Watching commentary on the world championship match between Nepomniachtchi and Ding is a somewhat futuristic experience. It has three human commentators, but also has an AI commentator usually referred to as “The Evaluation Bar”
The commentators - incredible chess players themselves - could sometime tell if a move is bad or good immediately by themselves. But they would often react to a move because of an adjustment in the evaluation bar: “Wow, what a mistake! Wait… why is that a mistake?”
They would then go on to explore the position and find the mistake pretty quickly. But the players playing the game - the best players in the world - often didn’t, even after looking for the best move for quite some time.
Turns out simply knowing there is a winning variation helps you find it. This is of course nothing new - any puzzle is inherently easier than solving real world problems because you know there is a solution. But it was cool seeing it in action so bluntly and also added a lot of action and drama to the experience. “Will he find this move?!” was often asked by players who wouldn’t have known that the move exists without the eval bar.
Random thoughts on the Eval Bar
It would be cool to have this principle incorporated to real life puzzles. By removing the certainty of a solution we could make solving puzzles more challenging or “life like” (e.g “Here are some tasks you need to solve, some have a solution and some don’t. Or maybe they all do… you don’t know. You have a limited time. Go ahead and try to solve as many as you can”).
In chess the Eval Bar only really moves when a player does a mistake. By default it presumes players make the best move. In some positions that is very unlikely which is why taking it as is if you are betting on the result of a game is not enough - it isn’t aware of the “humanness” of moves and does not take into account the likelihood a good move will be found.
It will be interesting to see two eval bars in the future - actual score, and predicted score based on the “humanness” of moves.I could see an Eval Bar for different sports in the future, even those that are not in a “closed system” like chess. Soccer could have a dynamic bar based on the position of the likelihood of a goal for example. The key is could it show moves that are not easy to find for humans, but are possible to find: “Wow, he had this pass, that would have lead to a 89% for a goal”. For that we would need game simulators for those games, which is harder for sports that aren’t as tightly organized as chess but not impossible.
Taking that idea farther, since the Eval Bar is such a powerful minimal interface, what else could an eval bar be used for?
Eval Bar in Life ideas:
A dynamic Eval Bar that monitors your behavior and is aware of your main goal. Updates dynamically based on the actions you take, and how you spend your time. it would combine the distance of your current state from your goal and calculate the probability of you actually going through the actions to achieve it.
One nice concept in the eval bar interface compared to showing “percentage of success” is that it has two sides, you and an enemy. Having an enemy you fight against could be a cool concept to nudge you and personify probability. It could be used to replace existing probability interfaces in some places.
There are probably more ideas I am missing here. Chess is ahead of the curve here as the first sport to have an AI evaluation that works quickly and readily available, but I suspect it won’t be the last.