What is the name of an AI system that learns by trial and error?



Imagine a system that controls dampers in a complex vent system that has an objective to perfectly equalize the output from each vent. The system has sensors for damper position, flow at various locations and at each vent. The system is initially implemented using a rather small data set or even a formulaic algorithm to control the dampers. What if that algorithm were programmed to "try" different configurations of dampers to optimize the air flows, guided broadly by either the initial (weak) training or the formula? The system would try different configurations and learn what improved results, and what worsened results, in an effort to reduce error (differential outflow).

What is that kind of AI system called? What is that system of learning called? Are there systems that do that currently?


Posted 2018-10-03T16:44:19.617

Reputation: 195

1You have asked the same question twice here, just with a different title. Please don't do that. We should close one of the questions - please suggest which one. – Neil Slater – 2018-10-03T19:11:50.340

1There is a subtle difference between the questions. One assumes a more traditional learning process before initial implementation and then refinement of that learning through use. The other question asks about the "trial and error" process where the machine may make random or educated guesses to try and optimize its results. – SchroedingersCat – 2018-10-03T19:41:17.450

You need to make the difference clearer then. In practice they look the same thing - phrasing like "the system would try different configurations and learn what improved results, and what worsened result" here, and " The algorithm CONTINUES to learn from its own empirical data," in the other question are not really different. It is no surprise that you have been answered with Reinforcement Learning in both cases. – Neil Slater – 2018-10-03T21:08:58.727

Thank you Neil. I try to make my questions here quick to read and answer if possible, in this case it took two parts to get to the desired result. Through both questions I got the level of detail I needed. – SchroedingersCat – 2018-10-04T13:20:59.000



I believe "Reinforcement Learning" is the term you are looking for (as mentioned by others as well) but keep in mind that the scope of your problem falls under the section of AI that is called Search.

Search algorithms are based upon experimenting with different actions (decisions) and selecting the one that minimizes an arbitrary cost function (reward), given the current and past problem states.


Posted 2018-10-03T16:44:19.617

Reputation: 259


Near solution to your problem definition is reinforcement learning. You can define some reward using the objective function and define some possible state space for the machine and finally solve the problem by reinforcement learning techniques (near to trial and error by learning the preferences).


Posted 2018-10-03T16:44:19.617

Reputation: 1 020


I think any learning algorithm probably uses trial and error and analysis of the results with the ultimate goal of maximizing utility.

It seems that the recent milestones in AI fall under the general umbrella of machine learning, which includes all forms of reinforcement learning. Essentially, any learning algorithm is using some form of statistical analysis.

  • For an umbrella term, I've been using "learning algorithm"

However, there is also a venerable history of less capable adaptive systems such as self-organizing networks. (See also optimal control.)


Posted 2018-10-03T16:44:19.617

Reputation: 5 886