AI the big picture

 AI isn't new but was researched since decades by multiple researchers. They have investigated andless amount of theories and algorithms for different subjects. To get a better picture what the AI community has researched in the past, the working thesis is, that there was a transition from closed systems in the past, to open systems in the present time. This working thesis should be explained briefly.

A closed system is the natural understanding in computing. It assumes that a software runs on a computer, and the programmer has to write down the source code including the algorithm. A typical example is a model predictive control algorithm which takes a physics engine to predict future states, or a path planning algorithm like RRT which searches for the shortest path. These approaches are imitating classical computer science paradigm which are working with the same technique.

The idea of a closed AI system is to grasp the reality in mathematical terms and write a computer program which solves a mathematical optimization problem. Such kind of appraoch was common in AI history until the 1990s. The only debate was about which algorithm was prefered, for example neural network or an alpha beta pruning algorithm.

It should be mentioend, that closed systems are not powerful enought to tackle advanced probloems. Especially in the domain of robot control, the paradigm fails every time, because of the state space explosion. There is no algorithm available which can handle millions of joint configurations of a biped robot. That was the reason why some pessimistic AI researchers in the past have assumed, that its not possible to solve np hard problems in AI.

A more powerful paradigm is an open system. Early examples are motion capture systems from the 1980s which are recording the position of markers in real time. Such a system is open because it tries to capture data from the environment, here mocap data. Another example of an early open system are text adventures like Zork I which puts also a great priority on human to machine interaction. Modern open systems developed after the year 2000 are using advanced interfaces based on text and sensory data. These systems are open because the input send to the computer is the most important information. A human operator might speak "Move to north and grasp the blue box". or another human operator might demonstrate a walking pattern in a motion capture suite and the robot has to repeat the trajectory. In open systems, the man to machine interaction stands in the center of attention. Possilble technologies like certain algorithms, a certain neural network or a database is groupoed around this principle. For example, a neural network might used to deterect the mocap markers, while a SQL database is used to store the realtime data, and then a rendering algorithm might fetch the database and paint the human pose on the screen.

From a technical perspective, these algorithms are trivial and most of them were available before the 1990s. The innovation is the context in which they are used which is human to machine interaction. The existing software libraries are not used to build closed systems e.g. a genetic algorithm which tries to improve itself, but they are used to parse textual input or annotate sensor data with textual [tags].