I had a similar experience in grad school with getting "behind the data". My husband David (also in grad school) had access to an apple orchard that was rather unique. Rather than being planted with a single cultivar, it was planted with dozens of different types of apples so that researchers could study how different apples fought off diseases. Since many trees are particularly vulnerable to infection during bloom, he wanted to check what percent of the blossoms of each variety were open at a given time point. So on paper, it was this dry list of percentages, but in reality it was a beautiful orchard with blooms in a dazzling pallet of rose, cream, burgundy, ivory, crimson and everything in between, plus a range of shapes and sizes.
With the rise of machine vision, I'm worried scientists will spend less and less time actually looking at their samples/subjects, instead relegating the looking to machines. Now, machine vision is great in that it lets you "look" at so much more data and find answers that would be difficult to get to statistical significance otherwise. But I am worried we'll lose something by distancing ourselves more and more from the reality we are trying to model.
I had a similar experience in grad school with getting "behind the data". My husband David (also in grad school) had access to an apple orchard that was rather unique. Rather than being planted with a single cultivar, it was planted with dozens of different types of apples so that researchers could study how different apples fought off diseases. Since many trees are particularly vulnerable to infection during bloom, he wanted to check what percent of the blossoms of each variety were open at a given time point. So on paper, it was this dry list of percentages, but in reality it was a beautiful orchard with blooms in a dazzling pallet of rose, cream, burgundy, ivory, crimson and everything in between, plus a range of shapes and sizes.
With the rise of machine vision, I'm worried scientists will spend less and less time actually looking at their samples/subjects, instead relegating the looking to machines. Now, machine vision is great in that it lets you "look" at so much more data and find answers that would be difficult to get to statistical significance otherwise. But I am worried we'll lose something by distancing ourselves more and more from the reality we are trying to model.