Please, cite this online document as:
Cobas C., Seoane F., Sykora S.,
Automatic Structure Verification (ASV) as an AI wizard: the milestones met and the challenges looming ahead,
Poster at 54th ENC, Asilomar (CA, USA), April 14-19, 2013, DOI: 10.3247/SL4Nmr13.002.
The community of NMR spectroscopists certainly does not need an explanation of the meaning of terms like 'structure verification' or 'structure elucidation' or, more generally, 'interpretation of a set of NMR spectra'. Likewise, the addition of terms like 'computer aided' or 'automatic' of the previous terms is quite straightforward and intuitive. But is it really? Actually, from the algorithmic point of view, each of these terms means something considerably different (though related). For example, it may be not quite clear to those who are not actually involved in software development how radically different is a 'computer aided' evaluation of a set of experimental, real-world data, and a completely 'automatic' handling of the same set. Or how much does ASV differ (using the same set of data) from the task of discriminating between several molecular structures. Likewise, while everybody understands that software products addressing these tasks should help the spectroscopist to carry out his tasks, very few think that it will be ever possible to eliminate a major portion of his today's everyday work. We believe that we have reached a point in the development of ASV where we do understand the real limits and can address the above distinctions and judge how far one can actually go along this road.
In this presentation we address the above topics, and also share with you some of the achievements, milestones, victories, tricks, and - why not - errors, we went through during this intense four-year work. We believe that the data-evaluation wizard we taught to stand on its feet and walk, and which we are now teaching to run, will become much more than an NMR tool. So many questions had to be addressed to make it work: data reduction into easily manageable forms such as the global peaks list (GSD), combined with the elimination of many artifacts, detailed editing of the reduced data, massive 'filtering' of useful information from the accidental, handling of real-world fuzziness and aspects of the data which still remain ill-defined, even after all the filtering, the way of incorporating specific know-how (in this case, the NMR 'book'), the way to ponder and combine fragments of information arriving from quite different quarters and based on different considerations, and a massive application of new algorithmic techniques suitable to mimic parallel 'thinking' and even 'intuition'.
Much of this experience, we believe, could be put to work also in areas other than NMR but that, of course, goes beyond this presentation. We do believe, however, that through the development of this kind of a wizard, NMR can significantly contribute to the understanding of how a real-world artificial intelligence should be built.
References and links:
 Cobas C., Seoane F., Sykora S., Global Spectral Deconvolution (GSD) of 1D-NMR spectra,
SMASH Conference, Santa Fe (NM, USA), September 2-10, 2008, DOI: 10.3247/SL2Nmr08.011.
 Carlos C., Stanislav S., The Bumpy Road towards Automatic Global Spectral Deconvolution (GSD),
50th ENC Conference, Asilomar, CA (USA), March 29 - April 4, 2009. DOI: 10.3247/SL3Nmr09.003.
Your comments are welcome and will appear here