For some people, retirement just does not seem to be an option.

Take Dr. Tom Smith, founder of Seismic Micro-Technology. A few years ago, Smith recapitalized the company, keeping a small piece, serving on the board of directors, and working four days a year. Then he was faced with the inevitable question: “Now what?”

For a man with no expensive hobbies, a joy of working with others, and an unending thirst for knowledge, the answer was simple. “I wanted to do some research that would be applied to working with people again,” Smith said.

Smith contacted the late Tury Taner, who had pioneered the use of neural network technology in seismic interpretation. Taner jumped at the opportunity because, as Smith recounted, Taner said, “This stuff really works!” He gave Smith literature to read, and the two eventually became a team, continuing to research the potential of the technology.

“When I was to the point where I could start making some contributions to Tury, it changed the flavor a little bit,” Smith said. “We just had a ball.”

Sven Treitel joined to make a threesome, and the team had several productive discussions. With Taner’s passing earlier this year, Treitel and Smith are continuing their research, though, Smith said, “with saddened hearts.”

What neural networks bring to seismic interpretation is the ability to study multiple attributes, a problem that is too complicated for the human brain. “In a workstation, when you get a seismic cube up and start looking at it, you’re going to look at two or three different numerical calculations based on your original seismic survey,” Smith said. “Quite commonly you’ll have at least five attributes, and sometimes maybe 25 or 30. I defy any ordinary human being to sit down and, in his or her head, look at one attribute, then flip over and look at another, and then look at a third. After awhile the mind can’t handle even four attributes at a time or really understand the relationship among them.”

He added that human nature is to build a play concept and select favorite attributes that fit that concept, ignoring those that do not fit. “It’s intellectually dishonest,” he said. “But we’re human.”

Smith explained that there are two types of neural networks, supervised and unsupervised. Supervised neural networks are trained to analyze and classify data; unsupervised neural networks are not. Many companies already use the supervised form, but Taner had Smith investigate the latter.

“If we’re looking at the problems of seismic interpretation, they’re very broad, and we’re working in areas where we sometimes don’t have any wells,” he said. “We let the unsupervised neural network technology find the things that are anomalous, find the things that are ordinary, and classify them.”

He likened it to a simple spreadsheet, where the columns represent the seismic attributes and the rows represent the time samples of the seismic trace at a bin center. “It’s about understanding what is common in the data, finding anomalies, and then reducing that down to something that a human brain can handle,” he said. “That’s what we’re having a real kick investigating.”

Smith tested the concept on a survey provided by FairfieldNodal and found anomalies around a salt dome. “It didn’t take us too long to realize that this is an automatic anomaly identifier,” he said. “Sometimes it’s the things that don’t classify well that give us a result.”

Having convinced himself that “this stuff really works,” Smith founded Geophysical Insights and currently is working with interpreters to tweak the application. The goal is to create a completely numerical, unbiased methodology that evaluates possible combinations of selected attributes. And for those who wrote off neural networks as wacky science 10 years ago, Smith assures the industry there will be no “white-shoed salesmen with the latest and greatest golden arrow that points to oil and gas.” Rather, he said, “It’s worth taking a look at because it runs unattended, and it has been demonstrated to deliver useful results in many other industries. And the bottom line is this is nothing more than another tool to use.

“Geophysical Insights will focus on developing neural network and other technologies for the oil and gas industry and apply those tools through consulting and training,” he said. “We believe neural network technology has the potential to be ‘disruptive,’ significantly reducing the risk and time of exploration by exposing deeper insights in the data.”

This disruptive potential, Smith said, is comparable to the fundamental revision of practices that occurred when seismic interpretation moved to the workstation in the 1970s. “The concepts that underlie this technology are as simple as investigating the information that you have on a spreadsheet,” he said. “If you can introduce data to a spreadsheet, it can analyze the natural clusters that reside in the data space and find their relationships. That is the essence of machine learning. That’s very fundamental.”