IEEE Technology and Society Magazine - March 2018 - 44

perform" (Lady Lovelace's italics). This argument has
real teeth as the robot scientists Adam and Eve are very
far from being autonomous agents seeking out their own
scientific problems. One counter-argument is that robot
scientists are programmed to learn novel scientific
knowledge, and as they learn from observations of the
physical world the conclusions are not purely deductive.
A variant of the lack of novelty argument is the argument that computers will only ever be able to do "normal science," i.e., within a paradigm, and will never be
able to do "revolutionary science" [23], [27]. Certainly
existing robot scientists are not capable of doing revolutionary science. However, very few human scientists
are either.
There is also an argument that such systems as
robot scientists are not truly autonomous, rather that
the systems merely are tools of scientists, and that there
is always a human-in-the-loop [28]. The existing robot
scientists are compliant with the definition of autonomous robots by IEEE P1872TM/D3 standard "a robot
performing a given task in which the robot solves the
task without human intervention, while adapting to
operational and environmental conditions." While one
may argue that there are certain shortcomings in Eve's
autonomy, the concept of a robot scientist, as originally
introduced [11], implies a complete autonomy in scientific discovery.

strengths of human and robot scientists, and to better
understand future working relationships between
humans and automation. These relationships occur at
many levels: from the most profound (deciding on what
to investigate, structuring a problem for computational
analysis, interpreting unusual experimental results,
etc.), to the most mundane (cleaning, replacing consumables, etc.).
One particularly interesting relationship between
human and robot scientists relates to the replication of
experiments. It has been proposed by Latour [31] that
there is a necessary trade-off between the communication of conceptual information and contextual detail:
scientists need to report their findings in the most
objective and abstract forms possible, so as to create
generalizable statements. However, the decision of
which detail is conceptual and which is contextual is left
to the individual scientist. Anthropological and ethnomethodological research indicates that this is a trait of
the human mind: scientists first decide on the experimental design in very broad terms, then later infer how
to empirically conduct the experiment, and lastly report
their results in mostly the same terms as the experiment was designed [33]. Robot scientists, on the other
hand, require a set of definite contextual elements to
work properly, and they always log these conditions of
the experiments they conduct.

An Anthropological Perspective

A Sociological Perspective

Compared to human scientists, robot scientists have an
fascinating mixture of super- and sub-human abilities.
Laboratory robots have traditionally been used to automate low-level repetitive tasks. Robot scientists inherit
this ability and have the super-human capacity to work
flawlessly on extremely repetitive tasks for days at a
time. In comparison humans perform badly at repetitive
tasks, especially those carried out during extended
periods [29]. We have confirmed this during our observational studies of human scientists, who routinely
make mistakes, particularly when subject to hindrances like stress, time pressure, or distractions. Robot scientists inherit from AI abilities that have traditionally
been regarded as high-level for humans, such as a
super-human ability to do logical and probabilistic
reasoning. However, robot scientists are sub-human in
their adaptability and understanding, and human scientists are still unequalled in conditions that require flexibility and dealing with unexpected situations, especially
those intuitive functions that might have otherwise been
considered low level [30].
Given the mixture of super- and sub-human abilities
of robot scientists, it is informative to investigate how
human scientists cooperate with their robot counterparts, both to improve the technology by playing to the

We argued above that there are convincing examples
now of scientists using computer programs to contribute to scientific knowledge. We are particularly interested
in the nature of that contribution and how AI-informed
scientific knowledge may differ from knowledge gathered without the assistance of computer programs. The
earlier description of what a robot scientist does contains two key processes that are critical to understanding scientific knowledge from a sociological perspective:
observation and interpretation. We will take each of
these concepts in turn. The aim is to offer a sociological
perspective on laboratory automation that is informed
by some key ideas introduced from sociology of scientific knowledge literature.
When scientists use the term observation they are
clearly making a connection between a sensory input
(e.g., "seeing") and the material world around them.
However, a great deal of phenomena in the sciences
cannot be seen in any literal sense. For example, when
a physicist says there are many ways to "observe" the
recoil of an atom, they are not referring to a process
that can be inspected in the same way that the color red
can be seen when a chemistry teacher asks students to
observe red fumes in a gas jar [34, pp. 1-2]. It is important therefore to understand exactly what is happening

IEEE Technology and Society Magazine

∕

MARCH 2018

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