How Human-like is Komodo 14 Human?

    In a recent post on my chess Blog I mentioned that Komodo 14 has a personality called Human which is “designed to play more like a human Grandmaster on a high level or like an amateur taught by a Grandmaster on lower levels.” The Komodo site explains that as a result “it may well perform better against most humans than the default engine on the same level.” 

    At the time of THIS game, Boris Baczynskyj was rated 2264 (Master) and I would expect that the Weighted Error Value scores would be lower when annotating with Komodo 14 Human, meaning that the closer to zero the score, the closer the player’s moves match the engine’s. 

    Analysis with Komodo Human showed Baczynskyj’s Weighted Error Value was 0.73 while Reshevsky’s was a precise 0.26, meaning that Reshevsky’s moves very nearly match those of the engine. Baczynskyj’s Weighted Error Value was 0.76 which is actually quite good. 

    One would expect the Weighted Error Value to increase when analyzing with Stockfish 16.1 simply because it is a stronger engine and humans would be less likely to select the same moves as the engine. 

    Using Stockfish 16.1 saw Reshevsky’s WEV to go up just a bit to 0.36, but Baczynskyj’s improved a bit to 0.69. I can’t explain it. 

    In a short match pitting Stockfish 16.1 against Komodo Human with a time limit of 5 minutes per game, Stockfish won 3-0. 

    In annotating one of the games (below) using Stockfish, it’s WEV of its own play was nearly perfect which is no surprise. What is interesting is that Komodo Human’s WEV was 0.72 which is pretty close to Reshevsky’s in his game against Baczynskyj.

    I am not sure what to make of all this except to sat that Komodo Human is clearly not close to Stockfish 16.1 in strength, but it appears that it does play at the Grandmaster level, but are the moves human-like? I can’t say. 

    To me this brings up the question, why pay for an engine when a stronger free program is available. Additionally, I do not see any benefit to the different Komodo personalities; they are, I think, sales hype. 

ChessBase

Komodo 14 Human–Stockfish 16.10–1C54G5PC2024Stocjfish 16.1

C54: Giuoco Piano: 4 c3 Nf6, main lines with 5 d4 and 5 d3 1.e4 e5 2.f3 c6 3.c4 f6 4.d3 c5 5.c3 a6 6.0-0 d6 7.b4 a7 8.a4 0-0 9.g5 9.bd2 e7 10.h3 c6 11.e1 g6 12.d4 is equal. Anand,V (2779) -Topalov,V (2740) Shamkir AZE 2019 9.a3 h6 10.b5 e7 11.e3 g6 12.xa7 xa7 13.a5 Black is better. Tharushi,T (1581)-Crocker,P (2135) chess.com INT 2023 9...h6 10.h4 g5 The position is equal. 11.g3 g7 11...e7 12.bd2 g6 13.c2 13.d4 d7 14.e1 f6 15.h3 exd4 16.cxd4 xd4 17.xd4 xd4 18.c1 de5 19.f1 e6 20.b1 ab8 0-1 (20) Dvoirys,S (2595)-Kaidanov, G (2629) Moscow 2005 13...g7 14.h3 h5 15.h2 hf4 Black is better. Jacimovic,S (1799)-Korban,K Herceg Novi 2008 12.bd2 e7 13.d4 h5 14.dxe5 xg3 15.hxg3 dxe5 16.h2 This is where white starts drifting into a bad position. 16.Nb3 keeps things nearly equal. g8 17.e2 f6 18.fd1 e7 19.b3 h8 20.f3 a5 21.xa5 h5 22.xg5? 22.d3 g4 23.a2-+ 22...h4-+ 23.gxh4 23.xf7 hxg3 24.xh8 g4+- 23...xh4 24.d3 g4 25.f3 h5 Weaker is 25...ah8 26.xh4! xh4 27.g3± 26.g3 ah8 27.e1 h7 28.f1 g5 29.h4 f6! 30.b5 30.b3 xh4 30...xh4 31.gxh4 xh4 32.g3 h2! 33.b6 33.f3 xf3 Xxf3 33...xb6 34.f3 xf3! 35.xf3 35.d1 f2+ 35...f2+ 36.g1 xf3+ 37.g2 xg3+ 38.xg3 38.xg3 38...f4+ 39.g2 f2+ 40.h1 f3# Weighted Error Value: White=0.72/Black=0.01 (flawless) 0–1

Download PGN