Jake

Idle Thumbs 161: The Eyes of Luigi

Recommended Posts

I guess I should also play Wolf, if I get the time.

 

You should, that's a good computer game!

 

3566-wolf-dos-screenshot-an-ingame-scree3568-wolf-dos-screenshot-the-wolf-select

Share this post


Link to post
Share on other sites

My android phone is usually surprisingly good at taking dictation, but the other day I was trying to set up a grocery aisle category with the name "ethnic" and I couldn't get the phone to recognize that word. It suggested "Catholic" and "attic" and then when I tried to say the word very slowly and distinctly it typed: "F Nick".

 

With that exact capitalization. I had never written or spoken that nickname to the phone before. 

 

I think that since we've now had over 100 episodes since Nick ruined Idle Thumbs forever by taking the Bethesda job, we could retire the "F Nick"name, but "Ethnic Breckon" is probably too good to resist.

Share this post


Link to post
Share on other sites

I think that since we've now had over 100 episodes since Nick ruined Idle Thumbs forever by taking the Bethesda job, we could retire the "F Nick"name, but "Ethnic Breckon" is probably too good to resist.

 

It's been said before, but countdown to an offensive Photoshop beginning... now.

Share this post


Link to post
Share on other sites

Anyone else tried the forest?  I thought I'd wait after Nick's impressions but Sean Elliot makes the AI sound fascinating:

 

Tremendously promising approach to AI behavior in The Forest.  Effective combat tactics reward experimentation and observation, not necessarily sustained direct damage. You discover what might ward a mob off through thoughtful play. Standing down a feint, for instance, seems to shake some. Parrying an alpha's attacks might wear at the party's resolve more than aggressively attacking the mob. Indeed, I've discovered several ways to discourage a sustained assault that supports greater nuance than simply attempting to flee. But you will overlook all of these by blindly attacking. The combat system can privilege de-escalation. Additional examples of conflict avoidance include creating a totem with sticks and dismembered cannibal parts to attest to your strength.  AI in The Forest sometimes stalk you, again rewarding awareness and ratcheting tension.  Because their behavior differs depending upon time of day, gender, etc., it encourages players to guess at their psychology.  Correct or incorrect, any theories we form on the cannibals' psychology contributes to our perception of its sophistication.  And few games achieve anything more than firm player grasp of hard and fast AI rule sets. Importantly, some AI seem to subvert the rules of thumb we arrive at... which keep us guessing and cause us to amend the theories of (rudimentary) mind we form. Not that I've parsed it all yet. At times, AIs engage in primitive ritualistic /religious behaviors. The thing is, The Forest is always suggesting there's something to learn through study. All that said, I wouldn't recommend purchasing The Forest Alpha at this time. (Unless you like paying to play QA.)

https://twitter.com/shawnelliott

Share this post


Link to post
Share on other sites

A lot of the discussion about what we want from artificial intelligence has been about the decisions the bots make when they fight you, but I also find the potential of how they react in non-combat scenarios fascinating. I'm the the player who plays stealthily just so I can find out what the inhabitants of a cave in Skyrim are doing when they supposedly don't know that an adventurer has wandered in. I love watching Skyrim's village NPCs have their awkward exquisite-corpse conversations. I remember well enough, what playing computer-games was like before I understood that A.I. typically is just a short list of conditional states in games. I used to think that all the bots were actually acting based off of their past experiences and hypothesizing about how best to accomplish their goals. That's an illusion that I seek out in games. I don't know how Civ 5's A.I. works, but I like to think they they hold grudges and will pander to you with a future back-stab in mind. In some cases, I like thinking that some of the leaders value trade over territory gain and build intentional relationships with other non-player leaders with those plans in mind. If this is a case, it's not a learning behavior as much as it is having a few baked-in strategies and goals within a system where they consider history or memory. The game I want to play is were the bots actually learn and adapt to each other's strategies so that I can watch it occur and manipulate their relationships. The best example I've seen of this is an iphone game not coincidentally named Artificial Life. In it, you have hundreds of paramecium-bots exchanging behaviors by mixing the conditions and methods into their children when mating. Over time, the larger populations can be seen learning as individual members with sucessful strategies are more likely to procreate. Colonies form and the most exciting moments are when a highly developed colony first begins to come into contact with another that hasn't mixed for many generations. The biggest weakness of this not-game is that examining the reasoning in this genetic-algorithm A.I. is very difficult; you have to have been paying attention to the behaviors that were expressing themselves the most during various generations to gain an appreciation for why they are reacting to each other the way they are. Ideally, if this type of system was in a game like Skyrim, the player would be able to interrogate an NPC to follow their chain of reasoning.

For fantasy-instance, you may see a milk-maid try to kill a shop-keeper.

You could ask her "Why are you doing your current activity?"

She responds with [These conditions were met].

You could ask "Why do you have [those conditions]?"

... and so on.

I'd love to play that. Especially if I could negotiate giving them a new activity in exchange for completing their current goal. "I'll bring you food if you clear out that dungeon for me."

Share this post


Link to post
Share on other sites

A lot of the discussion about what we want from artificial intelligence has been about the decisions the bots make when they fight you, but I also find the potential of how they react in non-combat scenarios fascinating. I'm the the player who plays stealthily just so I can find out what the inhabitants of a cave in Skyrim are doing when they supposedly don't know that an adventurer has wandered in. I love watching Skyrim's village NPCs have their awkward exquisite-corpse conversations. I remember well enough, what playing computer-games was like before I understood that A.I. typically is just a short list of conditional states in games. I used to think that all the bots were actually acting based off of their past experiences and hypothesizing about how best to accomplish their goals. That's an illusion that I seek out in games. I don't know how Civ 5's A.I. works, but I like to think they they hold grudges and will pander to you with a future back-stab in mind. In some cases, I like thinking that some of the leaders value trade over territory gain and build intentional relationships with other non-player leaders with those plans in mind. If this is a case, it's not a learning behavior as much as it is having a few baked-in strategies and goals within a system where they consider history or memory. The game I want to play is were the bots actually learn and adapt to each other's strategies so that I can watch it occur and manipulate their relationships. The best example I've seen of this is an iphone game not coincidentally named Artificial Life. In it, you have hundreds of paramecium-bots exchanging behaviors by mixing the conditions and methods into their children when mating. Over time, the larger populations can be seen learning as individual members with sucessful strategies are more likely to procreate. Colonies form and the most exciting moments are when a highly developed colony first begins to come into contact with another that hasn't mixed for many generations. The biggest weakness of this not-game is that examining the reasoning in this genetic-algorithm A.I. is very difficult; you have to have been paying attention to the behaviors that were expressing themselves the most during various generations to gain an appreciation for why they are reacting to each other the way they are. Ideally, if this type of system was in a game like Skyrim, the player would be able to interrogate an NPC to follow their chain of reasoning.

For fantasy-instance, you may see a milk-maid try to kill a shop-keeper.

You could ask her "Why are you doing your current activity?"

She responds with [These conditions were met].

You could ask "Why do you have [those conditions]?"

... and so on.

I'd love to play that. Especially if I could negotiate giving them a new activity in exchange for completing their current goal. "I'll bring you food if you clear out that dungeon for me."

 

I suppose that's what they were going for with the "Clementine will remember you said that" prompts in Walking Dead. I imagine it would work better in a non-story focused game, or at least a non-pre-written (rather than emergent) story game.

Share this post


Link to post
Share on other sites

FEAR actually uses a different technique for its AI than is typical for an FPS - it's called goal-oriented action planning.
 
Essentially, most FPS games use finite state machines (for example, Halo 2) where the AI idles/attacks/retreats, etc. where each state has a set of actions it will attempt, and triggers that will transition itself from state to state.

 

Goal oriented AIs instead have a specific result they aim for (say, the player's death) and then path-find backwards through their possibility space to their current state, forming a chain of actions that it thinks might achieve the result it wants.

 

My understanding is that the major reason why this sort of AI system isn't more widely used is that it's a huge hassle to debug or sculpt for specific encounters. Also, it's very hard to see the effect in most games, as there needs to be a wide, high-dimensional, interesting possibility space.

Share this post


Link to post
Share on other sites

That video is so good. Luigi just can't catch a break, huh?

 

I don't know if this has been linked yet, but this is a blog post about how it's a myth that players want intelligent A.I. The game dev describes a scenario in a first-person shooter of a player vs. a squad of three 'intelligent' A.I. enemies. One A.I. enemy would draw the players fire, while the other two flanked and attacked the player from the side. Players wouldn't end up seeing the flanking enemies and when they took damage, they assumed that the game was cheating by spawning enemies behind them.

 

The devs ended making a number of changes (adding barks when flanking, etc) all of which are described in detail in the blog post. Ultimately they concluded that A.I. should be "fun" to beat, and a focus on enemy variety may be more useful than better A.I. intelligence.

 

The blog post echos what many have pointed out in this thread already, that many players only want enemy A.I. to appear smart/reactive.

Share this post


Link to post
Share on other sites

Your comma troubles me Twig.

I can't imagine, why.

Share this post


Link to post
Share on other sites

I'm kind of amazed that he could ride that Kart, because the weight limit for those things is only 55 pounds.

Share this post


Link to post
Share on other sites

I love how almost everyone is like "what the fuck, br- oh, it's Luigi death stare, awesome!"

Share this post


Link to post
Share on other sites

It bothers me that he's looking at the camera and not at the person he hit.

Share this post


Link to post
Share on other sites

Holy shit, Goonies 2!  I loved that game when I was a kid, never finished it though, got stuck somewhere in it. 

Share this post


Link to post
Share on other sites

FEAR actually uses a different technique for its AI than is typical for an FPS - it's called goal-oriented action planning.

 

Essentially, most FPS games use finite state machines (for example, Halo 2) where the AI idles/attacks/retreats, etc. where each state has a set of actions it will attempt, and triggers that will transition itself from state to state.

 

Goal oriented AIs instead have a specific result they aim for (say, the player's death) and then path-find backwards through their possibility space to their current state, forming a chain of actions that it thinks might achieve the result it wants.

 

My understanding is that the major reason why this sort of AI system isn't more widely used is that it's a huge hassle to debug or sculpt for specific encounters. Also, it's very hard to see the effect in most games, as there needs to be a wide, high-dimensional, interesting possibility space.

 

This sounds very similar to the way rudimentary chess AI works. Each decision the AI or its opponent makes is a branch in a decision tree. Each decision the AI makes is evaluated via a heuristic (aka estimate-maker) to provide a weight to that decision based on how much it improves the AI's chance at winning. Then its repeated as many levels down the tree as the computer/difficulty setting can handle, and the AI picks the best (highest-weighted) outcome.

 

FEAR's sounds like the reverse of that, starting from the death of the player and working backwards in the direction of the current state of the game.

Share this post


Link to post
Share on other sites

Maybe the next FEAR game should have you be unwillingly telepathic so you always hear the enemies' thought processes.

Share this post


Link to post
Share on other sites

Maybe the next FEAR game should have you be unwillingly telepathic so you always hear the enemies' thought processes.

In slow motion.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now