AI intelligence is like every computer program a sequence of instructions or steps which when executed perform a sequence
of calculation as embedded in the program (instructions) and as defined by the persons who wrote (designed) the program. In short the program executes a sequence of mathematical calculations.
The general rule is when a program is executed again, using the same inputs, the results will be the same. This is the case when the program does not depend on the results on a random number generator nor is time critical.
This is also the case if you play two games of chess against a computer and in the second match you repeat exactly the same moves as in the first game. The result will be the same. What that means is that the computer has nothing learned from the first game.
If your moves are different in both games than 'ofcourse' the computer is forced to make also different moves.
A very interesting game to study AI is the game called Paint Monsters .
The name suggests that it is simple and is played by children. The reality is, that it is not. The game consists of 1605 individual games. When you have solved game #1, you can start with game #2 etc. Each game consists of moves in which the player makes certain selections. There exists a maximum number of moves. When the maximum number of moves is reached the game is terminated. The game is also terminated when a bomb explode etc. The next step for the player is to repeat the game, as many times as she likes, untill she has finally solved the game.
What makes this game interesting, is because it challenges the concept of human intelligence.
|
My definition of itelligence is: the individual capability of the human brain to remember what has 'happened' in the past and as a result in the future, using the logic inherent in my brain, performing the same tasks differently. In short: my individual intelligence, my logical capability, has improved. This is permanent.
|
When you compare this definition with a computer, or with the brain of a robot, than a computer consists of two parts:
A logical section and a memory.
- The logical section consists of a Central Processing Units (CPU) and a simulation program which consists of instructions. The CPU performs the calculations and logical decisions embedded in each instruction. Most instructions read data from memory (locations) and write the results back to memory.
- Memory contains the instructions and data.
The instructions can be divided in two major types:
- Instructions that are necessary to control the simulation of the game.
For example 1: the initial setup of the game as shown by the game "Paint Monsters" on an IPAD.
That data has to be translated to data understandable by the game simulation program.
For example 2: the answer of the game after each move by the simulation program.
This answer has to be entered on the IPAD which runs the game "Paint Monsters".
For example 3: instructions to manage the number of moves and the objective of each game?
- Instructions that are necessary to simulate the player
These instructions are the most difficult because they require to calculate based on the input (see example 1) to calculate the next move of the player (see example 2) based on the objective of the game, the layout of each game, the rules of the game specific on the rules of each type of paint monster and on the moves and results of previous games. These instructions define the strategy of the 'robot'
In order to chalenge who is the most intelligent we are going to play the game "Paint Monsters" with a human and with a computer program. That means computer program simulates a human player. In that case there is no human player involved but an opertor. The operator performs the moves as indicated by the computer program.
From a logical point of view you could say that a robot will play the game "Paint Monsters". The interface between the robot and the game is than an operator, which takes care for the communication between the two.
However to play an honest game neither the human player nor the engineer, who designed the computer program must have any knowledge in how to play the game "Paint Monsters"
When a human plays each game he will slowly learn, partly from his errors. As such his skill will improve. What the player should also do is to write down in how many tries he finaly solved each game. This allows for an honnest evaluation of the two players. To write down the total number of moves is an option.
What the engineer can do is to store all what he wants in the memory of the computer and use that data to make decissions on how to play the next game.
What the engineer can not do is to make modifications to the program, at least not make modifications to that part which make the decisions how to perform or to calculate the next move. What the enigineer also cannot do is to ask for advice over the World Wide Web. The human player is not allowed to do the same.
What the human player will experience at certain instances, when he performs a new game that also new monsters or targets (like bomb or arrow) are introduced. Such a new target can be a butterfly. Each time when a new monster is introduced normally there is a training game to explain the functionality of that new monster. What is important for the engineer that all that imformation will be available, so that in effect he can implement all the monster, in advance, during the design and initially testing of his simulation program. Those training games are not part of the evolution of either player.
My own prediction is that it is impossible to write one general program, when the designer of the computer program has on limited knowledge about what lies ahead of him, to finish all 1605 games, as implemented by the designers of the game "Paint Monsters". A more realistic goal is to write 1605 special purpose programs to finish each game. That is still not easy, because to define the specific strategies involved in each game is already tricky.
To write a computer program to play chess and beat a human player is much simpler if you have unlimited resources to calculate the next move. What makes this also simpler if the number of moves ahead you can calculate is also unlimited, including the number of processors. The point is that all these issues have nothing to do with the strategy to calculate the next move, which stays the same. In fact a computer which has more processors is not more intelligent.
Going back to the Paint Monster game it is much more challeging by two engineers to write each a simulation program and to try to find out which simulation is the most intelligent.
The question to ask is: if and to what extend AI can help us to understand or explain physical (chemical, biological) processes.
My understanding is, that all our understanding about the evolution of all natural processes comes from making observations and performing experiments. These observations and measurements are performed by using tools and these tools can include any type of measurement device like flow meters, pressure meters or a gas chromatography spectrometer or devices which use AI like deep learning. In all these cases the tools used should be tested and calibrated in order to demonstrate that what is indicated is accurate. From a principle point of view each tool can be handled as a black box, that means its internal functioning is 'not important'.
The question to answer is can AI be used to understand the details of a physical program.
When you consider a chemical process in its details, it is often a collection of chemical reactions which each take place depending about local temperatures, pressures or concentrations about the chemical elements involved. No AI can be used to describe these individual reactions.
In certain chemical reactions two photons are created. Photons are interesting because they can be polarised, which is consequence of the wave function of each photon. What makes certain chemical reactions specific of interest, is that when both photons are measured in the x direction, one photon will be +x and the other photon will be -x. That means the polarisation angle are each other opposite, the two photons are correlated. This is also called entanglement.
How can this correlation be explained. The question is can we use AI to explain this. I doubt that.
Suppose there are two explanations.
- The first explanation is that when the first photon is measured and the result is +x than immediate the state of the second photon becomes fixed as -x. When the second photon is measured the result will be -x.
This explanation requires immediate action a distance or a physical influence which moves faster than the speed of light.
- The second explanation is that immediate when the chemical reaction (collision) takes place and the two photons are created they are also correlated. That means the fact that both photons are correlated is a condition of the reaction. The two photon's are called L (for moving Left) and R (for moving Right). When the L photon is measured the result is -x. When the R photon is measured the result is +x. It does not matter which is measured first. The main assumption is that neither measurement influences the state of the other.
Which one is correct? IMO the second one because from a physical point of view, this is the simplest explanation.
Can AI help us to solve this physical problem? What you can do is to use all available litterature and count how many times solution 1 is indicated and how many times solution 2. Gives that the right answer?
IMO it is better to limit yourself to the documents who mention both solutions.
The central question of this reflection is: Can AI be used to improve our understanding of physical processes? To improve our understanding is primarily based on new experiments. To interpret these results is a human activity. To propose what to do next (and to make possible suggestions) is a human activity.
This is different than to compare the result of one experiment with the results of other experiments. Such a comparison can be automated.
If you want to give a comment you can use the following form