392

31

The Organization of Knowledge

record of the experimental conditions, as well as of the results, can be compiled auto-

matically; this, too, may be superior to the traditional hand-written laboratory note-

book, at least as far as long series of almost identical experiments are concerned. ¹⁸

This approach also has the advantage, compared with microarray experiments (which

are, of course, usually at least partially robotized), that conditions individually appro-

priate to each experiment can be applied, avoiding the possibility of errors due to the

uniform conditions applied to an entire microarray not being appropriate for some of

the reactions in some of the places on the array. The only downside would appear to

be the elimination of the possibility of the serendipity that has often been so valuable

for making discoveries.

A more ambitious development than running a preset workflow is to automate the

actual design as well as the execution of experiments, and even, if the promises

of unsupervised data mining are realized, to analyse the results, recalling Leib-

niz’s ambition to have a device that automatically separates truth from error, and

Hilbert’s “Wir müssen wissen. Wir werden wissen”. ¹⁹ The “robot scientist”, which

has access to background knowledge, devises hypotheses to explain extant obser-

vations and designs experiments to test the hypotheses, which are then run using

well-established technology, and the results interpreted. ²⁰ Given the vast scale of

manipulations required to elucidate gene functions and the like, this is a very neces-

sary development. It has been realized as a robot able to measure the growth curves

(defining the phenotype of a relatively simple microorganism like yeast) of selected

microbial strains (distinguished by genotype) growing in defined environments. ²¹

The problem to which the robot has been applied is the identification of the genes for

enzymes catalysing reactions thought to occur in the microbe. The robot was pro-

vided with extensive knowledge of metabolism, and software to produce hypotheses

about the genes and to deduce corresponding experiments to test the hypotheses.

These experiments were then executed by selecting strains from a collection given to

the robot, measuring their growth curves on rich medium and then inoculating them

into minimal medium to which additional metabolites, also selected by the robot,

were added, after which growth curves were again measured.

Such automation is well suited to answering questions of this nature, the frame-

work within which they are formulated being well circumscribed and carefully for-

mulated by the investigator who actually designed the robot: essentially it functions

as an extension of the brain and hands of the investigator. As such, it is an extremely

valuable aid and the proliferation of this technology will considerably accelerate the

18 Laboratory information management systems (LIMS) were apparently originally introduced

merely to automate billing for services carried out by contract laboratories, but nowadays are

capable of covering entire workflows, including tracking samples, recording the measurements,

and generating the reports of results.

19 Enunciated in his retirement address in 1930, one year before the publication of Gödel’s incom-

pleteness theorem, which rendered the ambition unattainable.

20 King et al. (2004).

21 King et al. (2009).