Portfolio

Elephants Don't Play Chess

 

Challenge

Teach the core concept developed within Brooks’ "Elephants Don't Play Chess" essay and Braightenburg's Vehicles experiment to fellow MFA Design students in under an hour (many of whom did not have technology/interaction backgrounds).

The concept we were looking at within our graduate-level interaction design class revolved around an artificial intelligence research methodology that emphasizes ongoing physical interaction with the environment as the primary source of constraint on the design of intelligent systems. Prior to this concept being introduced, AI architecture followed the traditional approach of emphasizing the abstract manipulation of symbols.

Role: Designer and instructor

Timeline: 1 week to develop, 30 min to teach

traditional architecture

 

subsumption architecture

 

Process

  • Initial readings and additional research to identify more accessible language
  • Developed a low-fi, participatory exercise for my classmates and instructors
    • Part 1 - Rearranging Architecture
      • Instructed classmates to arrange themselves (with very little guidance) using a set of prompt words and arrows into representations of the two methods
      • Unnecessary complexity thrived!
      • As intended, the input (from me) was not explicit enough to produce the exact behavior that reflects natural intelligence as is often the problem with looking at AI from a “top-down”, information processing perspective
      • Revealed key visuals surrounding the 3 main topics as we talked through the benefit of using behavioral modules in building AI architectures
    • Part 2 - Behavior Modules
      • Strategically selected our responsible and prestigious faculty to become our “simple agents” - to embody the moving parts and the situational sensor 
      • Through a series of discreet commands via ipad, continued to build up "behavioral modules"
      • Students asked to guess what type of command had been given based on the "simple agents" perceived behavior (progressed from easy to difficult)
      • "Simple agents" were eventually able to operate continuously and autonomously without confusion 
      • Turned "controls" over to students for a chance at "inputting behaviors" 

 

Success

  • Created an opportunity that allowed for my classmates to "embody" an AI and "feel" the unnatural reactions that would have existed within traditional architecture vs. more natural ones using simple, behavioral modules layered in such a way to allow for complex and highly intelligent "vehicles" to emerge
  • The second part of the exercise allowed for further discussion on the issues with conflicting behaviors suppressing input or inhibiting output and we were able to also discuss the difference between the initial layers (one could think of them as reflexes) and the subsequent layers (focused towards achieving the overall goal).