NN not managing to fit training data correctly

Question

I used the github repo https://github.com/suragnair/alpha-zero-general to get AI to play Shiftago(https://www.amazon.de/WiWa-Spiele-790023-Empfehlungsliste-Gesellschaftsspie