Is lower level AI sound play with random blunders added?

My ChessUp finally arrived yesterday and I love it. First impressions -
It seems to throw games away on the lower AI settings. I was ahead in position but level in material. So it was playing close to my standard (not great) then just blundered its Queen in a really obvious error.

Hi - Are you playing against the Ai with the app connected? Or playing against the board without the app?

No, just using the board’s AI.

1 Like

I don’t know how the board’s internal AI is implemented, so I can’t comment on that.

The phone apps use Stockfish, which does have a known method of “strength modulation”. Instead of looking for the single best move, it establishes the four best moves and their evaluations. It then adds random numbers to those evaluations, with ranges dependent on the strength setting - for a very low ELO rating, the randomness could be quite large. The move played is then the “best” move after those random perturbations were applied. There may also be limits on opening-book depth and endgame tablebase knowledge.

Different engines may have other methods of achieving a similar result. The usual method in the older dedicated chess computers was to change the time control, as the computer was only about as capable as an amateur player anyway. Rodent riffs on that by limiting the rate at which it can search positions.

I’ve tried with the app. It makes a difference, definitely better. Level 6 assist feels like cheating but it is instructive.

The embedded app on the lowest level may lose pieces, on the lower remaining levels may lose exchanges or miss tactics. On the higher levels it will make more strategic mistakes.

When using the app, we do not use stockfish’s built in levels - because they don’t actually work. They all play very strong and to make stockfish play lower it takes additional code (which we have done)