IEEE Technology and Society Magazine - March 2018 - 45

when a robot scientist is said to observe some aspect of
the material world.
A system designed to record data must be programmed within a set of parameters, and these parameters need to be established and agreed upon by a group
to be validated as correct indicators of the phenomena
of interest. For example, a case-based learning algorithm designed to perform class discrimination is normally first programmed by a group (humans) who have
decided which objects should be placed in to each classification, based on examples taken from previous experience of those objects. The objects that seem to
generate the most agreement as to the correct class in
which they belong can be labelled "exemplars" of that
classification. Thomas Kuhn's The Structure of Scientific Revolutions applied the theory of exemplars to scientific knowledge claims, arguing that major advances in
scientific fields often amount to a recognition that entities previously thought to be the same actually contain
important differences [34].
The language of observation, therefore, is a means by
which scientists can make distinctions between what is
known and what is not known. There are two important
implications of this argument for the use of robot scientists. First, the strength of any knowledge claim is linked
to how far observations are agreed upon and shared
among authoritative members of a knowledge community. The observations made by the robot scientist were
embedded within a well-established and respected
research program that employed skilled and knowledgeable scientists. It was those (human) scientists who were
recognizably authoritative and credible in their academic
discipline. It was mainly because of that credibility that
the research team was able to publish the AI-derived
data as a contribution to knowledge in their field.
Arguably then, even if robot scientists can be judged
to observe phenomena, the validity of the observations
takes shape when those observations are discussed
and defined through interaction between a social group.
Indeed, this process is the very definition of the second
concept of interest in this section: "the process... in
which individual responses are taken up in to patterns
of social interaction ... interpretation" [34, p. 17]. Can a
robot scientist be judged to interpret results according
to this definition? Here the sociological perspective
becomes useful.
Clearly, robot scientists are becoming acceptable as
tools for contributing to scientific knowledge among elements of the academic community. The observations
made by robot scientists are being taken up into patterns of social interaction via the human designers and
operators of those systems, and by the acceptance of
AI-derived methods by the wider academic community.
Therefore the sociological investigations of robot
MARCH 2018

∕

scientists and their human designers offer a rich seam
of possible empirical research: questions that speak to
fundamental issues about the role of computer programs for scientific knowledge making in contemporary
social life.

Formalization and Reproducibility
One important social and technological aspect of scientific discovery where a robot scientist can offer benefit is
reporting and documenting experimental results. This is
of particular importance in the context of "the reproducibility crisis": "The ability to reproduce experiments is
at the heart of science, yet failure to do so is a routine
part of research" [35]; "More that 70% of researchers
have tried and failed to reproduce another scientist's
experiments, and more than half have failed to reproduce their own experiments" [36]. There are many reasons for the non-reproducibility of experimental results:
the complexity of experimental and statistical methods,
"poor experimental design," the non-availability of raw
data, methods, and code, etc. [35], [36]. The use of natural languages to document experimental procedures
makes this worse. For example, consider this routine
natural language instruction: "inoculate 4 mL of liquid
YPAD and incubate with shaking overnight at 30 C." This
is ambiguous. It does not specify the speed and mode of
the shaking - is it 200 rpm, 600 rpm, and orbital or
reciprocal? What is the duration of the incubation - is it
8 h, 12 h, or does it not matter? Such information is vital
for the reproducibility. The reproducibility of experimental results can be improved through the recording and
execution of experiments using robot scientists [37].
Humans are reluctant to record every detail of their
cycles of hypotheses - experimentation - interpretations, because it is time consuming and also requires
knowledge of reporting standards. Humans are prone to
make errors in recording information, and they are
biased. Robot scientists are free from such limitations,
and can record all the information required for the
reproducibility at almost no additional cost and in
accordance with the best practices following the recommended standards.

Conclusions and Discussions
Going back to the Glymour paper that opens this article, "Kuhn said that scientific revolutions generally
meet fierce resistance - and the automation of discovery in science is no exception" [1]. However, it is to be
hoped, that the collaboration between human and
robot scientists will produce better science than either
can alone - human/computer teams still play better
chess that either alone. To understand how best to synergize the strengths and weaknesses of human and
robot scientists we need to better understand the

IEEE Technology and Society Magazine

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - March 2018