algorithms. Machine Learning. 3(2-3) 225-245.
Grefenstette, John J. 1989. A system for learning control strategies with
genetic algorithms. In Schaffer, J. D. (editor). Proceedings of the Third
International Conference on Genetic Algorithms. San Mateo, CA: Morgan
Kaufmann. Pages 183-190.
Grefenstette, John J. 1991. Lamarckian learning in multi-agent environments.
In Belew, Richard and Booker, Lashon (editors). Proceedings of the Fourth
International Conference on Genetic Algorithms. San Mateo, CA: Morgan
Kaufmann 1991. Pages 303-310.
Grefenstette, John J. 1992. The evolution of strategies for multi-agent
environments. Adaptive Behavior. 1(1) 65-90.
Grefenstette, John J. , Ramsey, Connie L. , and Schultz, Alan C. 1990. Learning
sequential decision rules using simulation models and competition. Machine
Learning. 5(4) 355-381.
Koza, John R. 1989. Hierarchical genetic algorithms operating on populations
of computer programs. In Proceedings of the 11th International Joint
Conference on Artificial Intelligence. San Mateo, CA: Morgan Kaufmann.
Volume I. Pages 768-774.
Koza, John R. 1992. Genetic Programming: On the Programming of Computers
by Means of Natural Selection. Cambridge, MA: The MIT Press.
Koza, John R. 1994. Genetic Programming II: Automatic Discovery of Reusable
Programs. Cambridge, MA: The MIT Press.
Koza, John R. 1994b. Genetic Programming II Videotape: The Next Generation.
Cambridge, MA: The MIT Press.
Koza, John R., and Rice, James P. 1992. Genetic Programming: The Movie.
Cambridge, MA: The MIT Press.
Koza, John R. 1995. "Codes 5510 / 5514." Orange-covered booklet
distributed at the 1995 International Conference on Genetic Algorithms conference
(ICGA-95) in Pittsburgh. Palo Alto, CA: Prodigy Press.
Langley, Pat. 1986. On machine Learning. Machine Learning. 1(1):5-10.
Langley, Pat, Simon, Herbert A., Bradshaw, Gary L., and Zytkow, Jan M. Scientific
Discovery: Computational Explorations of the Creative Process. Cambridge,
MA: The MIT Press, 1987.
Potter, Mitchell A., DeJong, Kenneth A., and Grefenstette, John J. 1995.
A co-evolutionary approach to learning sequential decision rules. In Eshelman,
Larry J. (editor). Proceedings of the Sixth International Conference
on Genetic Algorithms. San Francisco, CA: Morgan Kaufmann Publishers.
Quinlan, J. R. 1986. Induction of decision trees. Machine Learning.
1 (1): 81-106.
Schaffer, J. David. 1985. Multi-objective learning via genetic algorithms.
Proceedings of the Ninth International Joint Conference on Artificial
Intelligence. Los Altos, CA: Morgan Kaufmann. Pages 593-595.
Schultz, Alan C. and Grefenstette, John J. 1990. Improving tactical plans
with genetic algorithms. In Proceedings of IEEE Conference on Tools for
Artificial Intelligence. Los Alamitos, CA: IEEE Computer Society Press.
Pages 328-334.
Spears, William M. 1993. Crossover or mutation? In Whitley, Darrell (editor).
1993. Foundations of Genetic Algorithms 2. San Mateo, CA: Morgan
Kaufmann Publishers Inc. Pages 221 ­p; 237.
Author: John R. Koza
E-Mail: NRLgate@cris.com
Go to top of NRlgate Home Page
Go to Abbreviated Table of Contents
Go to Detailed Table of Contents
Go to Complaint Letter
to the Evolutionary Computation journal
Go to top of Previous Page