In which I reflect on my first meeting with the late Georges Grinstein.
In Memory of Georges Grinstein: Is There Magic in Visualization?
By Niklas Elmqvist, University of Maryland, College Park
I recently returned from IEEE VIS 2018 in Berlin, Germany, and during the conference I attended part of the evening session in memory of Georges Grinstein. Georges was a longtime member of the visualization community, and passed away unexpectedly in February 2018, so the VIS conference organized a special session in his honor.
Unfortunately, I had a dinner that conflicted with the second half of the session, so I didn’t see all of it, but I enjoyed the warm and thoughtful, yet quietly humorous tone of the evening. It seemed appropriate since Georges had a wicked sense of humor and was always ready with a lopsided yet good-natured grin. I didn’t have time to share any of my memories of Georges during the session, but the evening got me thinking, and I eventually remembered my first meeting with him. As it turns out, that meeting was quite formative to me, and sparked a fundamental question that has driven much of my career.
In the spring of 2006, I was still a Ph.D. student at Chalmers University of Technology in Sweden and visiting John Stasko’s group at Georgia Tech in Atlanta for the semester. During early spring, my then-girlfriend Helene and I rented a car and drove a couple of hundred miles to Little Rock, AR to visit my friend and collaborator Edi Tudoreanu at the University of Arkansas, Little Rock. We were going to plan out our next research project (which later was published at the IEEE VRST conference). As it turns out, Georges Grinstein was also visiting Edi’s department at the same time, and I was able to join in on Edi’s meeting with him, as well as go to lunch with the entire group.
When we had some time for talking just the three of us, Georges looked over at me with a sparkle in his eyes. “So, you’re a visualization Ph.D. student,” he said conversationally. “Tell me, do you think there is magic in visualization?”
I was caught entirely off-guard, and must have stammered an incoherent reply. Recall that I was just a mere student, still wet behind the ears, and meeting the famed Georges Grinstein in such an intimate setting was one of the highlights of my humble career at that point. I remember a triumphant origin story whooshing through my mind at the time. This is how I am discovered, I thought to myself. I say something profound, and Georges raises me to the skies.
“Magic, obviously,” I said when the drumroll in my head had ended. “Definitely magic.”
I remember that Georges grunted and looked disappointed. “Really?” he said. “Then I challenge you to come up with an insight that is only possible to make using visualization and that I cannot write an automatic query for.”
I puttered around for an answer to his challenge for a while, but could not come up with a satisfying answer. I said so sheepishly.
“So you see,” he said contentedly. “There is nothing magic about visualization in itself. If you can write an automatic query, it is by definition not magic. The only magic in visualization is that it allows you to answer questions you did not know you had.” A pause, a twinkle of the eyes. “And that’s not insignificant.”
Today, I know that there certainly are benefits to visualization that are difficult to replicate in an automatic query. For example, many clustering algorithms still struggle with cases that are trivial for a human, such as a ring of points surrounding an inner group of points (easily separated as the ring and the center for the human). However, these are likely not insurmountable obstacles, and in the end, I think that Georges was right on the money about the lack of magic in visualization. There is no magic, and that is fine, even good. We should not have romantic attachments to our research area that inflate its value and obscure its flaws. Rather, we should objectively reason about the strengths and weaknesses of visualization, and this is what I do when I teach visualization in the classroom.
Visualization is perfect for exploratory data analysis, as well as for communicating your findings. However, it is less useful for confirmatory analysis, and automatic tools will be much better and more efficient if you know the question to ask. There is no intrinsic superiority to exploratory or confirmatory analysis, just different situations when they are called for. This was something I certainly had not realized when I met Georges for the first time, but it was his provocative question that eventually led me to understand this fundamental insight. And for that, I am in his debt.