AI in Digital Entertainment Instructor: Rand Waltzman E-mail: [email protected] Phone: 790 6882 Room: 1430, Lindstedtsvägen 3 4 point course Periods I and II Who Cares? To start with, I hope that you care just because you think this stuff is fun. But wait, there’s more !!! Who Cares? Submitted Jul 19, 2006: Senior AI Programmer LucasArts Major Responsibilities: The Senior AI Programmer will be responsible for designing an automated system to control the behaviors, short and long term goals, and reasoning of AI This person will be tasked with implementing path finding algorithms and solutions The person in this role will create data driven, customizable, flexible, and robust code, systems, and algorithms Provide technical expertise to develop games or technologies in support of games. … Assist in the sharing of ideas and exploration of new practices to continually improve the quality of software development for the Company … Who Cares? Current Listings of specifically AI programmer positions from Gamasutra (www.gamasutra.com) – Midway Games -- Chicago, IL, United States – Offset Software -- Newport Beach, California, United States – Concrete Games -- Carlsbad, CA, United States – Petroglyph -- Las Vegas, Nevada, United States – Mad Doc Software -- Andover, MA, United States – NaturalMotion -- Various Locations, United States Course Literature Variety of web sites and electronically available papers. Join American Association for Artificial Intellilgence (www.aaai.org) – Online access to numerous conference proceedings including “Artificial Intelligence and Interactive Digital Entertainment” proceedings. – One of the best sources of materials on all aspects of AI. – International Student membership - $75 (about 550SEK) A bargain at twice the price!! Administrivia There is no tenta for the course! There is a final paper. – Design and analysis of some type of digital based entertainment that uses some type of AI technology to enhance the participants experience. Three homework assignments. Final paper is required to pass the course. Final grade will depend on how many successful (graded on a pass/fail basis) homework assignments you hand in on time. – 1 assignment Final grade 3 – 2 assignments Final grade 4 – 3 assignments Final grade 5 Details of the paper and the homework assignments will (soon) be found on the course web site. “The holy grail of game design is to make a game where the challenges are never ending, the skills required are varied, and the difficulty curve is perfect and adjusts itself to exactly our skill level. Someone did this already, though, and its not always fun. It’s called life. Maybe you’ve played it?” New Possibilities Application of AI techniques offer potential for new: – Media – Design field – Art form Different dimensions to consider: – – – – Cognitive psychology Computer science Environmental design Storytelling What is Fun? A source of enjoyment. All about making the brain feel good. – Release of endorphins into your system. – Same sorts of chemicals released by Listening to music we resonate to. Reading a great book. Snorting cocaine. Having an orgasm. Eating chocolate. Fun is the feedback the brain gives us when we are absorbing patterns for learning purposes. Subtle Approach One of the subtlest releases of chemicals is at the moment of triumph when we – Learn something – Master a task – Our bodies way of rewarding us This is one of the most important ways we find pleasure in games. In games, learning is the drug. Boredom is the opposite. – When the game stops teaching us, we feel bored. Experience vs. Data New data is used to flesh out a pattern. New experience might force a whole new system on the brain. – Potentially disruptive and not so much fun. Games must continually navigate between – Deprivation vs. overload – Excessive chaos vs. excessive order – Silence vs. noise How to Make a Boring Game Player figures out whole game in first 5 minutes. Player might see that there are incredible number of possible permutations. – Require mastery of a ton of uninteresting details. Player fails to see any pattern whatsoever. Pacing of the revelation of variations in the pattern too slow. – Or too fast. A Little Cognitive Theory The brain is made to fill in the blanks. – E.g., see a face in a bunch of cartoony lines and interpret subtle emotions from them. – Fantastic ability to make and apply assumptions. The brain is good at cutting out the irrelevant. – Show somebody a movie with a lot of jugglers in it. – Tell them in advance to count all the jugglers. – They will probably miss the large pink gorilla in the background. The brain notices a lot more than we think. – Put somebody in a hypnotic trance and ask them to describe something vs. – Asking them on the street! A Little More ... The brain is actively hiding the real world from us. – Ask somebody to draw something. – More likely to get the generalized iconic version of the object ... The one they keep in their head. – Rather than the actual object they have in front of them. Seeing what is actually in front of us is hard. – Most of us never learn how to do it. Chunking Compiling an action or set of actions into a routine. – Allows us to perform the action on autopilot. – Burning a recipe into the neurons. Example: Describe how you get to work in the morning. – – – – – Get up Stumble to the bathroom Take a shower Get dressed Drive to work. Easy enough, but ... Chunking What if I ask you to describe one of these steps? Example: Getting dressed. – – – – Tops or bottoms first? Socks in top or second drawer? Which pant leg goes in first? Which hand touches the button of your shirt first? You could probably answer with enough thought. – This operation has been chunked. – You would have to decompile and that would take time. More on Chunking ... We usually run on chunked patterns. – Most of what we see is a chunked pattern. – We rarely look at the real world. We usually recognize something chunked and leave it at that. When something in a chunk does not behave as we expect we have problems. – A car starts moving sideways on a road instead of forward. – We no longer have a rapid response. – Unfortunately, conscious thought is very inefficient. – If you have to think about what you are doing, you are likely to screw it up. 3 Levels of Thought Conscious thought. – – – – Logical Works on a basically mathematical level. Assigns values and makes lists. Very slow! Integrative, associative and intuitive. Non-thinking thought. – You stick your hand in a fire. – You pull it out before you have time to think about it. Integrative Thought Part of the brain that does the chunking. Can’t normally access this part of the brain directly. It is frequently wrong. It is the source of common sense. – Often self-contradictory. “look before you leap” “he who hesitates is lost” This is where approximations of reality are built. Appeal to Their Intelligences Some basic types of intelligence that entertainment can appeal to: – – – – – – – Linguistic Logical-Mathematical Bodily-Kinesthetic Spatial Musical Interpersonal Intrapersonal Internally directed Self motivated Fun is Educational Learn to calculate odds. – Prediction of events. – Qualitative probability. Learn about power and status. – Not surprisingly of interest since we are basically hierarchical and strongly tribal primates. Learn to examine environment or space around us. – Spatial relationships are critically important. – Classifying, collating and exercising power over the contents of space is crucial element of many games. Using spatial relations as basis for predictive models. Fun is Educational ... Learn to explore conceptual spaces. – Understanding rules is not enough. – To exercise power over a conceptual space we need to know how it reacts to change. – Exploring a possibility space is an excellent way to learn about it. Memory plays an essential role. E.g., recalling and managing very long and complex chains of information. – Provide tools for exploration. But, the trick is to strike a balance between Teaching players to rely on tools to overcome their own limitations VS Making people so dependent on tools that they can’t function without them. Fun is Educational ... Learn basic skills: – Quick reaction time. – Tactical Awareness – Assessing the weakness of an opponent. – Judging when to strike. – Network building. A very modern skill. As opposed to basic cave-man skills. Good Entertainment Thought provoking Revelatory – Good portrayal of human condition – Provides insight Contributes to betterment of society. Forces us to reexamine assumptions. Gives us different experiences each time we participate. Allows each of us to approach it in his/her own way. Forgives misinterpretations – Maybe even encourages them Does not dictate. Immerses and imposes a world view. From Game to Art For games to reach art, the mechanics must (one point of view) be revelatory of the human condition. – Create games where the formal mechanics are about climbing a ladder of success. E.g., mechanics simulate not only the projection of power, but concepts like duty, love, honor, responsibility. – Create games that are about the loneliness of being at the top. – Sample Titles Hamlet: The Game Working for the Man Sim Ghandi Against Racisim Custody Battle Example Your goal is the overall survival of your tribe. You gain power to act based on how many people you control. You gain power to heal yourself based on how many friends you have Friends tend to fall away as you gain power. So: – Being at the top and having no allies is a choice. – Being lower in the status hierarchy is also a choice Perhaps more effective Feedback: – Reward players for sacrificing themselves for the good of the tribe. – If they are captured during the game, they may no longer act directly but still score points based on the actions of the players they used to rule. – This could represent their legacy. What is Artificial Intelligence Can Machines Have Minds? Two Types of Goals AI and Computer Science Examples of AI Research Other AI Research Areas AI is Inherently Multi-Disciplinary Different Strokes for Different AI Folks AI Programming ACM Computing Classification I.2.0 General Cognitive simulation Philosophical foundations I.2.1 Applications and Expert Systems Cartography Games Industrial automation Law Medicine and science Natural language interfaces Office automation I.2.2 Automatic Programming Automatic analysis of algorithms Program modification Program synthesis Program transformation Program verification ACM Computing Classification •I.2.3 Deduction and Theorem Proving •Answer/reason extraction •Deduction (e.g., natural, rule-based) •Inference engines •Logic programming •Mathematical induction •Metatheory •Nonmonotonic reasoning and belief revision •Resolution •Uncertainty, ``fuzzy,'' and probabilistic reasoning ACM Computing Classification •I.2.4 Knowledge Representation Formalisms and Methods •Frames and scripts •Modal logic •Predicate logic •Relation systems •Representation languages •Representations (procedural and rule-based) •Semantic networks •Temporal logic •I.2.5 Programming Languages and Software •Expert system tools and techniques ACM Computing Classification I.2.6 Learning Analogies Concept learning Connectionism and neural nets Induction Knowledge acquisition Language acquisition Parameter learning ACM Computing Classification I.2.7 Natural Language Processing Discourse Language generation Language models Language parsing and understanding Machine translation Speech recognition and synthesis Text analysis ACM Computing Classification •I.2.8 Problem Solving, Control Methods, and Search •Backtracking •Control theory •Dynamic programming •Graph and tree search strategies •Heuristic methods •Plan execution, formation, and generation •Scheduling ACM Computing Classification •I.2.9 Robotics •Autonomous vehicles •Commercial robots and applications •Kinematics and dynamics •Manipulators •Operator interfaces •Propelling mechanisms •Sensors •Workcell organization and planning ACM Computing Classification •I.2.10 Vision and Scene Understanding •3D/stereo scene analysis •Architecture and control structures •Intensity, color, photometry, and thresholding •Modeling and recovery of physical attributes •Motion •Perceptual reasoning •Representations, data structures, and transforms •Shape •Texture •Video analysis ACM Computing Classification •I.2.11 Distributed Artificial Intelligence •Coherence and coordination •Intelligent agents •Languages and structures •Multiagent systems Quality bars of the near-future “Emergent” behaviors Believable characters 100x physics Portable avatars, persistent assets Communities Economies and money Camera POV and LOD drives gameplay Collaborative and dynamic intelligences AI could be a “killer app” feature of next gen Characters: – Awareness – Memory – Complex motives, simple commands Must be co-developed with animators! – Game AI must be acted out and seen – Expressions & gestures The Madden Test (of game AI) 1985: “That’s not football!” 1990: “I’d fire the coach!” 1995: “What are those guys doing?” 2000: “Rookie, you’re cut!” 2005: “That’s the way I designed it!” What EA learned from John Madden The Oakland Raiders playbook – Matchup strategy – 5 zones of field-position One Knee Equals Two Feet – Player ratings – All-Madden team Madden Football Genesis 23% Madden 97 Playstation 54% Madden 2001 Playstation 2 63% Madden 2005 PS2 83% AI is not critical…yet AI cited for 6/20 top PS2 games. – Metal Gear, NFL x 4, Soccer AI cited for 3/10 top PC games. – Half Life x 2, Civilization When AI is applauded… Appropriate non-player characters (npc) behaviors – Dynamic adjustments – Satisfying mistakes – Coordinated attacks/retreats Challenging opponent When AI is punished… Too easy – No cover Too dumb – Low awareness – “Deer in headlights” Observation #1 Maximize the ratio of perceived intelligence to internal complexity. -Put a red pyramid on a green block. >OK -Pick up a blue block >OK SHRDLU - Winograd -Men are all alike. >IN WHAT WAY? -They're always bugging us about something or other. >CAN YOU THINK OF A SPECIFIC EXAMPLE? -Well, my boyfriend made me come here. >YOUR BOYFRIEND MADE YOU COME HERE -He says I'm depressed much of the time. >I AM SORRY TO HEAR YOU ARE DEPRESSED -It's true. I am unhappy >DO YOU THINK COMING HERE WILL HELP YOU NOT TO BE UNHAPPY Eliza - Weizenbaum SHRDLU -Men are all alike. >IN WHAT WAY? -They're always bugging us about something or other. >CAN YOU THINK OF A SPECIFIC EXAMPLE? -Well, my boyfriend made me come here. >YOUR BOYFRIEND MADE YOU COME HERE -He says I'm depressed much of the time. >I AM SORRY TO HEAR YOU ARE DEPRESSED -It's true. I am unhappy >DO YOU THINK COMING HERE WILL HELP YOU NOT TO BE UNHAPPY Eliza 5 1 Perceived Complexity Actual Complexity 5 -Men are all alike. >IN WHAT WAY? -They're always bugging us about something or other. >CAN YOU THINK OF A SPECIFIC EXAMPLE? 1 -Well, my boyfriend made me come here. >YOUR BOYFRIEND MADE YOU COME HERE -He says I'm depressed much of the time. >I AM SORRY TO HEAR YOU ARE DEPRESSED -It's true. I am unhappy >DO YOU THINK COMING HERE WILL HELP YOU NOT TO BE UNHAPPY SHRDLU Eliza Observation #2 The player will build an internal model of your system. If you don’t help them build it, they’ll probably build the wrong one. Observation #3 The flow of information about a system has a huge impact on the players perception of it’s intelligence. Observation #4 From the players point of view there is a fine line between complex behavior and random behavior. Visibility of causal chains usually makes the difference. Observation #5 Mimicking human intelligence and maximizing the intelligence of an artificial system are 2 very different tasks. Observation #6 There are many applications of AI in games that don’t involve Opponents, Avatars or even human-like intelligence. Meta AI Peer AI Sub AI Meta AI Experience Peer AI Agents - Information Flow - Pacing - Simple Player Model - Behavior - Opponents/Avatars - Complex Player Model Sub AI Simulation - Physics - Tactile - Intuitive Player Model Meta Meta Meta Peer Peer Peer Sub Sub SimCity Meta Peer Sub The Sims Meta Peer Sub Sub Spore Meta Peer Sub Observation #7 Building a system that collects and reflects natural intelligence might be easier than replicating that intelligence. 45 Observation #8 Building a robust, internal model of the player can have huge potential value. From the player’s model of the computer…to…the computer’s model of the player Computer Understanding Player Story Adaptive Mapping Comedy Romance Horror Mystery Action AI Research & IE Practice IE has strong interest for systems that think, behave and interact like people. – Autonomous agents as supporting cast roles. Virtual Worlds: – NPCs Real Worlds – Companions – Collaborators – Opponents Good news for AI research community. – No simple non-AI engineering solution. Some Daunting Challenges Significant difference in the rate of development in AI and IE. – Progress in AI is slow – slower than ever. – IE experiencing explosive growth in both academia and industry. Slow progress of AI will not keep pace with academic and industrial interests. E.g., autonomous virtual animated characters. – Graphics researchers have provided animated character bodies approaching realism in visualization and animation. – Capacities for autonomous planning, control, conversation, and interaction are barely passable for most IE applications. Industry Can’t Wait IE has had to rely on fully scripted interactions with human players to support complex interactions. – Exception: Basic Combat One approach: – Have supporting cast members played by real humans. – In many ways, the rise of multiplayer and massively multiplayer IE forms has greatly reduced industry need for human-level AI. – But in other ways it has increased the need. E.g., how to help orient and retain new participants in a confusing virtual world. Social Preferences Interacting with other humans in a distributed online environment might be preferable for many. Result is increased interest in research in sociology and social psychology. – Social network analysis. – Personality profiling. – Perhaps more important than the fidelity of NPCs. Advice to AI Community Be happy that some of the pressure is being relieved! Broaden the scope of your expertise to include elements of the social sciences. Follow the Money! IE Industry probably has no intention of funding basic AI research. Traditional flow of software content: – Small developers – Filtered through hardware manufacturers – Large publishers. None of these has incentive to support individual basic research projects. – Not for industry-research collaboration either. Follow the Money! Developers probably have most to gain. But .. – Tight deadlines. – Slim profit margins. – Clash with academic models of high risk investigation. Ideas more likely to cross the divide than code. – Expect to see increased interest in academic prototypes. – Implies importance of research funding for prototypes. – Where will this funding come from? – Wait (!!) – it is the cavalry to the rescue ... Necessity is the Mother of Invention The military has been the most consistent source of AI research funding throughout its entire history. Increasing reliance on automation and information technology superiority. Steadily increasing interest in IE. – E.g., computer game technology for military Simulations Training Recruitment Existing comfort level with AI research has made it easier for military IE projects to have significant AI components. And the happy news is .. – the military is heavily into the tradition of the research prototype! A Couple of Suggestions AI should take advantage of the reduced need for human-level AI brought about by increased interest in multiplayer and massively multiplayer systems. – Use research-grade AI systems in the automation of supporting cast member roles that most humans would not find entertaining to play. Computational linguistics has been a notable exception in the slow pace of AI research. – Fueled by empirical and statistical methods. – Few IE researchers have capitalized on the potential offered by current technology. A Final Word If anything you have heard today has upset or discouraged you in any way, remember The Guide’s most important bit of advice: Don’t Panic!