AI challenges in Entertainment and Player Expression Doug Church AIIDE 1 June 2005 Who Am I • Programmer by Schooling • Designer/Programmer by Training/Practice • Interested in Technology enabling Design Worked on (i.e. coded, designed) a bunch of old PC games, recently have been doing highlevel goal setting/evaluation of (i.e. not real work) a variety of console games 1/6/2005 AIIDE '05 2 This Talk Initially hoping to do an end-of-week recap kind of thing, discussing trends I saw over the course of the conference and relating them to current industry situations But turned out I was speaking first 1/6/2005 AIIDE '05 3 So... Instead, I’ll talk about • AI related game design challenges I find most interesting as we head to next generation • My perception of how the state of the industry impacts AI work going forward This is a game design talk more than an AI talk Primary focus on single player games There are no pretty demos or pictures... sorry 1/6/2005 AIIDE '05 4 Topics and Takeaways • • • • • Changing landscape in games space How biz model messes with AI evolution Rise of entertainment content What experiences are we providing Opportunities/approaches for meaningful forward steps in the industry, risks 1/6/2005 AIIDE '05 5 Context Much is changing in the industry Changes in Market • Growing gap between top-sellers and the rest • Growing budgets and team size • Clear hook, 30 second experience pitch • Promise of direct distribution/ability to hit smaller markets with indie titles remains on the horizon, unfulfilled, but still discussed • Indie market would mean a very different talk 1/6/2005 AIIDE '05 7 Changes in Development • Content the real issue (budget, time, quality) – Most games now content led not tech led – Data driven everything in order to get done – Get programmer out of way, tight feedback loop • Team scale impacting process, creativity – Getting everyone on same page – Creatively collaborate at large scale • Less discipline-based, more task-oriented 1/6/2005 AIIDE '05 8 Changes in Gaming • For many years games were linear sequences of challenges – challenge itself as play-value/entertainment – “world” abstracted and simplified • Current trends – high world fidelity, and growing – open-ended worlds, player choice/customization – entertainment aspects/pacing, characters 1/6/2005 AIIDE '05 9 Changes in Fidelity • • • • Graphics: huge leaps in 20 years Scale and density of environments increasing Physics fidelity obviously much better Even design has several new “common styles” of play (trading/collectibles, openworld mission stacks, etc...) • Conversation systems? not so much... 1/6/2005 AIIDE '05 10 Changing Players • Online (XBox Live, etc) providing more multihuman gaming opportunities • MMO’s growing in popularity, providing large environments for “meaningful” player action • Single-player games containing more movieinspired moments, more watching 1/6/2005 AIIDE '05 11 The Mainstream Industry Getting Games Made The Industry Now • • • • Consolidation: Fewer games, bigger budgets Harder to get projects started, approved Risk management central to business Large public companies revenue driven, need big sales numbers regardless of dev cost • Licensing IP from other media, or sequel-ing established IPs, major part of forward plan 1/6/2005 AIIDE '05 13 Evaluating Games • Player Fantasy: what experience is being delivered for the player • Key Pillars: game feature hooks for players to get excited about and promise play value • Uniqueness: competitive landscape • Reality Check: can it be done? on time? – market windows, competitive product timelines, budget analysis, dev team experience, etc... 1/6/2005 AIIDE '05 14 Relevance to AI • Improved cover finding is very nice • Guys who don’t get stuck on corners are nice • Non-magical following of a racing line is nice ... none of these makes much of a 30 second TV ad, or a quick sound byte for Newsweek ... and if the character only lives for a minute, how much fidelity can we even perceive 1/6/2005 AIIDE '05 15 Market Reality • Tons of interesting and hard work goes into making modern game AI’s able to work at all – but it is expected by the market, and really is mentioned most when it fails, not succeeds • Even in specialist press, we’ve basically bragged our way out of meaningful claims – “realistic characters respond to your actions” has been said for 10+ years about games – and often scripted stuff has more impact/is remembered more by players 1/6/2005 AIIDE '05 16 The Result • AI code can be hard to build, and innovating and improving it is seen as risky – So potentially valuable/interesting features often cut as scope/risk reduction • Press/users outcry for better AI usually trivial – i.e. pathing and grenade dodging kind of things – hard work, sure, but not some new innovation • Hard to make case for future looking AI investment given risk profile and low ROI 1/6/2005 AIIDE '05 17 That Said • Consumers do get excited when something combines new AI with new play idea/concept – Black and White, The Sims, first RTS games – Many of these had struggles to approve/release • Can we encourage/enable more of that? 1/6/2005 AIIDE '05 18 What is needed to Pitch it? • Attach AI feature to compelling player fantasy • Identify and show a unique player experience/mechanic the AI feature enables • Some other pain point of development (cost, time) that the AI technique will improve • Evidence it wont require leaping off some 10year research project cliff into total unknown 1/6/2005 AIIDE '05 19 Technology Status What challenges? What can we do? AI Tech Situation • As fidelity of worlds (graphic environment detail, lighting, terrain complexity) grows, challenge of just keeping up ratchets up – – – – – pathing on a 2d tilemap with 90 degree walls easy pathing on an arbitrary polygon mesh, not as easy switching between idle and combat sprites easy managing 100+ bone blended model, not as easy and so on... • Just keeping old features working is hard 1/6/2005 AIIDE '05 21 Some example AI Tech • • • • • • • Pathfinding Map Analysis (cover, opportunity, shortcut) Group coordination and management Actions (climb rope, fire gun, plus world use) Expression (what do i say, what anims...) Senses Traits and Characterizations 1/6/2005 AIIDE '05 22 Some example AI Tech • • • • • • • Pathfinding Map Analysis (cover, opportunity, shortcut) Group coordination and management Actions (climb rope, fire gun, plus world use) Expression (what do i say, what anims...) Senses Traits and Characterizations 1/6/2005 AIIDE '05 23 Enabling Players • Unique DNA of gaming is interaction • Currently, we provide lots of microinteractions (move, shoot, dodge) • Lack of support for larger scale player choice • Hence drive towards online (where other humans provide the reaction to the choices) • AI is the obvious tool to enable more player flexibility and expression 1/6/2005 AIIDE '05 24 Who Cares? • Plenty of games are fine w/current level of expression (Tetris works pretty well, etc...) • And movie style games are plenty fun • But we are missing out on a huge range of possibilities, and ones that are uniquely us • More reactive worlds with more payoffs and meaning to choices will be more human 1/6/2005 AIIDE '05 25 Entertainment Aspects • Big moment payoff in FPS these days rarely comes straight from the systems/AI core – scipted overrides with hand-placed triggers/events – AI core supports the set piece by moving actors about, reacting and sensing, but that is it • Similarly in many other types... AI controls the mundane character actions, and then in big events the character is put into auto-pilot • Pure systems behavior often seen as flat 1/6/2005 AIIDE '05 26 Entertainment and Reality • Often as our AI’s/NPC setups get better, they become worse as game foils – hard to tell what is going on, why – opaque actions, no sense of agency • Need better demonstrations of NPC traits – emotes, drawing attention, some sort of feedback • AI needed to support director’s goals and feel, not impose reality – AI to help pacing, variety, etc... lots to try 1/6/2005 AIIDE '05 27 AI in Games What are we using this for? AI Styles we use a lot • Opponent – Simulates another human player – i.e. enemy fighter in SC, general in an RTS, etc • Manage – Simulates independent agents to attempt to direct – i.e. RTS troops, Sims in the Sims, B&W creature 1/6/2005 AIIDE '05 29 Why these styles • Opponent – Strategize against player abilities – Pick challenges (shoot a guy who is in cover, chase a guy, drive faster than guy) and build AI around that optimization goal • Manage – Build systems based AIs with limited but repeatable capabilities – Gameplay about player learning to use them 1/6/2005 AIIDE '05 30 Styles and Expression • Opponent and Manage provide clear expression for player in micro-actions taken • Presented sequence of small goals, have freedom on how to get there using toolset • Often provide a very small-task oriented approach to completion • Much like a job or homework... checklists 1/6/2005 AIIDE '05 31 Styles we do not use much • Negotiate – limited use in RTS environments • Converse – very very limited conversation tree models, primarily, almost universally prescripted branching • Choice and Consequence – Occasional forays into faction based/multiple valid path, but mostly still save and reload based play 1/6/2005 AIIDE '05 32 Steps toward other styles • Mercenaries – very basic faction model, and somewhat opaque and low on consequence (at least for first several hours), but still has NPC reactions to player choice • Fable – very shallow NPCs, but they pay attention to player actions and shade their opinion/behavior • Nintendogs – Ok, just another pet sim, but it is non-combat character interaction, and a nice “step” on a path 1/6/2005 AIIDE '05 33 Design Challenges • Growing AI complexity makes expression harder in some ways – what were they reacting to? did what i do matter? – is he my friend or did he just not see me or am i wearing a disguise or maybe... • Add a complex behavioral/sensory model and getting meaningful player feedback is tough • Another reason pure opponent model is nice 1/6/2005 AIIDE '05 34 Design Needs • What tools address this, can designers use them, what feedback do they get? • Tension between automated response and precise controls, where is sweet spot? • AI’s need to be Robust, Contextually aware, and Controllable... not an easy balance • Esp. as world systems grow in complexity • Getting good at this for micro-tactical combat setups... but not much for levels above this 1/6/2005 AIIDE '05 35 Going Forward What are we doing? Optimistic View • Keeping up, AI can manage characters in increasingly complex settings • Games to keep moving forward • Complex settings give options for expression • More scope for entertainment as we can create more compelling and full worlds 1/6/2005 AIIDE '05 37 Pessimistic View • Existing trajectory continues to evolve a small set of games, ignoring many more – Great, we master low-level tactics – No player-driven character interactions, scripting the choice for memorable entertainment moments • Single player games become movies with tactical/combat/physics challenges to “turn the page” to the next scene • Single player as “training dummies” for online 1/6/2005 AIIDE '05 38 Proactive Response • Reality somewhere in the middle • Can’t wait for more interesting AI integration and adoption to “just happen” • Need to address current industry needs (entertainment, risk management) w/o giving up on pushing other AI types • Note: Nothing wrong with better pathing/etc, but is that all we can do? 1/6/2005 AIIDE '05 39 So we need to get ahead • Finally good at doing games in Year 2000 – 20 person teams, static worlds, a few characters – Ooops, a bit late, eh? • Where do we need to get ahead – Tools: get on par with graphics/world fidelity • middleware? more sharing of tech ideas? – Tools: better blend of AI systems behaviors and “script-like” entertainment elements... – Risk: need manageable steps, solid path 1/6/2005 AIIDE '05 40 Some Pain Points to address • Dev Costs – w/o better AI tools won’t be able to build content, due to pain of path management and scripting • Necessity of more complex worlds – if-then structures will become unmanageable, get more flexible and robust solutions going now 1/6/2005 AIIDE '05 41 Opportunities to go for • New styles – Hard to make them work, but when you do, and connect them to a market, you are very happy • “Real AI” – planning, learning – manageable but meaningful steps into games • Entertainment – flexibility to allow big moments to be attached to real choices would be very compelling – Take gaming big moments back from the movie people, make them interactive 1/6/2005 AIIDE '05 42 The End