Agree broadly with everything in the article. However, I have one point to add:
> Physical models that reproduce behavior are limited by the physics of the world, while computer models have much looser bounds.
This is also one of the weak points of computer modeling. The outputs are strongly influenced by modeling assumptions, and without feedback from reality, one can easily be deluded towards wrong models. The computer simulation encodes the map not the territory. This is particularly true in hard to model fields like economics & finance. As long as one appreciates that, computer modeling is a powerful tool.
Regarding what would make modeling more accessible to computer users, I think the answer is an environment where users have access to the full live system (few boundaries) and can tweak it in real-time, with right feedback loops. Eg: Smalltalk, Lisp.
I like how Paulien Hogeweg phrased it too: "thinking in the most interesting simplification" [0]
I always thought the 'but models are wrong' is quite a weak excuse. Newtonian physics is a wrong model of how the universe works. Using it though; we put a man on the moon. Einsteinian physics are incomplete, but it allows us to use satellites moving at roughly 10KM/s to pinpoint our location to within a few metres.
A model's usefulness is not about how accurately it describes what is happening, but about whether it describes what happens at a certain meso-scale accurately.
>> Einsteinian physics are incomplete, but it allows us to use satellites moving at roughly 10KM/s to pinpoint our location to within a few metres.
Funny. I know how GPS works and that knowing the SAT locations is critical, but I never thought about how fast they are moving. They need to transmit their own position and time with great accuracy. In particular, don't they have to transmit their time and position at the time of transmission? That is a more interesting problem than I realized.
GPS satellites aren't really transmitting their location at any particular time, they're transmitting the so-called ephemeris and almanac data that has information about their orbits so that you can calculate their position at the time of transmission.
And if you have a ADALM-PLUTO, you can download the brdc file (ephermis) and spoof all 12 satellites to a location of your choosing here: ftp://cddis.gsfc.nasa.gov/gnss/data/daily/2019/brdc/
As someone who have computer modelling as a core of my research it makes me happy to see this quote mentioned here.
People on other science forums are generally fast to dismiss computer models as 'unrealistic' and 'non-representative', disregarding their utility.
For those interested in models I always recommend looking at Schelling's segregation model for starters, the premises and conclusions drawn from the experiment are quite interesting.
For myself, whenever I read about someone's new model and their predictions, I'm always looking to see what their model was verified against. If they've got a model, and it is verified to touch reality in some place we can check it, then I'm willing to at least listen to what the model says about where we can't touch, see, or test. It could be wrong, but it's at least the start of a conversation.
But if it's been verified against nothing, I don't have much interest in it.
Some examples of that include almost every model that claims to describe some element of human social interaction, which never seem to be checked against anything real that I can see, and I recall once reading an article and paper on modeling exoplanets and their likelihood of having life on them (or the likelihood of it being hospitable, not sure which) that flew so far beyond our science and data that the peer reviewers should be ashamed they ever approved it.
(Of course, if you want to publish your unvalidated model as an interesting piece of math, go nuts.)
> Physical models that reproduce behavior are limited by the physics of the world, while computer models have much looser bounds.
This is also one of the weak points of computer modeling. The outputs are strongly influenced by modeling assumptions, and without feedback from reality, one can easily be deluded towards wrong models. The computer simulation encodes the map not the territory. This is particularly true in hard to model fields like economics & finance. As long as one appreciates that, computer modeling is a powerful tool.
Regarding what would make modeling more accessible to computer users, I think the answer is an environment where users have access to the full live system (few boundaries) and can tweak it in real-time, with right feedback loops. Eg: Smalltalk, Lisp.