ally by experimentation based on the model’s predictions. This process of falsification is the defining characteristic of scientific models. 65 If it is successful, we have discovered a property of the system that is not captured correctly by the model, which may lead to an improved model and to a new falsification cycle. If it is unsuccessful, the model stands, which does not necessarily mean that it is correct: it simply means that it has not yet been proven wrong (and we should keep trying). A main feature of the scientific method is that the “truth” is in the system, while the model is in principle never fully correct.

This Popperian approach has been adopted as part of the recent idea of a Turing test aimed at biological modelling, 40 where the model is deemed valid if it cannot be told apart from the actual biology. Here, of course, we do not advocate comparison of the actual material, but just of the behavior (as is the case for Turing’s original test for machine-generated intelligence). However, in contrast to Turing’s original test, falsifying the model here is something that we actually strive for, since it is a wonderful way to encourage further research.

The engineering method starts by producing a model (for example, a blueprint, or a specification) of what we want to build, and proceeds by building it. We then aim to show that what we built is in fact an implementation of the model. Such a verification process compares the outcomes of the system to the predicted outcomes of the model, by testing and model checking, which is in many ways similar to scientific experimentation. If this is unsuccessful, it means that we have discovered a property of the model that is not correctly implemented by the system, which may lead to an improved system and a new verification cycle. If it is successful, the system stands, which does not mean that it is correct: it simply means that we have not yet found the next bug (and we should keep trying, by making the model/specifica-tion more complete). A main feature of the engineering method is that the “truth” is in the model, while the system is in principle never fully correct. Thus, science and engineering work in opposite directions.

Incidentally, reverse engineering,
the process of deriving an unknown
model from an existing system, fol-
lows very much the scientific method.
Conversely, (direct) engineering could
be also called reverse science. As an ex-
ample of the difference within a single
discipline, consider systems biology,
which is largely a scientific enterprise,
as opposed to synthetic biology, which
is largely an engineering enterprise. Of
course, there are strong interactions
between science and engineering, with
one inspiring the other. Many engi-
neered systems are inspired by biologi-
cal systems that have been scientifically
investigated (for example, genetic algo-
rithms, neural networks), and converse-
ly, as we have argued, modeling biologi-
cal systems can be inspired by modeling
techniques in engineering.

Modeling a complete organism. We feel that it might be beneficial to use the ideas and methods discussed here to model a complete biological system. In fact, a “grand challenge” of modeling a full multicellular organism has been proposed, 39 motivated by the belief that unprecedented depth of understanding life and its mechanisms will result from such a model. The dream is to model the organism as a reactive system, the backbone of which would be its multitude of cells and their interactions, but to include the relevant inner behavioral aspects of the cell on the molecular and biochemical level as well. The 1000- cell Caenorhabditis elegans nematode worm, better known simply as C. elegans, was suggested in Harel39 as a possible system to model.

Obviously, this is less ambitious than modeling, say, the entire population of a species, and more ambitious than modeling a mere cell. The choice of which system to address is a matter of taste, but our feeling is that an organism would be a good compromise that would yield enormous benefits, if it can indeed be done satisfactorily. The question of when to stop, that is, when is the model deemed valid or complete, is a very interesting one, and we have proposed that the Turing test mentioned previously could be a good first approximation: We are done when the model’s behavior cannot be distinguished from that of the real thing, in which case the model can be said to be a theory of the organism; see Harel. 40

This whole organism project (WOP) would take many years of work, and would entail using a variety of methods and to interconnect them all smoothly into a full, true-to-all-known-facts, 4-dimensional model of the creature. We would want the model to be easily modifiable and extendable as new facts are discovered, to have an animated, anatomically correct front-end, which would have to be tightly linked to a reactive system model of the organism. The front-to-back linking could be done using the idea of reactive animation. 24 Most importantly, the model would enable realistic simulation of the organism’s development and behavior (this is the fourth dimension), and would lend itself to the kinds of analysis techniques discussed earlier. All of this could help uncover gaps, correct errors, suggest