How to externalize intelligence

The classical understanding of artificial intelligence was that the intelligence is always located inside a robot. Inside the robot there is a microcontroller, on this microcontroller is running a software, the software is based on algorithms and intelligence is located in the algorithm.

With the advent of embodied AI since 1990, this strict philosophy was modified. A more recent understanding is, that intelligence can be located on multiple locations like inside the robot, in the environment of the robot or in both locations. A typical example for externalizing intelligence is a line following robot. The robot does not decide about the path but the decision was given by the line on the ground. The robot follows only the line.

A more advanced version of externalizing intelligence is corridor with qr codes. The robot navigates along the qr codes until he reaches the goal. Its possible to guide the robot to a different goal by changing the position of the qr codes. The software on the robot remains the same.

A teleoperated robot has no internal algorithm at all but its decision making process is determined entirely by the environment. In other words, the intelligence is located outside of the robot. Such sort of machinery is very easy to program, because intelligence offloading means that there is no need to code intelligence in software, but intelligence is already available in the environment.

Intelligence offloading generates a new problem which is how to communicate with a robot. In case of a line following robot, the robot needs a sensor to detect the black line, a robot who follows qr codes needs a camera, and a speech enabled robot will need a microphone plus a language parser. None of these technical artifacts have its root in artificial intelligence but they are standard components used for signal transmission. This makes it easy to build such robots.

Instead of designing advanced algorithms and increasing the onboard intelligence of a robot, the new goal is to invent advanced human to robot communication interfaces which makes it easier to submit commands to the robot. The most advanced interface available is based on natural language. Its more advanced than a light seeking robot because natural language works on a high abstraction level. A short sentence triggers a complex reaction of the robot.

Let me give a practical example for human to robot interaction. 
a) the human operator asks the robot to move to a certain position in a maze
b) the human operator asks the robot to follow a sequence of waypoints in a maze
c) the robot should solve a complex pick& place problem like "transport all the boxes in room A"

These examples are sorted by its difficulty. Programming the language interface for case a) is easy. The robot needs only a list with possible locations in the maze and simple path planning algorithm to reach the goals. In contrast, the cases b) and c) are harder to implement. Parsing and executing the commands is a demanding task for a robot.

Strictly spoken a remote controlled robot doesn't need onboard intelligence but the intelligence is located outside of the robot. There is no algorithm which provides intelligence, but the decision is made by the environment and the robot responds to the decision.

A light seeking robot is a good example for an external controlled robot. The robot follows the light, but do not solves motion planning problems. Such a robot works only if a light source is available in the environment. If the light source is at north, the robot will move towards this direction. According to the strict definition of a robot, such a light seeking automaton is not a robot but its an automaton. It doesn't has a program and its doesn't execute an algorithms. A light seeking robot can be realized with out a microcontroller but as an analog circuit. So its not a computer.