The claim that the AI has been recently changed to target healers, or weak heroes, or Wu Kong, or key heroes preferentially isn’t a new claim. The AI becoming suddenly smarter was first suggested back in mid 2017.
Popular theories of the AI’s new targets include:
“Key” hero ( although how it determines who is “key” is never clear)
What is fascinating is the frequency with which people claim that this is “new behavior.” The healer-targeting logic appears to get reintroduced with almost every patch, judging by how often it is claimed as new behavior.
July 2017 (AI targeting healers is a new thing):
July 2018 (targeting healers):
August 2018 (Wu, then healers,and it’s a new thing):
August 2018 (targeting healers, just starting to see this trend):
December 2018 (targeting Wu, and it’s a new thing):
December 2018 (raid defense getting smarter, targeting key heros):
December 2018 (weak heroes, plus targeting healers is a new thing):
January 2019 (targets talented heroes):
January 2019 (targeting weakest heroes with multiple hits on Atlantis, and it’s a new thing):
February 2019 (completely random targets):
February 2019 (targeting healers, and it’s a very new thing):
March 2019 (targets healers and key heroes, new in the last few months):
April 2019 (adapting to new strategy; didn’t originally target Wu, but suddenly does):
April 2019 (targets lowest health):
May 2019 (weak heroes):
May 2019 (targets healers, and it’s a recent change):
All true @Garanwyn but sometimes people who are time poor and/or in no mood to run multiple tests post on the forum to check if others have experienced the same or to be quickly refuted so that they can then go about their business. This process is particularly useful if the answer can be quickly obtained.
*This makes moderators’ jobs that much more mundane obviously
Sure. There are a few, like you, who asked if this was a real thing, or pointed out that it could be confirmation bias. You stated out with “Is it just me, or…?” @HolaQLeros also clearly said “Or it might be confirmation bias on my end?”
But the point is not at all to single out individuals, but to observe the pattern: people very frequently perceive a “new” change when nothing has actually changed. And they are often (but not always) very confident in making claims based on those perceptions.
This perceiving is a very human thing to do, so it’s silly to fault people for it. Where I do find fault is with the very strident claims that something definitely has changed, despite there being no substantive evidence to support the claim.
I’m gathering data to determine if a) there is bias in the target selection–initial indication is that there probably is, but I need a bit more data to be sure–and b) determine how the AI is biased. I’ll publish here or in a separate thread when I’ve got enough data to make an inference with a reasonable level of confidence.
Excellent! I look forward to seeing that. How the algorithm picks targets is a fascinating question.
My completely unscientific opinion is that it probably has some biases toward low health and high mana, but it’s clearly not an every time kinda thing, because I have recorded instances of the AI ignoring low health, low defense healers to pound on someone much sturdier (think: ignoring Sabina at low health and high mana to tag a full health Frida who has less mana).
Memory is a very fallible instrument, and mine is as fallible as anyone’s. Yesterday, I was 100% confident that I remembered one of my mother’s doctors had put her on a particular medication. And I insisted it was so–right up until she found the prescription proving that it was another doctor who had prescribed it.
What can I say? Evidence is king. Memory is shockingly sucky.
If the prevalence of propaganda which is wielded in an attempt to ignore or discredit evidence is the defining characteristic of “post-truth” in a particular era, we have been there before and undoubtedly will be there again.
One of my all-time favorite observations is that this was written over 2000 years ago, in Ecclesiastes:
“What has been will be again,
what has been done will be done again;
there is nothing new under the sun.”
Ever since I started playing I disagreed that the AI isn’t very smart. The AI has always seemed to target the right heroes and make fine decisions. The only thing is that Gravemaker might only hit 2 when he could hit 3. But yeah they usually seem to target key heroes, in raids my healers are first to die and against titans Wu kong will be the primary target until he’s dead lol. Not every single time obviously but often enough that when kiril is fully charged and has only half HP and the defending Lianna is fully charged, I know who she’s gonna hit.
Personally, I’ve found that in my typical raid squad, Proteus and Skittleskull get attacked more often than my maxed out Boldtusk, Evelyn, Caedmon, or Aeron. This leads me to believe that the targeting is likely HP based. Also, the number of times that two enemy AI target the very same hero who’s special JUST became available…
Those are two of my hypothesis: the AI views a hero with their special available as a threat, and thus attacks them… absent a threat (or if there are multiple special-threats), it then looks at HP. Might even look at HP + Defense? Who knows… but it does seem like weaker heroes get targeted first, as do those with their special ready to use. This is why Wu gets targeted, not because he’s Wu, but because of his low health compared to the rest of one’s team.
Response to the main point above
Agreed. People seem to make this claim without recognizing that this behavior is consistent with how it has been for as long as I’ve played the game.
@SWEG, @CheTranqui As I said, my unscientific opinion is that the AI is doing SOMETHING that isn’t just pure randomness. But I would agree with you both that whatever it is doing, it has almost certainly been doing it for a long, long time.
I’m not sure I take quotes from different opinions and time frames as evidence. ( there is a wide range and always opposing opinions at all times). I do agree the ai isn’t predictable 100% of the time meaning there isnt a single factor that chooses a specific target 100% of the time. I do think there is an “overall” quotient of “power” “wins/losses” that decides outcomes though.
Each time players play and their healers (or their Wu) don’t get targeted 5x or more in a row, means they just have to keep playing more, and the more data they put into the past increases their probability that next game their healer (or Wu) will get targeted 5x or more in a row. And since the past predicts the future with increasing certainly, then after seeing no results for the longest of times, it shouldn’t take much longer to eventually see it happen. I think I’m slowly getting the hang of this math thing.