The AI as a model of human managers

[TEHLUDDITE]

First off kudos on this release. I found some enjoyment in FHM2014 but FHM2 is a big leap forward on all levels. As you read further, please keep in mind that I think the devs have done a great job, generally, in programming the AI. I am starting to wonder, though, whether the future development of the AI for FHM won't require more of an overhaul than some tweaking.

I've been tinkering with the game, controlling Kingston (OHL) and have been goofing around with the trade AI. Here's what I was able to get the AI to agree to at the OHL trade deadline around Jan.8, 2016:


Ottawa gives up Travis Konecny for the rights to Sam Bennett plus Kingston's 2016 draft picks in rounds 12 through 15.


Both Konecny and Bennett are NHL drafted and signed. Bennett (19) is in his final year of CHL eligibility (not counting overage) while Konecny (18) has a year remaining after this one. Bennett is playing currently with Calgary in the NHL. Konecny is with Ottawa in the OHL.

I would have hoped that the AI would know that there was no chance of ever having Sam Bennett in the OHL again (Bennett has exceeded the games-played threshold this season for being returned to junior, having played 40 NHL games to date). Similarly I would have hoped that the AI would value Konecny very highly since he is a highly effective OHL player with a little more than a season of CHL eligibility remaining but does not look ready, ratings-wise, to make the jump to the NHL. The likelihood of his return for his final year of CHL eligibility should make him a valuable property.

Making matters worse, the upcoming OHL draft has only 138 players listed as eligible. At 20 teams per round, that makes less than 7 full rounds before the list is exhausted. So the picks included in this trade are completely worthless. I think it goes without saying that the AI should know that.

It looks like the AI is evaluating this trade in a snapshot fashion, looking exclusively at the current situation and failing to project for the future. A human manager would do that as well but would also be calculating projections for the future of each asset in the trade. Presumably the AI's failure here to properly project the draft pick values is a solvable programming issue. Chances are the AI rejects this trade if it understands that the picks it is receiving have no value.

Still, for the series to eventually have the dynamic and effective AI we all want it will need to think more like a human. Human GMs have plans. My experience with the game is far too limited to say that the current AI doesn't do much or any planning, but that is just my early impression.

This could be an opportunity for the devs to utilize the community - gather feedback from users on how they evaluate GM decisions, step-by-step. Make a questionaire with a bunch of scenarios in order to gather the data. Build a list (or fine-tune the one that already exists) of heuristics from that data that can be used by the AI algorithm.

Apologies to the development team if this is already happening behind the scenes but this is a case where consultation with the community could prove very valuable.