Heuristics and Frame Problems

Life Tips & Miscellaneous Navigation of this blog Travel and History Zen and Life Tips Machine Learning Digital Transformation Artificial Intelligence Deep Learning Economy and Business 

Heuristics and Frame Problems

Heuristics is a convenient or heuristic method used to solve a problem or make a decision on an uncertain matter when there is no clear clue to do so.

In contrast to heuristics are algorithms. For example, the formula for finding the area of a triangle is a good example of an algorithm, and the area of a triangle can always be found by applying the formula (base x height)/2.

Heuristics, on the other hand, refers to proverbs and sayings such as “hurry up and get around” and “just try it.

Most of the time, heuristics are used when we want to get a somewhat satisfactory or complete answer quickly and without much effort. When solving such uncertain decision-making problems, such as predicting future possibilities based on current conditions, it is important to consider probability. When we use the word “probability” in our daily lives, it is often used to express “prospects” such as the expectation of an economic boom or the probability that one side will win in a game. While ordinary probability should be judged objectively based on some kind of evidence, probability in this heuristics is mostly judged intuitively and is called “subjective probability.

This subjective probability is used theoretically in Bayesian theory. In this theory, the combination of relative probabilities (prior probability and posterior probability), rather than absolute probability values, is the basis of the theory, and the theory that we can approach more certain probabilities by taking steps has been formulated and utilized in various situations.

Also, since heuristics are not perfect solutions, they can sometimes lead to outrageous mistakes, and the subjective probabilities mentioned earlier can be far from objectively correct. This is called “bias”.

In response to such biases, slogans such as “throw away the stereotypes” that are not confined to heuristics are sometimes advocated, and heuristics are often perceived as unnecessary.

However, in real world cases, it is difficult to consider all the requirements in full (all the requirements need to be lined up in order to calculate probabilities), and even if they could be lined up, there are so many things to consider that it would take an enormous amount of time to calculate them. The future is too uncertain to draw definitive conclusions, and there are often more processes that require decisions based on heuristics.

This is particularly true of the artificial intelligence frame problem. This is the result of a thought experiment by philosopher Daniel Dennett, in which he conceived of a robot that would take the place of a human being in a room and perform a dangerous task: to remove a valuable work of art from a room that had been bombed.

At this point, Robot 1 realized that it should move the cart to take out the art, so it pushed the cart to take out the art from the room, but the bomb was planted on the cart and the bomb exploded. However, a bomb was placed on the dolly and the bomb exploded. This was a failure because Robot 1 understood the purpose of taking out the artwork, but did not understand the side effects of doing so (carrying out the artwork and simultaneously carrying out the bomb, which exploded).

However, if we create such an algorithm, it will start to think about all the things that may occur as a side effect, and it will continue to think infinitely. However, if you create such an algorithm, you will start thinking about all the possible secondary occurrences, and you will end up thinking endlessly, and the bomb will still explode. This is because there are an infinite number of possible side effects, and it would take an infinite amount of computation time to take them all into account. Translated with www.DeepL.com/Translator (free version)

Furthermore, based on the failure of robot No. 2, if we develop and operate robot No. 3, which has been improved so that it does not consider irrelevant matters in carrying out its purpose, it will continue to think infinitely in an attempt to identify all matters that are irrelevant to its purpose, and it will stop operating.

The frame problem can be defined in this way: “Unless the space to be considered is finite, we are forced to think about infinite possibilities. If you throw a problem of infinite possibilities at a machine, it will continue to work without stopping, resulting in a “runaway” state. In contrast, the human heuristic approach has the advantage of finding an answer to an infinite number of problems, no matter how uncertain the future may be.

As I mentioned earlier about some patterns of reasoning, the approach to breakthroughs in frame problems with heuristics described here is one of them, a non-deductive approach such as abduction.

In order to create a true artificial intelligence (that can solve any problem), it is essential to create such a heuristic algorithm.