For something over ten years, I was something of a Madden fanatic. I eagerly awaited Maddenoliday every year, and even those years when my Eagles were not contenders, I rooted for them every week because I knew that if the IRL Eagles did well, the Madden Eagles eventually would, too.
If you don’t know about Madden football, well, you’re probably not even reading this. But in case somehow you are reading it anyway, it’s important to understand that the players in Madden are modeled from the actual players. Players move on and off the Madden roster as they join and leave the real teams, and each player has maybe a dozen different statistics (speed, throwing power, tackling ability, etc) that determine his performance in the game. Other sports games do this, too, but for me Madden was the sports game for a long time.
Although it continues to be a standout game, I have to say that I did ultimately fall out of love with Madden, to the extend that this year — for the first time in maybe fifteen years — I didn’t buy the new version. Mostly it has to do with the complexity of the game. If you want to be really, really good at this game it takes a very long time to learn all of its nuances. I mean, it’s incredible that you can control things like stunts on offense and shading for individual defenders, and the incredible level of granularity you have with your pre-snap adjustments. (It’s my understanding that many elite players basically call the same few plays over and over and do all of the adjustments before the snap.) In reality, no single person has the responsibilities that a Madden player has, and eventually it got to be too much.
There are some other reasons that I lost interest in Madden, but one that I’d like to speak about here is one that I hadn’t thought much about after giving up the game until recently. With “artificial intelligence” all the rage now (though more and more it is more accurately being called “natural language modeling”), I began to try out Chat GPT and was reading a lot about its potential for generating content (like, I suppose, a blog post, yikes!). The more I learned, the more I began to remember Madden, and here’s why…
One of the disappointments I had with the game is that, for all its complexity, it cannot model the real world in one important way: imagining things we haven’t encountered yet.
Say my team has just traded for a backup quarterback who has been languishing on some other team’s roster, holding a clipboard and never seeing any game action. I happen to know this QB and I think he’s a diamond in the rough, far more talented than anyone gives him credit for. On my team, in combination with the other players on offense, he’s going to be magic: far better, in fact, than the starting QB we have now.
So, do I shuffle the Madden roster so that the kid can lead the team? No chance! Because the game may be called Madden, but the governing ideology comes from Bill Parcells, namely: “You are what your record says you are.”
If your player’s stats are lower than some other player, then they’re not going to perform as well. Period. Now, there is a little bit of nuance there, because players are a combination of stats and not one single number (though there is a stat called “overall” that does a good job of summarizing), but nevertheless, it’s a straight-up numbers game. In some years, Madden’s publisher tried to get around this by introducing the concept of hot and cold streaks, as well as randomly giving some players a little bit of a boost or handicap (called “nerfing” in game-speak) to make it a little more mysterious who might be the best pick for any given situation.
I do think these were great ideas, but they generally only applied to single games, so without any more information it would still be unwise to start a player that you just felt was underestimated over one that you might feel is overrated. In Madden, you are what your record says you are.
So, what’s the connection to my Chat GPT adventures? I was having lunch last week with a friend who works for a company that, among other things, is working on detecting ai-generated content, and he shared with me some of the characteristics of the work product of these applications. It occurred to me that the great weakness of these models is that they are unable to actually create anything. They can only put together combinations of things that have already been done.
Now, it’s true that many (if not most) inventions are just combinations of prior inventions and that very few things are created completely out of whole cloth. In that light, perhaps the ai applications limitation that it can only put together elements that it has ingested from prior work doesn’t seem like such a problem. But the thing is that the ai programs don’t really understand the things it’s combining. Worse, it creates the illusion of intelligence by combining the most commonly combined things. Ask Chat GBT a question, and it will probably give you the right answer, not because it actually knows the answer but because it can see what the most common answer is and give you that, assuming (and trusting) that is the correct answer. Write enough articles (or, say, post in the comments of your favorite blog) asserting that the Earth is flat, and these chatbots will ultimately agree with you.
I have no idea how Madden could possibly solve the “diamond in the rough” problem it has, absent the invention of a Time Machine. The developers already face a huge problem when they try to calculate the stats on rookies (ie: that franchise quarterback you just drafted but hasn’t played a down of football yet), with equal numbers of people complaining that the ratings are too low or too high. Likewise, I don’t really know how computer programs are ever going to truly create things that are not just other things put together (and, worse, put together in the most “average” way possible). What I suppose is important now is that we recognize that they can’t, we stop expecting them to, and we turn our attention instead to the things that they can do well. What I’m saying is, when it comes to the ai applications, this is not a miracle advancement.
They are who we thought they were.
Leave a Reply