How accurate is a NBA2K simulation in terms of what we might see in a month’s time? The video we’ve placed at the top of this text looks like basketball, but how much of it is basketball?
That question leads to other questions — questions like, “Is it … realistic that Kelly Oubre is backing down Javale McGee in the post?” “Does Dwight Howard really participate in five out that much?” “Would Howard really have trouble backing down Devon Booker?” — and it runs straight into the deep variance that exists at the heart of sports.
You can measure the trajectory and entropy of ball movement in basketball and American football. There is a relationship between competitive balance and the popularity of an event. You can have a go at measuring who’s a predictable hitter in baseball and who’s an unpredictable hitter. NHL ‘underdog’ teams tend to do statistically better than basketball teams.
And there’s a whole lot more. But these kinds of questions also might explain why a software engineer named Oluwatobi Popoola landed on this notion —
Why was I assuming that NBA2K was actually going possession-by-possession, accounting for a bunch of insanely specific scenarios for all ten players on the court? How we get to simulated results doesn’t matter, as long as the results look realistic and remain realistic over a series of simulations.
We’re not looking to have Dwight Howard’s or JaVale McGee’s size and strength accurately simulated vis-a-vis a mismatch. We’re not looking at an AI trying out different offensive or defensive strategies on each other. We might be looking at something that just ends up vaguely accurately estimating someone’s overall statistical tendencies.
But is that still the game? Especially as we contemplate firing up a machine so we can watch Klay Thompson’s swiftness continue uninjured and unencumbered?