Stealth Change on AI's behaviour?

Pylgrim wrote:

Playing Thick a Thieves I've started noticing a much increased occurrence of the opponent sitting on AP when it's available for a cheap ability, if there's a more expensive ability in the team. This is something that used to happen very rarely but in the games I've played during the last day, for example, Cmags has several times avoided using Magnetic Translocation as soon as he got 10 AP allowing IM35 to eventually (thanks to goon AP) cast Ultra-Freon Beam. Other examples: Ares sometimes refusing to cast Onslaught to allow the Empiricist cast Merciless Crush, Venom missing on Devour to let Yelena do Payback Mission and not using Symbiote Snare in Favour of Moonstone's Gravity Warp.

Anyone has noticed this too?

They always do this, but due to the low % they have toward saving for the bigger move, what usually happens is the AI invariably uses up AI for the cheaper move long before they hit the more expensive move unless they're force fed AP.

Case in point, let's say there's a 75% Magneto will defer to IM35 for Ultra Freon Beam (it's almost certainly a lot lower than this based on experience). We know the AI has no particular priority on the color it has an active ability in, so in that particular matchup they have red, blue, yellow. Let's say the AI is equally capable of getting any of those 3 colors, then it'd take 9 turns for the AI to go from 10 AP to 19 AP to use Ultra Freon Beam. The chance of the AI saving for 9 turns in a row is (0.75)^9 ~= 7.5%.

On the other hand, in the actual node we have AI getting at least 2 blue AP a turn. Let's say the AI doesn't even make a blue match (and also doesn't spend it on Cooking the Books). It takes at most 5 turns to hit 19 AP, so the chance of that is (0.75)^5 = 24%. Realistically, if the AI also made any blue matches then it'll hit that in 3 turns (10 + 3 + 2*3 = 19), so it could be as high as (0.75)^3 = 42%. Now the % the AI saves up for the more expensive move is anyone's guess, but I hope this shows how the chance of the AI end up having enough for the bigger move is much higher when it has someone to force feed AP.

Vhailorx wrote:

So it seems that AI has a hierarchy if there are more than one abilities available on particular color. So sometimes it does eschew lower cost powers to save up for higher cost powers (though it typically goes the other way).

The thing that's harder for me to figure out is what factors might make the hierarchy change. Obviously, if the character with the higher cost skill dies, the lower cost skill moves up in priority. So if the AI is saving red to bast photon blast, and you kill moonstone, then that red might immediately shift down to headbutt the next round. But I have also seen the ai wait to cast something until I leave a squishy character out front. I don't know if that is coincidence, or if the ai factors things like target hp.

Nor is it clear if there is any logic in which abilities the ai will choose to prioritize (I would guess it's random). some have speculated that the ai just uses enemy position and ability order to prioritize, but I can't confirm that.

First of all there are some completely inexplicable cases where the AI doesn't use an ability at all when it can. This seems to usually happen after you down someone. It's usually pretty easy to see in say a Muscle/Thug/Thug formation where you downed the Muscle and see that the enemy has 30 yellow and yet the next turn it'd choose to do nothing despite having more than enough AP. This very rarely happens to non goons but I have no idea what triggers it, but it seems to be related to downing someone.

But overall the AI evaluates ability like this:

For a given color, if they have enough to use the most expensive ability, they'll use it. If there's a tie for the most expensive ability (Thunderclap or Headbutt), it appears to be random which one they use (this is practically impossible to test reliably).

Otherwise, if they have enough for a lesser move but there's one that can be saved up, they'll evaluate the chance to save up based on (higher AP cost move's AP cost - lower AP cost move's AP cost). The higher this number is, the more likely they'll save it. So for example, Ultra Freon Beam at 19 versus Magnetized Particles at 10 gets you 9, which is a relatively large difference, so they're more likely to save up.

Note that if there are multiple abilities on the same color, this check occurs multiple times. That is, in the IM35 + Magneto + Don node, if the AI has say 10 AP, it's going to evaluate Cooking the Books (8) and Magnetized Particles (10) separately. Depending on what numbers it rolled, all these 3 outcomes are possible:

1. Use Cooking the Books. (good for you)
2. Use Magnetized Particles. (generally very bad)
3. Use neither. (probably okay)

It's important to remember that the without an external source of AP, the % basically works out that saving up for the more expensive move is almost never possible. This is what people think as the 'AI is waiting for OBW/The Hood/whoever to be in front". All that happened is the AI happened decide to save up a few turns before giving up and happened to hit whoever you're trying to protect. In fact, if the move they're trying to guess is not going to KO whoever you need to protect, I recommend just ignore trying to protect the character at all because you're often passing up a better move while the AI is choosing when to use its move completely at random.

Pylgrim wrote:

I know all this. What I am trying to say is that it seems as though they raised that %. (Definitely not 75%, I'd say it was more like 25% then and it feels like 60% now) I have played TaT every time since it was created and I had never got Ultra-Freon Beam'd this many times. I also noticed it with other characters. Not getting Onslaught'd before the AI got Merciless Crush AP was extremely rare. Now it's happening quite frequently.

The chance of them saving up at all is always very small even with AP pump so each person's experience may difer wildly. I think I saw Ultra Freon Beam used twice in this event (without having Magneto down), and that seems awfully high, but certainly within what can be explained by luck. By the way, that % I gave is per turn. Remember, the AI re-evaluates its options every turn unlike a human. If you're making the same decision and opted to save for Ultra Freon Beam under the identical situation, you wouldn't immediately change your mind next turn, but the AI does. If their base chance is 50%, then you're looking at 0.5^9 = 0.2% normally, 0.5^5 = 3.1% with a 2 blue AP pump but no matches, and 0.5^3 = 12.5% if they made a blue match after having enough for Magnetized Particles. The first two are effectively 0%, and if this time around the AI just happened to make a blue match after having 10 blue AP, bumping its chance to 12.5% and get a bit luckier than usual, you can easily get hit by Ultra Freon Beam more than all your previous playthroughs put together. I'm pretty sure the two cases I took an Ultra Freon Beam indeed has the AI making a blue match after they've 10 blue AP, but if you're not keeping track of these things you might not be aware that making that extra blue match greatly increases the chance of having enough for Ultra Freon Beam.