Most AI researcher argue, that if AI has been invented it is possible to use this technology to increase the productivity. The assumption is, that Artificial Intelligence is equal to a higher productivity, this will lower the costs and make the economy faster. But how exactly is AI connected with higher productivity? It make sense to ask if this assumption is true what the conditions are.
To research the topic a bit further we have to investigate a situation in which Artificial Intelligence is available but the productivity has become lower. A typical example is the non player character in a game computer game like Pacman. The funny thing is, that the ghost who are controlled in the game autonomously make the life not better but they will hurt the player. How can this be? It contradicts the story, that Artificial Intelligence will increase the productivity in any case. The so called Non player characters were invented with the opposite purpose, the social role of the ghost is, to attack the human player. His task is to prevent the player of become successful. And if we are going a step forward, the Pacman game itself was created with the purpose to lower the productivity. Somebody would argue, this is normal because a game is different from work. I think, this point is important. The AI in the pacman game works as a non player character and it is working in a game.
Let us go back to the initial problem. The main reason why Artificial Intelligence is researched is because of increasing the productivity. For reaching this goal, two conditions has to be fulfilled, first the AI is not located in the non player characters but is replacing the human player, and secondly, the AI is not located in a computergame but in a real world application. Is this combination possible? Unfortunately not. The only AI which is available today is built in into a computer game and is working always against the human player. So called ingame AI is the most successful technology and it is available today. The problem with all the automatic controlled ingame AI which is working in games like F.E.A.R., Starcraft, and racing simulation is, that the AI is trying to beat the human. In most cases the software is able to do so.
But this doesn't solve the problem. Because it's a game and because the AI is working against the human. As a result, the productivity will becomer lower but not higher. To understand this paradoxon we have to go into the details of how to create an Ingame AI. Suppose the game is called Starcraft and the non-player character is controlling a swarm of units. The swarm has the goal to make the play for the human more difficult. At first it is important to know, that current Finite state machines in combination with GOAP planning techniques are able to realize this goal. In most cases, the human player will loose against a computer controlled Starcraft Swarm. The reason is, that the programmed algorithm will work 24/7, never make a mistake and the programmer can implement in the algorithm all the knowledge he has. The situation is the same like in a game of computer chess. The player makes a move, the computer makes a move, and at the end, the human will have lost the game.
On the first look, the AI community can be proud of their AI software because they have mastered in complicated problem. But has this technology improved the productivity? The first thing to know is, that computergames are additionally to work tasks. That means, in real life there is no game such chess or Starcraft. Even if the computer is beaten the world champion in Chess, he hasn't produced anything useful. And the second thing which is important is, that the AI has not helped the human but the AI has beaten the human. What does this mean? It means, that if the human deactivates the AI in the game, he will reach the goal much faster. The AI is programmed to become an obstacle.
What engineers are prefering are machines who make the human more productive. The car is good example. Using a car is better than not using a car. Or using a the electric light is superior over not using electric light. The assumption is, that using an AI will increase the productivity. But exactly this is not true. Using an autonomous car will lower the productivity.
Let me give another example. The theoretic assumption is, that an automatic Unmanned Air vehicle which can deliver a box to the customer will increase the productivity. The customer has to press the “buy” button on the smartphone and 10 minutes later, a drone lands with the pizza box on his balcon. The bill was payed electronically, and the customer can eat the pizza. That is in short the assumption how a UAV is working.
Now we should describe the situation a bit more realistically. The first thing to know is, that a UAV can't replace the normal logistics. But it is an additional service. Secondly the price for the service is higher than the normal one. And last but not least, the customer has to read a complicated manual and he will loose his time because he is not familiar with an UAV. But how can this be? Isn't Artificial Intelligence automatically an improvement? Exactly this is the problem. That AI will results into higher productivity is the hope but it was never proven. The only thing what was invented yet, is AI software which makes life more complicated.
To understand this phenomena we have to search for an AI application within the computing domain. A normal Windows PC contains of many programs. Which part of the PC can be improved with the help of Artificial Intelligence? Can the Chess engine or the Starcraft playing bot used for automating Windows task?No it can't. The only thing what is available in modern operating systems are background scripts, but they do not contain Artificial Intelligence, but are programmed more simpler. For example, one of the scripts is checking for updates and the next script is searching for device drivers. That means, the Windows PC itself doesn't profit from ingame AI software.
Which domain will profit from it? Exactly this is the problem. The assumption is, that ingame NPC bots can be used in real life cases for urseful purposes, but they don't. Let me give an example. In theory, it is possible to control a swarm of real robots in a maze similar to what the Starcraft Game is about. That means, physical robots are driving in maze and the control signal was generated by a computer. That means, the starcraft game was transferred into the real world. But how exactly will this demonstration increase the productivity? Right this is the bottleneck. There is no application for 10 self-driving robots in a maze. A real application is not available in which such a swarm make sense.
The funny thing is, that this works independent from the technology itself. Even if the AI Controller was improved, the robot swarm will remains useless. The reason for this paradox situation is, that in reality no control task is available in which an AI can be used meaningful. Or to explain it more direct: the assumption that the reality contains of control problem is wrong.
Norbert Wiener has postulated in the 1950s that the world is equal to a large computer game which contains of control tasks. A person who is grasping an apple is solving a grasp task and a car driver who control the steering wheel is dealing with a line changing problem. This description of the world is trying to sell Artificial Intelligence. If somebody has invented a computer program which can steer a car, he will explain to the public, that they have such a problem. This kind of aggressive marketing is visible today. Somebody has invented a robot controller in ROS, and now he explains to the factory worker, that they have a control problem on the assembly line and the software will solve it.
The hypothesis is, that AI in general will lower the productivity, and will always act in a computer game as a non player character. That means, it is not possible to utilize AI for increasing the productivity but the result is the opposite. To test this hypothesis we can try to sell robotics fair. The promise is, that a robot in a factory will increase the costs, will lower the productivity and will lower the quality. Can this promise be fulfilled? Oh yes, most robots are working in that way. They are programmed as non player characters, they are working against the human worker, they steal the humans time, money and energy.
This pessimistic assumption has a counter example. I've found in the history of robotics a device which can improve the productivity. This device is called a printer. Printers are widespread used in modern offices and most of them are helpful to the employees. Printers are usually not recognized as robots. They are forming are closed system in which the human has no access. And they were designed for a special purpose. The interaction between a printer and the human user is minimized. That means, the human provides the command “print 10 pages” and the printer is doing that job.
The reason why a printer is perceived as a useful device is because the design focus is on the hardware side. The printer software is not very important in most products. The control infrastructure for driving a printer is built ontop of the existing machine.
No comments:
Post a Comment