Google Earth Community: Search for Ron Boychuk

Thanks for the responses.

I have a log-in at InternetSAR, so I have read the briefing many times. My questions are left over after reading it. An experienced SAR person may have written it, but it leaves out the most important details from a forensics perspective. SAR and accident forensics go hand-in-hand--before we search we must have theories as to what might have gone wrong, in order to narrow the search area and make best use of resources.

I understand that coding is a complex and time-consuming task and that features will come. In the interim, since a search (or two) is in progress, manual editing of the mission briefing is called for. I strongly suggest getting the system up and running with manual editing, then automating the tasks as we go along.

I understand confidentiality; it can be protected absolutely while providing searchers with necessary information. Personal names and identifying information are unnecessary. Just the facts.

I am doubtful that waiting for spring for more imagery is a good plan, but I understand that there may be no other option. Some of the likely locations are permanently snow covered. The winter's snow mass will progressively hide any disturbance of the regular snow pattern, and the spring melt cannot be expected to uncover any wreckage in an area of permanent cover; indeed, it may complete the obscuration of a site.

In an ideal world, a jet with a high-res camera will be dispatched on a mapping mission immediately following any disappearance, or at the first weather opportunity. Handled properly, this technology will supplant conventional SAR. This is not the reality, but "more images, sooner" should be the mantra of Internet SAR.

To the two posters who dissed the process of wild-assed guessing, you have missed the point completely. Humans function by processing the best information available, making a hypothesis (or wild-assed guess), then testing it and reevaluating the hypothesis based upon the results. We do very, very poorly at monotonous and repetitive tasks that require monitoring of complex systems (like overlays in Google Earth.)

This fact has been demonstrated repeatedly in careful studies, and has led to changes in flight crew training, and ultimately to the science of cockpit resource management, or CRM.

The way to use many human volunteers to search effectively for a lost plane in a bank of high res images is to give them the best information possible, and let them take their wild-assed guesses, search the predicted area (which they will do with vigour, since they are testing their own theory.) Then, when they find nothing, as most will, let them refine their hypothesis and try again. As the available information improves, hypotheses will be refined and search areas will change.

Inter-individual variation will ensure that everyone's guesses are a little different. Some people won't have theories and will search at random or follow a grid.

The function of the managing computer system is to ensure that all this random activity efficiently covers the entire grid as thoroughly and systematically as necessary. If areas remain unsearched, it must present incentives to people to search them, for example, a button that says "click here for the most neglected areas."

Expecting people to grind through image after image systematically and with little input or no plan of their own is simply unrealistic. Even if they do it, they are likely to miss objects of interest because humans make poor robots.

What we need is thousands of wild-assed guesses, informed by an ever-improving information base, refined by real results, and managed by a system that sees the overall picture and gently guides and encourages the searchers.

That's the way real SAR works, not at the level of searching the grid, but at the higher level of defining the grid to be searched based upon information, expertise and, yes, wild-assed guesses.

Conventional SAR doesn't have the resources to search every square mile between Relevstoke and Qualicum, and neither do we.

I'm willing to bet that someday wreckage will be found that was the subject of an Internet search (perhaps Steve Fossett's aircraft) and it will turn out that the wreckage was clearly visible on one of the images that was searched, and yet every human missed it because they were just grinding through a list of images, rather than testing a hypothesis.

Here's an example: Say Steve Fossett suffered a heart attack while flying and lost control of the aircraft, placing it in a tight spin with a high-impact crash and subsequent fire. If one searcher looks through the image containing the wreckage expecting to see wings and a fuselage, they may see nothing and move on. On the other hand, if some weirdo like me hears that he drank three cups of coffee, say, at 8 am, departed at 9am, and the a/c cruised at 110kts, I'll draw a circle from the Flying-M Ranch to the maximum range he could have achieved by the time his predicted blood levels of caffeine maxed, resulting in coronary vasoconstriction. Then, I'll request images along the max range circle and examine them, specifically looking for a dark, chaotic blotch that was absent on the earlier imagery.

As a sole search strategy, this would be awful, except if it happened by chance to be correct. But multiply this by thousands and monitor and guide it with a system that keeps the overall pattern in reasonable order, and you have a recipe for using humans to do SAR quickly and effectively while retaining maximum enthusiasm and participation.

This way, InternetSAR can become the new Guitar Hero, or something. It all depends upon getting the maximum reliable information to the searchers, categorizing the information from most to least reliable, and managing the overall search for effectiveness.

If you want to eliminate hypothesis testing, write an AI to poke through every pixel. Once it is proven better than humans, take the humans out of the loop. I think that's a ways off yet, though.

B.

Post Extras