Automation of decisions

It would be interesting to explain artificial intelligence in terms of "automation of decision making".

• Humans think and make decisions on the spot.
• Human beings verbalize "how to make decisions" in advance, and human beings make decisions accordingly.
  • New part-timers who do not understand the nature of the work
• Humans verbalize "how to make decisions" in advance, and the machine works accordingly.
  • Most programs, expert systems

The human being verbalizes in advance, "What is noteworthy in expressing the situation?

• Prepare a lot of "situation data" and "the right decision in that situation",
• The machine calculates in advance "how to combine the situation data to make a decision that is close to the correct answer,
• Decisions are made based on the results of these calculations. Supervised learning

The human being verbalizes in advance, "What is noteworthy in expressing the situation?

• The goal is to verbalize "what kind of condition is good."
• Prepare a lot of "situation data",
• The machine calculates in advance "how the combination of data of interest is likely to be in good condition",
• Decisions are made based on the results of these calculations. Reinforcement Learning

The human being prepares a tremendous amount of "situation data" and "correct judgment in that situation" in advance,

• The machine calculates in advance "what is important out of the situation data" and "how to combine them to make a decision that is close to the correct one,
• Decisions are made based on the results of these calculations. : Deep learning, image discrimination by DCN, etc.

The "what kind of state should it be" is verbalized in advance by humans,

• Prepare a tremendous amount of "situational data,
• The machine calculates in advance "what is important out of the situation data" and "how to combine them to achieve a good situation".
• Decisions are made based on the results. ： Deep learning + reinforcement learning, Atari games could be played better than humans.

Deep Learning has eliminated the need for "humans to verbalize 'where to look'", although It's conditional on "if you have a huge amount of data." And without data, of course, a human being would have to verbalize it, The need to "collect data" in the first place has not gone away.

This page is auto-translated from /nishio/判断の自動化 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.