I mentioned a few months back, near the start of the season, that I was concerned about the galloping pace of AI advance being bound to ruin our little game of Fantasy Premier League before long. I was finding that Google's now infuriatingly AI-forward search function repeatedly offers to help me pick my team whenever I ask anything FPL-related.... and I'm sure many FPL managers, alas, will not be strong enough to resist such offers.
Well, I thought I'd give it a whirl over this international break. Just out of curiosity, you understand; just to see how bad it is. I wouldn't dream of following any of its suggestions. (The greater danger for me is that I would feel unable to accept any of its tips, and thus might be disadvantaged if it said anything sensible that I'd already been thinking of doing....)
And this screencap at the head of this post is what it came up with - supposedly working on the basis of a Gameweek 11 Wildcard selection with an available budget of 102 million.
Well, this is actually about its third go. Its opening attempt was so blandly 'template' that I derided it for just copying what was being most commonly shared on the forums rather than actually doing any analysis. (It duly confessed to this; but, of course, the only strong and consistent thread in its behaviour is to flatter the user with constant praise and affirmation, so that might not mean anything....)
I'm also not sure how closely it's actually sticking to the budget assigned; it may be 'more difficult' to access current data (surely not impossible; but it's not very resourceful in its problem-solving, poor lamb), so it seems to be leaning on the original pre-season player prices. Its 'awareness' is also somewhat loosely anchored in time: its grasp of which season we're in, let alone which gameweek, can sometimes be a bit shakey. At one point, for example, it was trying to recommend Van Hecke to me as a 'Brighton double-up' - apparently unaware that the Dutchman had transferred to Spurs over the summer.
It also seemed inclined to pay little or no attention to recent form or upcoming fixtures, the looming risk of suspension for players who've racked up a lot of yellow cards already, or the potential impact of injuries not just on the players ruled out but on the teammates around them. After I'd schooled it on a few of these points, it began to 'notice' potential issues like Raya and Gabriel being 'overpriced', or Konsa being at risk of being benched when Saliba returns from injury, or De Cuyper perhaps being hit by minutes-sharing when Mitoma is back in play, or Manzambi and Joao Pedro still having injury flags against them - and whether that might have a positive or negative impact on the prospects of players like Jackson and McGinn, Palmer and Rogers. But even when it acknowledged these factors, it didn't always seem to pay them any heed - and I was not convinced that there was any real analysis behind any of its selections.
So.... here we have it sticking with Rogers - despite having earlier seemed to prefer Palmer, but then having admitted that probably neither of them were worth having until Joao Pedro has returned. We have it doubling up on Vuskovic and De Cuyper, despite having acknowledged that double-ups are risky and that De Cuyper's minutes might soon be under threat. And we have it doubling up at the back also on Spurs, despite admitting that they've so far established no convincing or consistent form (being decent at the back isn't good enough if you can't score any goals...). And it's gone for Wirtz and Rudoni - WHY??
It doesn't look to me as if this AI has any clue how to play the game as yet. Others, Claude and ChatGPT and so on, are probably better; though I doubt if any of them are good enough to pose a serious challenge to the best human players yet. But in a way, they can perhaps be even more harmful when they're bad. Being bad won't stop lazy FPL managers leaning on them; and having large numbers of people follow lazy, bad advice - that's largely shaped by what lazy, bad managers who constantly look for tips online are doing - is only going to further narrow and ossify the dreaded 'template'.





