Feasibility and Challenges of Machine Intelligence in Geophysical Systems
Fern Geophysics LLC
To date, machines that have exhibited intelligence of some form are all of the Turing type built on the architecture of Von Neumann. They allow sets of logical operations to be performed based on a finite set of axioms, or rules, and are intrinsically algorithmic in nature. Problems in geophysics, and all others that are related to the subsurface, can be amenable to such an approach and, in fact, have been quite often exploited as such. However, equally often, within the same problem, situations are encountered where a logical set of rules and procedures do not work well. Take the problem of subsurface imaging, for example. Migration is algorithmic, while velocity analysis may not necessarily be posed as such. Indeed, geological considerations and the absolute need for understanding to make stepwise progress in geophysics, perhaps, makes an algorithmic approach intractable for the same reasons as well.
Yet another feature of subsurface datasets that sets it apart from all other forms of data is the richness of noise that is not only varied in character but is also observable at all scales. Noise, which may be defined as nonconformance to a set of rules and logical constructs, includes absenece of data as well, must be accounted for correctly for any machine with intelligence to work well; that is to say, it should be able to distinguish signal from noise intelligently. With the exponential increase in compute power and storage in recent years, large volumes of data are now being stored and processed extremely efficiently today. The question that may then be asked is whether these modern machines can be made intelligent for geophysical systems, or, given the nature of the problem, are there fundamental limits that must be admitted regardless of how powerful the machine is? Using examples from the geosciences and elsewhere, I explore some of these issues in this talk.
Debashish Sarkar is a geophysical consultant. His industry experience includes working at Conoco Research, GX Technology/Ion Geophysical, Lumina Technologies and GE/BHGE Research with technologies related to seismic imaging, AVO, borehole acoustics and production engineering for both conventional and unconventional reservoirs. He has a PhD in Geophysics from the Colorado School of Mines.