Embodied Task Competence as a Necessary Condition for Artificial General Intelligence

Abstract

Claims of Artificial General Intelligence (AGI) increasingly rely on performance in disembodied, symbolic, or virtual tasks while explicitly excluding embodied activities. This position statement argues that such exclusions are unjustified. We propose that any embodied, economically useful task demonstrably performable by a given robotic platform under teleoperation by a minimally trained human must be included in the required task set for AGI, provided the AI system has access to the same sensors and actuators. This criterion is operational, embodiment-neutral, and closes a critical definitional gap in current AGI discourse.

1. Background and Problem Statement

AGI is commonly defined as the ability of an artificial system to perform a wide range of economically valuable cognitive tasks at human level or better. In practice, however, many definitions implicitly restrict these tasks to disembodied domains such as language, games, or abstract reasoning. This restriction creates a category error: many economically valuable human cognitive tasks are inherently embodied, including construction, maintenance, logistics, food preparation, and domestic labor. Excluding such tasks is a historical artifact of benchmarking convenience rather than a principled distinction.

2. Teleoperation as a Proof of Cognitive Sufficiency

When a robot successfully performs an embodied task under teleoperation, it is empirically established that the robot’s sensors and actuators are sufficient, that the environment admits a viable control policy, and that the task can be solved through general reasoning, perception, planning, and adaptation. What is absent is not embodiment but internalized cognition. Teleoperation therefore provides a proof of cognitive sufficiency of the embodiment.

3. Reference Teleoperator Definition

A Reference Teleoperator (RTO) is a human operator who (1) has completed no more than a fixed, task-agnostic amount of teleoperation training on the target robotic platform (e.g. 20–40 hours); (2) uses only the standard sensor, actuator, and feedback interfaces available to the system under test; (3) demonstrates task performance within the interquartile range of a statistically representative cohort trained under the same protocol; and (4) receives no task-specific rehearsal or prior exposure to the test environment.

4. Core Criterion

An artificial general intelligence must be capable of autonomously performing any embodied, economically useful task that can be reliably performed by a given robotic platform under teleoperation by a Reference Teleoperator, using the same sensors and actuators.

5. Implications and Conclusion

Adopting this criterion makes embodied robotics tasks a mandatory component of AGI evaluation where feasible and renders disembodied-only AGI claims incomplete. Teleoperation provides a clean empirical boundary between embodiment and cognition. Where teleoperation succeeds, autonomy is a matter of intelligence rather than mechanics, and any credible test of AGI must reflect this fact.