Commish For A Day II: #6 Automated Strike Zones

(Another issue is how much we trust these machines to get it right. Is the margin of error used in QuesTec acceptable for games? I think so, but I can imagine the PR nightmare if people find that the system gives too much of the corner of the plate, or something similar.)

One of the biggest problems with the game is that the umpires don’t enforce the zone, and many have “personal” strike zones, and the players know it. Worse, umpires give calls to individual players based on reputation. Heck, pitchers and catchers have skills like framing pitches, expanding the zone, and so on, which are really ways of tricking the umpire into calling more pitches in their favor. The strike zone is a fixed area, and should be enforced as such.

I think there are two main arguments against this:

First, people may complain about removing the human element from the game. I have 2 responses to that:

  1. We’re not removing it entirely- you’ll still need umpires to make decisions about out/safe calls, whether a batter checked his swing, calling infield flies, and all kinds of other things that either require judgment calls, or that no amount of camera work or automated processes can do better than an actual human on the field.
  2. Calling balls and strikes is one instance where the “human element” has gone way to far, and taken the game away from where it should be. Umpires are messing the game up and making it too much about themselves.

The other issue is, “Can we do it?” I’m not an expert in this technology (and a lot of it seems to be proprietary, so only the people working for the companies know the details), but let me take a crack at it anyway.

Here’s where I think we are:

  1. We have (from here) processing software that: knows from a camera angle (or several) whether a ball thrown from the mound crosses the plate. Presumably (again, I don’t know the details), it knows enough to deal with different arm angles of the pitcher, different batter heights, the fact that the strike zone is three-dimensional, etc. This, to me, is the largest processing hurdle, and it seems to be solved.
  2. TV broadcasts have something (maybe QuesTec, you see a lot of vague reports about how it’s “used in broadcasts”, but not how or how often) that show you the result of many pitches right after the pitch.

If we want to implement this, then these are the questions that need to be answered (that I assume Selig has the answers to):

  • Q1: Is the version we see on TV as accurate as the version used to grade umpires? If not, how slow is the “real” version (or, how inaccurate is the TV version?) This is the key question, I think. If the TV version is really inaccurate, and the “accurate” version takes 1 minute or so per pitch, my idea isn’t workable. I don’t know what the truth is here.
  • Q2: Is the few-second delay we see on TV between the pitch and when we see the graphic a necessary delay because the computer is crunching the data? Or is it there because it fits better in the broadcast?

What these questions boil down to is “Suppose we want to find out whether the pitch is a ball or a strike (down to the accuracy used to grade umpires) and do it within some given time frame (say, .5 sec). Can we do it?” If we’re already pretty close, then some combination of:

  • Throwing more hardware at the situation (a few million dollars for supercomputing power shouldn’t be much for MLB),
  • Telling the folks at QuesTec to modify their program a bit, and
  • Waiting for computers to get faster as they always do

…should be enough to get us there. But while I can see these incremental improvements taking time time to process a pitch from a few seconds to under a second, I can’t see them doing something as large as shaving minutes off of the processing time. To do that, you need a whole different approach, which is beyond the scope of what we’re talking about.

(As an aside, I think the answers to both of these questions is- or can be- “We can do it really fast“. The reason for this is that I’m reminded of the stupid FOX hockey-puck tracker of a few years ago. People hated it as a TV device, but from a technological standpoint, look at what it did- we had a way to show, on screen, the position of a physical object while it was moving around, in real time. That was several years ago, and I’m sure that with a fixed (for each batter) strike zone, we could do something similar to a baseball, and then see if the path of the ball went through the fixed zone.)

  • Q3: Is it possible to do every pitcher and every batter in the time frame? Or do “non-standard” situations (exaggerated Rickey Henderson crouches, submarine pitchers, knuckleballs, people forcing you to move your camera because someone hit a home run off of it,…) affect the amount of processing work done? The plan only works if we get the result fast in all situations. I’m kind of 50/50 on whether this is likely.

One solution that’s not workable here is to force every pitcher and batter to pitch or swing in front of the camera, so the system can be programmed to learn about each hitter’s swing, and each pitcher’s arm angle. This won’t work because it encourages players to “cheat” and, for example, pose for a smaller crouch in front of the camera than they use in an actual game to shrink their strike zone. So, we need to see if the software can figure out the strike zone in any situation without knowing beforehand anything about who’s pitching or who’s hitting.

  • Q4: (this is more of a “deciding policy” issue..) What if there is a glitch (could be hardware, could be communications, could be something like a beach ball in front of the camera) and the automated call doesn’t make it to the umpire. What do you do?

These are just off the top of my head, there may be more, or I could be missing something important. But either way, I think it’s a conversation worth having. You don’t ever hear anyone say “We’d like to do it, maybe in a few years when we get the processing power”, or “Yeah, that’s a goal of ours, when we clear some technological and logistical hurdles”. Instead, you hear “We can’t remove the human element!”, like people are afraid that automating this will cause the rise of the Matrix, or something.

Thanks for giving us some interesting food for thought, Sean. Automated strike zones? Calls being made by computer, like in tennis. More accurate, less “personal interpretation” of the strike zone, less favoritism, less “human element”.

Other “CFAD” submissions

  1. Commish For A Day II: #1 Expand Playoff Teams
  2. Commish For A Day II: #2 Roster Expansion
  3. Commish For A Day II: #3 Expand the DH to the NL
  4. Commish For A Day II: #4 Umpire Statistics
  5. Commish For A Day II: #5 Salary Cap & Floor
  6. Commish For A Day II: #6 Automated Strike Zones

About @Jason_IIATMS

IIATMS overlord