Human Level

Over the last 30 years, research in AI has fragmented into more and more specialized fields, working on more and more specialized problems, using more and more specialized algorithms. This approach has led to a long string of successes with important theoretical and practical advancements. However, these successes have made it easy for us to ignore our failure to make significant progress in building human-level AI systems. Human-level AI systems are the ones that you dreamed about when you first heard of AI: HAL from 2001, A Space Odyssey; DATA from Star Trek; or CP30 and R2D2 from Star Wars. They are smart enough to be both triumphant heroes and devious villains. They seamlessly integrate all the human-level capabilities: real-time response, robustness, autonomous intelligent interaction with their environment, planning, communication with natural language, commonsense reasoning, creativity, and learning.

If this is our dream, why isn't any progress being made? Ironically, one of the major reasons that almost nobody (see Brooks et al. [2000] for one high-profile exception) is working on this grand goal of AI is that current applications of AI do not need full-blown human-level AI. For almost all applications, the generality and adaptability of human thought is not needed--specialized, although more rigid and fragile, solutions are cheaper and easier to develop. Unfortunately, it is unclear whether the approaches that have been developed to solve specific problems are the right building blocks for creating human-level intelligence. The thesis of this article is that interactive computer games are the killer application for human-level AI. They are the application that will need human-level AI. Moreover, they can provide the environments for research on the right kinds of problem that lead to the type of incremental and integrative research needed to achieve human-level AI.

Given that our personal goal is to build human-level AI systems, we have struggled to find the right application for our research that requires the breadth, depth, and flexibility of human-level intelligence. In 1991, we found computer-generated forces for large-scale distributed simulations as a potential application. Effective military training requires a complete battle space with tens if not hundreds or thousands of participants. The real world is too expensive and dangerous to use for continual training, and even simulation is prohibitively expensive and cumbersome when fully manned with humans. The training of 4 pilots to fly an attack mission can require over 20 planes plus air controllers. The military does not even have a facility with 20 manned simulators, and if it did, the cost in personnel time for the other pilots and support personnel to train these four pilots would be astronomical. To bypass these costs, computer-generated forces are being developed to populate these simulations. These forces must integrate many of the capabilities we associate with human behavior--after all, they are simulating human pilots. For example, they must use realistic models of multiple sensing modalities, encode and use large bodies of knowledge (military doctrine and tactics), perform their missions autonomously, coordinate their behavior, react quickly to changes in the environment, and dynamically replan missions. Together with researchers at the University Southern California Information Sciences Institute and Carnegie Mellon University, we set off to build human-level AIs for military air missions (Tambe et al. 1995). In 1997, we successfully demonstrated fully autonomous simulated aircraft (Jones et al. 1999), and research and development continues on these systems by Soar Technology, Inc. Although computer-generated forces are a good starting application for developing human-level AI, there are extremely high costs for AI researchers to participate in this work. It requires a substantial investment in time and money to work with the simulation environments and to learn the extensive background knowledge, doctrine, tactics, and missions. Furthermore, much of the current funding is for building and fielding systems and not for conducting research.

Computer Games

In late 1997, we started to look for another application area, one where we could use what we learned from computer-generated forces and pursue further research on human-level intelligence. We think we have found it in interactive computer games. The games we are talking about are not Chess, Checkers, Bridge, Othello, or Go, which emphasize only a few human capabilities such as search and decision making. The types of game we are talking about use the computer to create virtual worlds and characters for people to dynamically interact with--games such as Doom, Quake, Tomb Raider, Starcraft, Myth, Madden Football, Diablo, Everquest, and Asheron's Call.

Human-level AI can have an impact on these games by creating enemies, partners, and support characters that act just like humans. The AI characters can be part of the continual evolution in the game industry toward more realistic gaming environments. Increasing realism in the graphic presentation of the virtual worlds has fueled this evolution. Human-level AI can expand the types of experiences people have playing computer games by introducing synthetic intelligent characters with their own goals, knowledge, and capabilities. Human-level AI can also recreate the experience of playing with and against humans without a network connection. Current players of computer games are driven to networked games because of the failings of the computer characters. In massively multiplayer online games, human-level AIs can populate the worlds with persistent characters that can play the game alongside humans, providing opportunities for interesting interactions that guide players in the game and enhance the social dynamics between players. Our hypothesis is that populating these games with realistic, human-level characters will lead to fun, challenging games with great game play.

From the AI researcher perspective, the increasing realism in computer games makes them an attractive alternative to both robotics in the real world and homegrown simulations. By working in simulation, researchers interested in human-level AI can concentrate on cognitive capabilities and finesse many of the pesky issues of using real sensor and real motor systems; they must still include some sensor modeling to get realistic behavior, but they don't have to have a team of vision researchers on their staff. They can pursue AI research in worlds that are becoming increasingly realistic simulations of physical and social interactions, without having to create these worlds themselves. Computer games are cheap ($49.95), reliable, and sometimes surprisingly accessible, with built-in AI interfaces. Moreover, computer games avoid many of the criticisms often leveled against simulations. They are real products and real environments on their own that millions of humans vigorously interact with and become immersed in. Finally, unlike military simulations, we do not need to hunt out experts on these games; they surround us.

Another reason for AI researchers to work in computer games is that if we don't start working in this area, the computer game industry will push ahead without us (Woodcock 2000). Already there are at least five AI Ph.D.s working in the industry (Takahashi 2000). AI researchers have the opportunity to team with an aggressive, talented, and caffeine-charged industry in the pursuit of human-level AI. Here is a list of reasons for AI researchers to take the computer game industry seriously (Laird 2000a).

First, computer game developers are starting to recognize the need for human-level AI. Synthetic human-level characters are playing an increasingly important role in many genres of computer games and have the potential to lead to completely new genres.

Second, the computer game industry is highly competitive, and a strong component of this competition is technology. AI is often mentioned as the next technology that will improve games and determine which games are hits. Thousands of new computer games are written every year with overall development time averaging nine months to two years, so technological advances sweep through the industry quickly. Already, many computer games are marketed based on the quality of their AI. This field is one in which AI will have a significant impact.

Third, game developers are technologically savvy, and they work hard to stay current with technology. AI programmer is already a common job title on game development teams.

Fourth, the game industry is big. In terms of gross revenue, the computer game industry is bigger than the movie industry (Croal and Totilo 1999).