Might a machine win a Nobel prize?


In 1997, the IBM computer Deep Blue wins a chess-game vs Garry Kasparov. This is considered a milestone in Artificial Intelligence research. Now, a second milestone dates April the 3rd, 2009 with Science publishing two reports on automating science. In the first one, Schmidt and Lipson (Cornell) propose a computational approach for detecting physical laws from experimentally collected data. As a principle for the identification on non-triviality, they first numerically calculate partial derivatives between variables from the data, then they generate candidate symbolic functions by randomly combining (and iteratively re-combining) mathematical building blocks. They finally compare the derivative expressions with the derivate data and score the best pairs according to parsimony criteria.

Given the dimensionality and the complexity of current “omics” data, the computation time required to detect solutions is probably near to 1000-10,000 hours, however the algorithm’s search seems highly parallelizable and very appealing for distributed approaches. What is more astonishing, is the following step. In a second report King and colleagues (Aberystwyth University) extend the concept of “artificial scientist” by generating ADAM:

this is a physically implemented laboratory automation system that […] executes cycles of scientific experimentation. (ADAM) automatically originates hypotheses to explain observations, devises experiments to test these hypotheses, physically runs the experiments by using laboratory robotics, interprets the results and then repeats the cycle.

As a proof of concept, they applied Adam to the identification of genes encoding orphan enzymes in the yeast.

Despite the abundance in data, theoretical gaps still exist in systems biology and integrative physiology, automatic science can potentially increase the rate of scientific progress. At the end of this provocative paper, the authors wonder:

Might this process diminish the role of future scientists?

Quite the opposite: does chess-software diminished the number of chess-players?

Schmidt, M., & Lipson, H. (2009). Distilling Free-Form Natural Laws from Experimental Data Science, 324 (5923), 81-85 DOI: 10.1126/science.1165893

King, R., Rowland, J., Oliver, S., Young, M., Aubrey, W., Byrne, E., Liakata, M., Markham, M., Pir, P., Soldatova, L., Sparkes, A., Whelan, K., & Clare, A. (2009). The Automation of Science Science, 324 (5923), 85-89 DOI: 10.1126/science.1165620

Ruppy, the first fluorescent-dog

ResearchBlogging.orgA Korean team report the generation of  a RFP-transgenic beagle. Dogs exhibits 224 genetic diseases similar to those found in humans making them one of the closest known models for various human hereditary diseases. However, experimentation with animal -which should be at the service of the whole mankind -  raises strong and acute ethical challenges, particularly if the experimental model is a pet.

Although still prototypical, the concept of "reporter animal" arguments toward a new use of animal experimentation based on the generation of a knowledge based on the non-invasive observation of physiological events in living animals at molecular detail. This vision is still in its infancy and several ameliorements steps need to be undertaken. One of them, is the development of better transgenic abilities to safely introduce a genetically-encoded reporter into mammals. Due to the technical dif´Čüculty in obtaining fertilizable eggs and the unavailability of embryonic stem cells, no transgenic dog has been generated so far. Hong et al., report now the use of Somatic Cell Nuclear Transfer (SCNT) to generate, from a stably-transfected fibroblast, a dog carrying the red-fluorescent protein. This will be probably a debated proof of concept.

Hong, S., Kim, M., Jang, G., Oh, H., Park, J., Kang, J., Koo, O., Kim, T., Kwon, M., Koo, B., Ra, J., Kim, D., Ko, C., & Lee, B. (2009). Generation of red fluorescent protein transgenic dogs genesis DOI: 10.1002/dvg.20504