What is the frame problem?About important issues in AI | AI Specialized News Media AInow

When you first learn the word "artificial intelligence" or "AI", many people may first imagine intelligent robots such as "Doraemon" or "Astro Boy".

The robot that behaves like a human is supported by Japan's proud anime culture and is famous worldwide.

However, unfortunately, robots such as "Doraemon" and "Astro Boy Atom" are unfortunately not currently being developed.

The reason why such robots have not been developed are that there is an important problem called "frame problem."

So, this time, I will introduce the "frame problem", an important issue in artificial intelligence.If you read this, you may find out why Doraemon is not born.

▶【関連記事】ドラえもんをつくるという夢の現在地 ー研究の壁を超えて>>

▶【関連記事】AI・人工知能が搭載されたロボットまとめ>>

▼ Click here for those who do not know more about AI (artificial intelligence)

table of contents

What is the frame problem?

The definition of "artificial intelligence (AI)" varies, but to say roughly

It can be said.

Autonomous driving and chatbot will be easy to imagine.

This AI is mainly classified as "weak AI" and "strong AI".

Almost all AIs currently spread apply to "weak AI".The debate based on this classification is the "frame problem".

The frame problem was proposed in 1969, the first AI boom, by John McCarsey, a computing machine scientist, the name of the AI, and Patrick Hayes, a cognitive scientist.

What is "frame problem"?

Is a problem.

To understand the meaning of this problem, words are words, such as "What is a finite information processing ability?" "What kind of behavior is a robot that behaves?"It will be easy to understand when disassembled.

"Bomb and robot" are expressed in an easy -to -understand manner in focus on these three elements that have been disassembled.

From here, I will introduce what the "bomb and robot" are.

Daniel Denette's "bomb and robot" thinking experiment, developed in a paper "Cognitive Wheel Artificial Intelligence and Frame Problem", is a very famous example that expresses frame problems in an easy -to -understand manner.

The characters are "Dr." and "Robot (R1) invented by Dr.".The purpose of this thinking experiment is for the robot to take out his battery from the warehouse.

Dr. gives instructions to R1 Robo.

Dr. "Because the battery is going to run out soon, bring the battery from the warehouse. However, say that there is a time bomb on the battery."

R1 Robo took out the battery from the warehouse, but brought it to a timely bomb with the battery.And the explosion.

Dr. will challenge the development of the R1-D1 so as not to repeat the same mistakes again.

Dr. has added a new feature to the developed "R1-D1".It is to calculate everything that can occur in the intended meaning.

In detail, in addition to the original purpose, "Remove the battery from the warehouse", for example, "What if a time bomb is on top?", "What if the battery and the bomb are on the same wagon?"Calculate everything that can occur in the process of achieving the purpose, and deal with each.

That way, the idea is that you can process even if a bomb is on the battery.

Dr. makes the same instructions as last time a robot.The robot goes to the warehouse and tries to remove the battery from the warehouse with his own hands, but explodes because there are too many problems to deal with.Since the issues that could occur in the secondary side were infinite, the amount of calculation became enormous, and as a result, the time bomb was up and the time was up.

Dr. will not give up and challenge the development of "R2-D1".

Dr. who decides that the previous problem was to deal with "unwanted problems" add new features.

It is to distinguish between the relevant and unrelated meaning.

In other words, the original purpose of "removing the battery from the warehouse" and the meaning that has nothing to do with it ("What is the color of the wall?", "What is the radius of the battery?")However, we ignored unrelated things and performed the purpose.This robot is "R2-D1".

Dr. gives instructions in the same way.

Dr. "Because the battery is going to run out soon, bring the battery from the warehouse. However, say that there is a time bomb on the battery."

Then the R2-D1 Robo stands in front of the warehouse and explodes.

It is thought that nothing was done as a result because we continued to calculate what to ignore, that is, what the conditions with unrelated meaning were.

In this way, this thinking experiment expresses that no matter how flexible it is to make a robot, it is not possible to deal with all the problems that can happen.

Now, read the example of "bomb and robot", and then look at the frame problem from another perspective.

Hitoshi Matsubara, a Japanese information engineer, has an approach that connects tacit knowledge and frame problem.Before that, I will introduce "tacit knowledge".

In fact, the robot that appears in the "bomb and robot" has the name of the name.

"R" of "R1-D1" stands for "ROBOT", but "D" means "DEDUCE", that is, "deduction".

The robot in this thinking experiment means deductively dealt with the problem in a completely blank state without any rule of thumb from the beginning.

The opposite of "deduction" is "induction".In other words, it means thinking based on various rules.Based on these, let's connect the frame problem with the "tacit knowledge".

First, why can people move when they face unknown problems?

For example, let's replace the robot part of "bomb and robot" with humans.Probably in the case of humans, I think that I will hurry the time bomb and take out the battery from the warehouse.

However, what can be done, however, is that the empirical knowledge of living is naturally integrated in the head, and the use of that knowledge may be dealing with various problems.From.

This knowledge is called "tacit knowledge."The presence or absence of "tacit knowledge" is a very important concept in distinguishing humans and computers.

In addition, the "tacit knowledge", which is integrated with the knowledge engraved by the gene and the rules of experience accumulated from everyday experience, is considered to be enormous.

The idea is that humans are simulating frame problems because humans are dealing with all issues after affirming the existence of tolerant knowledge in the person.

I personally think that it is very interesting to think about frame problems by relatively inductive people and deductive robots.

Now, when you look at the frame problem in detail, let's look at what this problem shows in the real world.Let's consider the frame problem closer.

How to solve the frame problem?

Here are three elements that are considered necessary to solve the frame problem.

1. 計算量の問題
2. コグニティブ・ホイール
3. 暗黙知の計算機

I will explain in detail each.

John McCarthy and Patrick Hayes, who first advocated the frame problem, stated that the essence of the problem was calculated.

The original definition states that "it is troublesome to describe an event that does not change when trying to describe (that is, a computer) on a calculator (that is, a computer) in a logic."

Therefore, the solution approach needs to be reduced in consideration of "description+processing".

If you take a blackboard as an example, it is better to write a small blackboard as small as possible.In other words, it is desirable to solve the versatility and calculation issues that are referred to in computer science.

As mentioned above, the processing has been solved a small amount, but it is still said that there are still many descriptions.The GPT-3, probably closest to the tacit computers, is also famous for having many parameters.

The "cognitive wheel" in the title of the dissertation means "invention of wheels", one of the historical inventions.

The invention of wheels is considered to be the oldest invention in human history, and it is a well -known story that the Ubaid Dynasty of Mesopotamia civilization was used as a portrown portrack in the 5,000 BC.

It is believed that the wheels were invented by humanity as a model of other creatures.

Similarly, Daniel Denettes implicitly shows the direction of artificial intelligence as an independent calculator, rather than a model of humans.From there, the discussion was active in the "weak AI" and "strong AI" introduced at the beginning.

The focus is on whether or not the implicit knowledge can be performed, that is, an artificial intelligence that can be treated based on the rules of experience on a computer, that is, to realize artificial intelligence that can deal with any problem, that is, to realize artificial intelligence that solves the "frame problem".It will be.

So far, the process has not yet been clear how humans select appropriate implicit knowledge and resolve frame problems in a pseudo manner.

Therefore, it is currently not possible to express tacit knowledge with a calculator, and it is considered difficult to develop a robot that can break through the frame problem.

ただ現在では深層学習の研究も活発に行われているため、目的を遂行するための絞り込みは少なからず改善されているIt can be said.

The case where AI won the Go World Champion is a specific example.

At the end

生き物はみな、日常でもフレーム問題に直面しているIt can be said.

For example, "What to do today's dinner", "How to write a university report", "What do you want to do in your future?"

In such a case, most people have solved them by gathering already information, such as looking at recipes posted on the Internet, collecting content that they want to write from the references, and choosing the course that the adults around have walked.More.

In terms of "frame problem", it does not solve the problem.Because it only solves the problem in the frame.

However, I think everyone has solved the problem with a quick decision from the rule of thumb.Therefore, it can be said that all creatures have solved the frame problem or not.

Now, when thinking about how to solve the problem in front of you, it may be a good idea to measure the presence or absence of a frame.

◇ Ainow intern student ◇ We are also sending on Twitter.◇ I write my blog while doing an intern with Ainow and studying writing.