Return to “Suggestions”

AI Goals

#1
Hey everyone,

What different kinds of goals/personality types will the AI have in LT? Make money is an obvious one, but what about Get Power? Bloodlust? What sort of things can we expect?
Live long and prosper.

Re: AI Goals

#2
Not sure if you're aiming to suggest this as a discussion or literally asking a question. If the latter, than I don't believe the details have been anything close to finalized, so we should make a discussion of it. ;)

Personally, I'd like to see a 'Great Builder' personality. Obviously dependent upon making money, but beyond that this would be an AI with a particular interest in grand scale projects, built to impress and amaze the galaxy even in the absence of said project being considered 'profitable'. (given that it's not likely to bankrupt itself, at least.)

Re: AI Goals

#3
This is the place where I mention my article on personality styles and gameplay modes. ;)

The short version is that a number of psychologists and game designers have seen a core set of about four basic motivations, which express themselves in games in particular patterns of play:
  • Power through Performance (kinetics): manipulating the world and the people in it to feel intense sensations
  • Security through Persistence (mechanics): following the rules to win the game and accumulate the most stuff
  • Knowledge through Perception (dynamics): exploring to understanding the complex world
  • Identity through Persuasion (aesthetics): interacting with people and stories to become more fully human
(Note: Individual people don't express just one of these; they are fundamental patterns of behavior that emerge when you look at what motivates large numbers of people. So targeting an individual gamer with a system like this usually won't work too well... but it can be useful when you're making a game that you hope a lot of people will play, because then you can satisfy the core motivations of play with targeted game systems.)

In other words, this is a suggestion to key NPC personalities to Power (action), Security (achievement), Knowledge (understanding), and Identity (personal meaning), and to key gameplay predicates roughly to Performance (Combat), Persistence (Trade), Perception (Exploration), and Persuasion (Diplomacy). That should IMO yield characters whose behaviors feel plausible -- they'll goal-seek (from their motivations) and act (based on gameplay predicates) in ways that feel recognizable to us.

So there's that suggestion about personality-driven AI for Limit Theory that I've been wanting to get off my chest. :)

NPC Characteristics

#4
In the Update 9 announcement thread, Josh wrote:
Josh Parnell wrote:...I will soon implement a "personality" factor that will change how each individual values certain actions. This will mean that the "optimal" course of action will become subjective, so each NPC will have his/her own idea of what the best plan looks like. These factors will include considerations like how much the NPC likes money ('greed'), how much it values its own life ('safe'), how much it values personal interaction ('social'), how much it is willing to expend other life ('aggressive'), etc. Definitely a place where we can have a lot of fun with the AI :)
I think it would be quite interesting to throw out some ideas of what these personality types could be.

There's been some discussion over similar ideas, though it was often at the faction level rather than for individual NPCs. The most obvious set of attributes that we can consider was Gazz's list of Naval characteristics. Josh also indicated some of his thinking with safety, greed and aggression.

I started making a list of attributes, but I wonder whether it would not be better to present a few sliders that bias the character towards different behaviours:

[Offensive]---------------|----[Defensive]
This would govern bonuses to attacking large numbers of ships or finding ways to avoid fights. This would be similar to Josh's Aggression and Safety attributes.

[Command]--------------|-----[Pilot]
This would govern bonuses to whether the NPC is a better as leader or a lone pilot. This could also be "Strategic-Tactical".

[Science]---|----------------[Engineering]
This would govern bonuses to tech research or blueprint development, as well as the capability of the NPC to perform repairs in the field.

[Business]--|-----------------[Military]
Not sure about this one, but the goal is to bias the NPC towards military action or business.

[Loyal]------------------|---[Opportunistic]
This would govern whether the NPC would stick to their faction instructions/plans or take opportunities when they are presented.

This list of five sliders should allow plenty of complex behaviour but still be easy to present to the user and understand.

Thoughts?

Re: NPC Characteristics

#5
If possible, I'd like to borrow from psychiatry and use one of the models out there (OCEAN (the Big Five), Myers-Briggs, etc). It would definitely fit more towards Josh's vision of increasing levels of abstraction.

Taking OCEAN for example:

Openness - desire for new experiences, adventure, new ideas
High - favors exploration, combat, research
Low - favors trading, mining, construction

Conscientiousness - self-discipline vs creative spontaneity
High - likely to make rational decisions in combat, pay bribes when threatened, stick to old trade runs
Low - likely to "run for it" when ransomed, ransom other ships, try new trade runs

Extraversion - sociability, obviously (the most important trait in psychiatry)
High - much more likely to participate in faction, enlist in fleets, cooperate
Low - much more likely to favor self-driven activities - mining, trading, pirating alone

Agreeableness - aggressiveness, obviously

Neuroticism - emotional stability. The most controversial Big Five trait.
Could be confused with conscientiousness, but reflects more of an internal state rather than an external one.

As an example:
Highly neurotic, highly conscientious - most traders on Wall Street
Low neurotic, highly conscientious - seasoned military veteran
Low neurotic, low conscientious - that drunk guy at the bar

Re: AI Goals

#6
Flatfingers wrote:This is the place where I mention my article on personality styles and gameplay modes. ;)

The short version is that a number of psychologists and game designers have seen a core set of about four basic motivations, which express themselves in games in particular patterns of play:
  • Power through Performance (kinetics): manipulating the world and the people in it to feel intense sensations
  • Security through Persistence (mechanics): following the rules to win the game and accumulate the most stuff
  • Knowledge through Perception (dynamics): exploring to understanding the complex world
  • Identity through Persuasion (aesthetics): interacting with people and stories to become more fully human
(Note: Individual people don't express just one of these; they are fundamental patterns of behavior that emerge when you look at what motivates large numbers of people. So targeting an individual gamer with a system like this usually won't work too well... but it can be useful when you're making a game that you hope a lot of people will play, because then you can satisfy the core motivations of play with targeted game systems.)

In other words, this is a suggestion to key NPC personalities to Power (action), Security (achievement), Knowledge (understanding), and Identity (personal meaning), and to key gameplay predicates roughly to Performance (Combat), Persistence (Trade), Perception (Exploration), and Persuasion (Diplomacy). That should IMO yield characters whose behaviors feel plausible -- they'll goal-seek (from their motivations) and act (based on gameplay predicates) in ways that feel recognizable to us.

So there's that suggestion about personality-driven AI for Limit Theory that I've been wanting to get off my chest. :)
Awesome! Thanks for that, Flat. I think it's a really nice and compact summary of personality dimensions :D :thumbup:

I think, in the end, the goal for all NPCs will be the same: "maximize value." IMO this is the goal that every rational agent follows. The key is that "value" is subjective and means different things to different people. In the demo I just made the goal "maximize money," which is basically making the assumption that everyone values money above all else. But if we incorporate, for example, Flatfingers' personality traits, we will see that for certain NPCs, maximizing the value of their state will involve forming new relationships with other NPCs, or acquiring more knowledge about the universe, etc.

So I hope that, in the end, the divergent behavior of the NPCs will be entirely controlled by their personalities, so that I don't even have to worry about giving them different goals!
“Whether you think you can, or you think you can't--you're right.” ~ Henry Ford

Re: AI Goals

#8
Flatfingers wrote:This is the place where I mention my article on personality styles and gameplay modes. ;)

The short version is that a number of psychologists and game designers have seen a core set of about four basic motivations, which express themselves in games in particular patterns of play:
  • Power through Performance (kinetics): manipulating the world and the people in it to feel intense sensations
  • Security through Persistence (mechanics): following the rules to win the game and accumulate the most stuff
  • Knowledge through Perception (dynamics): exploring to understanding the complex world
  • Identity through Persuasion (aesthetics): interacting with people and stories to become more fully human
(Note: Individual people don't express just one of these; they are fundamental patterns of behavior that emerge when you look at what motivates large numbers of people. So targeting an individual gamer with a system like this usually won't work too well... but it can be useful when you're making a game that you hope a lot of people will play, because then you can satisfy the core motivations of play with targeted game systems.)

In other words, this is a suggestion to key NPC personalities to Power (action), Security (achievement), Knowledge (understanding), and Identity (personal meaning), and to key gameplay predicates roughly to Performance (Combat), Persistence (Trade), Perception (Exploration), and Persuasion (Diplomacy). That should IMO yield characters whose behaviors feel plausible -- they'll goal-seek (from their motivations) and act (based on gameplay predicates) in ways that feel recognizable to us.

So there's that suggestion about personality-driven AI for Limit Theory that I've been wanting to get off my chest. :)
Let's see if I can map these to the Myers-Briggs personalities:

Power - ESFP/ESFJ - Performers and Providers - presidents, those with a strong moral compass, who call it 'as it is'
Security - ESTP/ISTP - Promoters and Crafters - learners and analyzers.
Knowledge - INTJ/INTP (the classical scientist types) - Masterminds and Architects - independent, creative, autonomous types
Identity - ENFP/ENFJ - Champions and Teachers - value harmonious relationships, enthusiastic, optimistic

That still leaves 8 or so personality types to consider. My own personality type for instance (ISTJ on most days) isn't on there, even though it "should" fit under Knowledge.

PS Oops, saw your article. Glad that we mostly agree.

PSS The OCEAN types actually do map reasonably well to the Myers-Briggs types. This table on wikipedia for instance:

Extraversion Openness Agreeableness Conscientiousness Neuroticism
E-I −0.74 0.03 −0.03 0.08 0.16
S-N 0.10 0.72 0.04 −0.15 −0.06
T-F 0.19 0.02 0.44 −0.15 0.06
J-P 0.15 0.30 −0.06 −0.49 0.11

The closer the number is to 1.0 or −1.0, the higher the degree of correlation.

Re: NPC Characteristics

#9
To provide a contrarian view, I would suggest that and actually appreciate it if the organization of NPC characteristics stayed away from the rigid classifications of pseudoscientific astrology like MBTI, Enneagram, and the works. I think personality dimensions or traits are just fine, you can even model them after the seven virtues and sins if you want, but I don't agree with the intuitive notion of grouping them by traits and creating distinct identities out of each, as is (especially) MBTI's wont.

INFJ, Socionics ILI here. I assure you, If I didn't think the whole stuff was a true waste of time, I would talk Shadow functions with you all day.

Re: NPC Characteristics

#11
A couple of quick comments:

jimhsu, my impression is that Big Five is useful mostly in a clinical setting for helping to understand people who are outside the norms. The Neuroticism index presumes some degree of brokenness. By contrast, MBTI emphasizes that styles aren't better or worse than each other; they're just different. I find that more useful for understanding regular people, as well as in a game design context.

That's not to say you couldn't design an NPC motivation system based on the OCEAN indices, or Dr. Oldham's 14-style model based on the DSM-IV, or some other clinical instrument. As long as it's fun for your target audience, it's a useful starting point for game design. My preference for the MBTI -- actually, for Keirsey's more generalized "temperaments" version -- as a starting point for a model of NPC motivations is because it's detailed enough to be useful but general enough to avoid excessively complicated systems that have to be modeled and tested as a game.

And alpan, at the risk of being branded as some kind of True Believer impervious to reality ;), I'll just say that I don't lump the MBTI/temperament model with things like astrology/enneagram/blood type. The former has the virtue of being a theory generated by analyzing data (aggregated results of standard questions about what actually motivates people) and identifying broad patterns, where the latter are theories that have gone looking for data to try to confirm them (and thus don't).

That said, I don't disagree with the idea that you could base the motivations of AI actors in a computer game on darn near any system, or even none at all (i.e., just an unordered list of high-level goals). I happen to like the four-temperament model for the reasons given above -- it's simple enough to be programmable while still (IMO) reflecting real and discernably distinct ways of seeing the world. If I were looking for a pure theory of personality, I might go a different route, but this is about developing some system that supports computer-based entertainment. Temperament is just a convenient starting point.

And I'm an INTP Architect, if that wasn't already embarrassingly apparent. :ugeek:

Re: NPC Characteristics

#12
...I would suggest that and actually appreciate it if the organization of NPC characteristics stayed away from the rigid classifications...
...I don't agree with the intuitive notion of grouping them by traits and creating distinct identities out of each...
I am largely with Alpan here. While I would very much like distinct NPC personalities, I would prefer if said personalities emerged as much (if not more) from probabilistic outcomes as strict categorization. In my view in a sandbox an element of randomness (unpredictability, if you will) in NPC behavior is not merely refreshing but almost a necessity for extended playability.
I know not what life is, nor death.
Year in year out-all but a dream.
Both Heaven and Hell are left behind;
I stand in the moonlit dawn,
Free from clouds of attachment.

Re: NPC Characteristics

#13
I definitely do support a more stochastic approach towards defining the characteristics of NPCs -- your actions should define your type, not your type to your actions. That said, personalities (not necessarily MBTI, but whatever system) serve both a) as a way for the player to identify NPC as potential allies/customers/adversaries, as well as b) other NPCs with personality types that are the same or different to evaluate whether another NPC is worth collaborating with.

Re: NPC Characteristics

#14
Not to argue, but to explore these ideas, I believe it's important to bear in mind that Limit Theory is a computer game in which AI actors (whom we might as well call NPCs) will do things for reasons.

In other words, the AI model described for LT at this time is one in which the actions that an NPC chooses to take will flow from the NPC's high-level goals, and Josh has indicated that those goals will be influenced to some extent by the "personality" of the NPC.

It's on this basis that I suggested the fairly well-understood model that I did. Again, its value is that it's simple while expressing reasonably well some fundamental and familiar motivations for behavior. Basing the "personality" of NPCs on this model is thus relatively simple to program and test as well as being easy for the general assembly of players of a computer game to recognize and work with.

Tom, from a personal perspective I agree with you completely on preferring some variation in NPC behaviors, especially in a sandboxy game. At the same time, too much variation just looks like randomness, which is usually not desirable in non-player characters with whom functional interaction is a key gameplay element.

Suppose in the game we meet an NPC flying around mining. Why is he mining? Maybe it's because he has a goal of "make money." OK, why does he want to make money (since that information is what will drive his choices of game predicates)? I know it's very tempting to want this NPC's actions to be based on a story -- his ailing mother needs medicine, and Bob the Trader at Outpost #31 said there was good money in mining, and money can be used to buy medicine, and Mary the Scout said there was good ore in System Blivet-9, etc. -- but how does that work as a generative system for a game filled with thousands (or more) of NPCs? It could be awesome for Josh to work on that... but what's not being developed while he's tweaking that actions-from-stories design and code?

Starting with a simple personality system -- such as power, security, knowledge, identity, but other systems could also work -- for high-level goals, and basing planning decomposition on those goals offers a practical solution to the need for a way to define what NPCs want. And it does so in a way that's quick to understand for both developer and players -- NPCs do things for comprehensible reasons -- while still allowing some flexibility in actual behaviors expressed. Using such a core motivation system does not imply that the actions of every NPC will be entirely deterministic. The actions expressed by NPCs in LT will (I think I'm safe in saying) also be conditioned by environmental factors ("what's around me?"), equipment ("what gear do I have?"), factional relationships ("how do I feel about who I'm interacting with?"), and even a degree of randomness ("I think I'll go this direction today!").

Having a systemic internal model of personality doesn't mean that every NPC you see exploring will have "Explorer" tattooed on their simulated brain. It means that NPCs who are naturally interested in discovering things about the world will tend toward actions that increase that satisfaction, whether through research or interacting with lots of other NPCs or flying around the edges of their space to see what's out there. How they do those things is still open for interesting variation, as I agree it should be.

But again, I'm not stuck on any particular approach for getting there. If someone would like to propose a specific alternative model for NPC goal-seeking that does a better job than the one I suggested of being codable, testable, comprehensible, and expressive without determinism, I know Josh will give it the same fair consideration he does all the notions we toss out here.

Gotta start somewhere, though. :)

Re: NPC Characteristics

#15
Wouldn't it make more sense to construct NPC personalities around things that need to be done to keep the universe running and interesting? At the end of the day, we're still looking at a game, with the AI being required to facilitate a breathing, living universe.

While it would probably be fun to talk to a navel-gazing AI about the vagaries of its virtual existence and wax philosophically about where it does come from and what the meaning of its virtual life is, it's not what makes a good game (well, unless you want to take a shot at the Turing test...). At the end of the day, raw materials must flow, get converted into hardware and consumables, and then used and destroyed to make place for more and prevent the universe from dying, either due to complete breakdown of production chains or massive oversaturation.

I'm not above suggesting to blatantly cheat in the AI department if necessary to achieve these ends, or rather this state of equilibrium. As such, an AI "personality" in my opinion needs to succumb to metagame necessities.
Or, to put it in more concrete works, it needs to be ensured that the AI actually keeps the universe running, instead of bumbling blindly through the dark with some vaguely-defined personal goals. It would probably also be good if faction AIs follow some established archetypes, simply to prevent the factions from becoming too similar to each other. And most important: faction hostility needs to be kept on an extremely tight leash, else we get AI Darwinism, where the "fittest" (or probably, the most aggressive) AI eats all of the lesser ones over the course of a few days. I imagine the game to become incredibly unfun when all you encounter is xenophobic megafactions that have adopted a "kill on sight" philosophy on anything that doesn't belong to the faction, and is large and willing enough to throw hundreds of ships at the player.

I'm also quite unclear on how to tie all these AI shenanigans together. It's pretty clear that not all NPCs will be created equal, as there needs to be a chain of command somewhere in order to get the AI to accomplish larger goals. And then there's always the problem with repopulating an area after a local faction ceased to exist.

Faking the illusion of a working, living universe is a lot easier than constructing one that actually works. Even more when it's randomly generated. Depending on the scope of AI influence, I can see universes breaking down at an alarming rate if left unsupervised. Quite frankly, I see quite a few problems with the planned AI approach as far as the sustainability goes. I wish Josh would elaborate a bit on the actual mechanics of production chains, factions, entity spawning and the economy in general, since these subjects are what the AI needs to deal with on a regular basis. Once we have established what the AI can/must do, we can wax poetic about implementing personality along those lines.
Hardenberg was my name
And Terra was my nation
Deep space is my dwelling place
The stars my destination

Online Now

Users browsing this forum: Bing [Bot] and 2 guests

cron