Video: Embracing the dark art of mathematical modeling in AI
[Note: To access chapter selection, click the fullscreen button or check out the video on the GDC Vault website]
February 20, 2013 | By GDC Vault Staff 3 comments More: Video
Courtesy of the GDC Vault comes a free lecture on utility-based AI and the mathematical modelling used to calculate AI's complex behavior, from specialists Kevin Dill of Lockheed Martin Global Training & Logistics and Dave Mark of Intrinsic Algorithm.
In a follow-up to their presentation at GDC 2010, Dill and Mark return to GDC 2012 to offer a more advanced lecture on how utility-based AI can handle large numbers of selections dynamically. They champion programming modular decision making "considerations" for AI (such as calculating distance), which can then be reused in various parts of the game design.
Company Name(s): Lockheed Martin Global Training & Logistics, Intrinsic Algorithm
Track / Format: AI Summit
Overview:Utility-based AI is a widely-used approach, particularly for games with deeper or more complex behavior. While new users may find utility functions complex and intimidating, experienced users see them as a natural and comfortable way to express behavior. In a follow-up of their 2010 lecture, Kevin Dill and Dave Mark will show how simple problems can be laid out quickly and easily using common design patterns. Additionally, they will show how complex situations can make use of utility functions to express more nuanced behavior. They will then walk through real-world examples, showing how they would be expressed in a utility-based architecture.
About the GDC Vault
In addition to this presentation, the GDC Vault offers numerous other free videos, audio recordings, and slides from many of the recent GDC events, and the service offers even more members-only content for GDC Vault subscribers. Those who purchased All Access passes to events like GDC, GDC Europe, and GDC China already have full access to GDC Vault, and interested parties can apply for the individual subscriptions via a GDC Vault inquiry form.
Group subscriptions are also available: game-related schools and development studios who sign up for GDC Vault Studio Subscriptions can receive access for their entire office or company. More information on this option is available via an online demonstration, and interested parties can find out more here. In addition, current subscribers with access issues can contact GDC Vault admins.
Be sure to keep an eye on GDC Vault for even more new content, as GDC organizers will also archive videos, audio, and slides from other events like GDC China and GDC 2013. To stay abreast of all the latest updates to GDC Vault, be sure to check out the news feed on the official GDC website, or subscribe to updates via Twitter, Facebook, or RSS.
Gamasutra and GDC are sibling organizations under parent UBM Tech.
Using the analogy of learning pointers to describe learning this form of AI is apt. Thanks for this, Dave and Kevin; this video made me feel smart.
At the ICT, we began to use mathematical modeling for more responsive conversation agents, after years of working with static dialog trees. In the process of learning what this new paradigm was all about (to eventually design an accessible authoring tool for this system, called Roundtable), a programmer locked me in a room for an hour and laid out what this was, buckets of possible decisions. It suddenly clicked, and I kept the programmer in the room another two or three hours excitedly talking about the possibilities for massaging those probabilities, combining weighting and ranking, and even implementing ranking states; that is, alternate ranking values to follow the context of conversation.
Trends like this, and artificial evolution, are what make AI an exciting field to me. Like!
At the ICT, we began to use mathematical modeling for more responsive conversation agents, after years of working with static dialog trees. In the process of learning what this new paradigm was all about (to eventually design an accessible authoring tool for this system, called Roundtable), a programmer locked me in a room for an hour and laid out what this was, buckets of possible decisions. It suddenly clicked, and I kept the programmer in the room another two or three hours excitedly talking about the possibilities for massaging those probabilities, combining weighting and ranking, and even implementing ranking states; that is, alternate ranking values to follow the context of conversation.
Trends like this, and artificial evolution, are what make AI an exciting field to me. Like!