Section 2: Visualizing Scientific Models (Some Assembly Required)
The purpose of models is not to fit the data, but to sharpen the questions.
— SAMUEL KARLIN
Tinkering in the Model Shop
Whether it’s gaining a better understanding of brain function or galaxy formation, scientists increasingly employ data-intensive simulations based on a combination of observations and theories. Visualizations of these models (which we’ll explore in a few moments) don’t need to embody every possible aspect of a system to be valuable investigative tools. For example, in one visualized model example that follows, you can see some effects on a star that is undergoing the gravitational tug of a black hole. These visualizations just need to capture elements of complex systems and dynamics that people can probe, test, and predict. Some of the uses include:
- ? Investigating theories that are very difficult, if not impossible, to observe directly
- ? Accelerating the probing and testing of a hypothesis and its parameters (for example, how much dark matter is required to give a galaxy its observed shape?)
- ? Providing a way for scientists to use visual representations to explore complex systems
- ? Highlighting different aspects of a system by using different kinds of visualizations
- ? Revealing the potential missing pieces and hidden connections in the data
- ? Enabling better collaborations between researchers
- ? Conveying scientific concepts to general audiences
- ? Providing literal “hands-on” experience with conceptual models
Yet, how can scientists be certain their simulations truly represent reality rather than capturing quirks of code or faulty assumptions? Perhaps the simulations are revealing something completely new, true, and unexpected. Well-constructed visual simulations can be compared with like representations displaying alternate models. When possible, the simulations can also be matched against direct observations of the actual subjects of investigation.
Making model visualizations that encourage exploring, testing, and even playing with scientific hypotheses demands paying attention to various details of the users’ experience. Think about it: who would want to spend much time playing with a toy train that continually falls off the track or a toy plane that crashes back to the ground? An interactive visual model of a scientific concept that was glitchy or frustrating would not invite engagement or the kind of exploration that leads to important insights.
In general, that means the responsiveness of the visualizations needs to be crisp and the interactions easy. They need to be fast and simple enough to encourage pushing and probing of the data. Given the typical scope and scale of data in neuroscience and cosmology, that’s no mean feat. Decisions have to be made about how and where the numbers are crunched and the manner in which images are served up and rendered.
There’s often an ongoing tinkering and tuning of code and hardware. For example, how much work will be demanded of the user’s own device versus the machines serving up the simulation? Mapping the data, in terms of their relationships and the organizational structures in which they reside, is an integral part of the process.
Morgan MacLeod, a graduate student at the Department of Astronomy and Astrophysics at the University of California, Santa Cruz, has been working with visualizations of physics models to help researchers — and the general public — develop a sense of what happens during certain interactions between stars and supermassive black holes (SMBHs). For example, if a star strays too near to an SMBH, gravitational forces can rip it apart and the stellar material can be stretched into two tails. These are big, but rare, events that occur very far from Earth, so scientists can benefit from models and simulations to get a good view of these epic encounters.
One of the choices that astrophysicists have to make regularly, MacLeod says, is whether to use representative or exact system models and simulations. He explains that in an exact system approach, the scientist tries to re-create the entire system with every known element included. However, this might not always be the best and most direct road to gain a clear understanding. “There are many times when people can only show or digest one slice at a time,” MacLeod says. He adds, “It should not be the ambition of a visualization to always be complete.” It just needs to be able to clarify a point. You want to show the constraints of the information but also convey something.
There can be obvious differences in the approach to designing a targeted visualization for focused researchers versus one geared for lay audiences. Some kinds of visualizations can be tuned to show a highly specific aspect of a system that will be meaningful only to people steeped in the subject matter. With that said, it can be easy to fall into the trap of assuming great knowledge of a scientific area necessarily translates into proficiency with other kinds of technology.
For example, a group of scientists might have very deep knowledge about their subject area but be unfamiliar with the nuances of how to work with a certain kind of interactive chart. Crucially, they might not be able to gauge how much to trust the representation they are looking at or what to do next to evaluate it. Lay users might be proficient at working with the software but not have enough context to understand what they are seeing. Many subtle differences can emerge between different types of users and these should be considered in the design process.
“There’s trade-off in the people who are willing to invest a minute versus the people who are only willing to invest a few seconds. I often ask myself whether I should throw in x, y, z aspects of a model into the visualization. Even with scientific graphs there’s a certain degree of salesmanship, in that you want to make a point and have people see and believe it,” MacLeod says. Simplifying models that isolate specific aspects of the data and turning this into visual and physical forms is not a new impulse. Centuries ago, people created clockwork embodiments of the solar system to help give the viewer a sense of the relative positions and motions of bodies in the solar system with the heliocentric orientation (the Sun rather than Earth as the center of the system.) In some respects, MacLeod, who at the time of this writing is living in Rome, traces a lineage between the visualizations of the dynamics of black holes he’s helped to create and the orreries (an example is shown in Figure 7-6) he’s seen at the Museo Galileo at the Pellazo Castellani.
Figure 7-6. An example of an orrery
Although these orreries are not built even remotely to scale or entirely representative of planetary motion, they still serve a useful function to convey the best understanding of the time.