One of the options in the game is to simulate the election cycle. The player just chooses two candidates and sees how they do. Now, you won’t get the exact result every time, but it’s fun to see how a given candidate will do if run by the AI. This mode helps us test against any unintentional biases in the issues.
Since no one knows how the election will go, it can be difficult to draw the line between when candidate A is simply more viable than candidate B. In the 2016 version, we had to issue a patch. You see, Trump kept winning and that was impossible. Goes to show what we know.
So let’s take a look a Sanders vs. Trump using the current data set we have (and we will be continually updating the data set between now and release).
Run #1: Sanders vs. Trump (AI vs. AI)
The stats we give to each candidate are subjective. But since our goal is to make the game as accurate as possible, we do our best to be fair to each.
From here we can see what mistakes the AI made. For one thing, Bernie not even visiting New Hampshire is a glaring error. Most of the time, the errors the AI makes are ones of taking certain states for granted. Also, in this game, the AI did not focus on Medicare for All.
Run #2: Sanders vs. Trump (AI vs. AI)
Fundamentally, the AI players will make different choices each game. They are the choices they think, at that very moment, will help them win but because each game is a little different (town halls and such come up in different places and we have a margin of error on the various issue data that we use to adjust the issues from game to game slightly.
Let’s see what happens game 2.
Mid-campaign in game 2. One thing that the game concludes is that “Medicare for all” is a losing issue in Florida. When he picks that issue, Florida goes red.
We dug into this and (and again, this is a game) but seniors in Florida tend to have higher than average incomes and don’t tend to favor Medicare for all. We will have to get more information on this.
Anyone into politics knows that these results are impossible. There’s no scenario where Trump is going to get Massachusetts or New Hamphire. So what’s the deal? It is a simulation issue or an AI issue or both?
In both cases, it’s an issue with AI. Bernie just isn’t visiting these states. Same with Wisconsin. So then the question is, why isn’t he visiting those states?
Run #3: Sanders vs. Trump (HUMAN vs. AI)
So this time, I’m playing as Sanders.
I also pick Amy Klobuchar as my running mate.
Playing as Sanders, I’m beating Trump pretty effectively. However, I’m also not running my campaign as Sanders probably would. I focus on Wealth Inequality, the Environment and Diversity. I stay away from talking about medical insurance or college loans or guns.
End of game: Sanders victory
So what did I do differently than the AI?
For one thing, I didn’t ignore my core states. I took advantage of AI Trump ignoring Oklahoma and edged Trump out there. But the big thing was focusing on the wealth gap issue and addressing climate change and not so much on Medicare for all.
In the real world, Sanders won’t get Oklahoma. But he didn’t actually need that to win. He didn’t even need Florida. I did manage to sneak Tennessee and Kentucky away from AI trump because, again, he didn’t start to visit there until it was way too late.
What we are learning
Trump is a difficult candidate to beat. This is mostly because he puts certain states in play that most Republicans would not. While I won Pennsylvania, Wisconsin and Michigan, I had to put quite a bit of effort to do so. I had to give up Florida to ensure victory in those states.
From a game point of view, it all means there’s lot of replay value. All the game’s key stats are in text files so players can play around with the numbers all they want and share them with others.
In the meantime, we continue to get more and more data and the team continues to improve the AI.