Is Subsumption Architecture a Credible Alternative to Classic AI
Disney Research is currently studying real-time AI and subsumption architecture, which uses sensory data to select actions in autonomous robotics. Subsumption architecture can provide a robot with an emotional spectrum, just as humans can with sensory input. The subsumption architecture, often called GOFAI, is a control architecture offered as an alternative to classic AI. It links sensory information to action choices in an intimate and bottom-up way, rather than using symbolic mental representations of the world to guide behavior. In addition, the The look of the Disney robot is more realistic due to the subsumption architecture.
Subsumption architecture approaches the problem of intelligence in a fundamentally different way than conventional AI. Four basic concepts guided the objective:
A basic concept of situated AI is that a robot should be able to react to its environment in a manner similar to that of a human. Rodney brooks believes that a placed mobile robot should not model the world internally and then function according to this model. Rather than modeling the environment, he asserts that “the world is its own best model,” meaning that appropriate configurations from perception to action can be used to interact with it directly. Despite this, each module / behavior continues to model the world in proximity to sensorimotor signals, albeit at a very basic level. These simple models should make hard-coded assumptions about the world but avoid using memory to predict the world’s behavior, relying instead on direct sensory feedback.
The first benefit is that it pushes the designer to test and build an integrated physical control system rather than theoretical models or simulated robots that may not work in the real world. The second is that it is able to solve the symbol anchoring problem, which many classic AIs face, by directly connecting sensory data to meaningful actions.
In light of the progress of evolution, Brooks argues that acquiring the abilities of perception and movement is an essential prerequisite for the creation of a human-like intellect. Moreover, by excluding top-down representations as a possible starting point for AI, it appears that “intelligence is determined by the dynamics of interaction with the world”.
The individual modules are not considered intelligent in themselves. As determined by observation of the agent and its environment, the interaction of such modules is generally considered intelligent (or not). Thus, “intelligence” is “in the eye of the beholder”.
- Emphasis on iterative development and testing of real-time systems in their target domain.
- Emphasis on the direct link between limited and task-specific perception and the expressed actions that require it.
- The emphasis on distributive and parallel control, which integrates the systems of perception, control and action in the manner of an animal.
- The challenge of building a flexible action selection via a highly distributed system of inhibition and suppression.
- The lack of substantial memory and symbolic representation of architecture, which seems to prevent it from understanding language.
When the subsumption architecture was invented, its new configuration and new approach allowed it to be successful in a wide variety of critical areas where traditional AI had failed, including real-time interaction with a dynamic environment. However, its lack of extensive memory storage, symbolic representations, and central control puts it at a disadvantage when it comes to learning complex behaviors, deep mapping, and understanding language.
Subsumption is a robot control technique that differs from the “traditional” method. This is a simple approach. Rather than a series of complex individual “vertical” activities, the subsumption approach views the control problem as a series of “layers” ordered “horizontally”. Each layer is meant to implement a “skill” or the ability to display behavior. Thus, each layer is quite capable of leading the robot independently. Additionally, the adaptability of robots can be increased by using subsumption control, allowing them to work in complicated and unstructured environments.
Join our Discord server. Be part of an engaging online community. Join here.
Subscribe to our newsletter
Receive the latest updates and relevant offers by sharing your email.