# Can Fuzzy Logic Control Self-tune as Other Learning Systems Do?

This inquiry appeared in the comments to one of the answers of this question, but is actulaly not related to emotional intelligence, so it is reproduced here as a separate question

I thought (maybe incorrectly) fuzzy logic was mostly used in sort of pre-programmed control systems, as opposed to AI or learning systems that regress on some datasets.

Programming, Pre-programming, and Parameterized Processing

Regression during training and distinctly prior to use is a form of programming just as the previous PLC (programmable logic controller) had been.

Consider the input of a compiler as the parameters that drive the arithmetic and control on a VLSI processor. That is the way Shannon, Turing, and von Neumann thought of it. That is the conceptual core of CPU design today, even with libraries, frameworks, caches, and hardware acceleration added.

The only difference between a list of execution instructions, when though of as parameters given a CPU (central processing unit) and those learned through training an artificial network is structure. The parameters learned during training are also directives for processing but through a different mechanism. They are like application configuration but with greater general flexibility.

Branching Rules as Data

One might argue that CPU controllers do more than arithmetic. They branch based on predicate logic. But that is true in artificial networks through a more streamlined mechanism. For instance, a zero in a learned parameter tensor effectively shuts off signaling from an activation function result in network layer $$i$$ to the addend in a sum presented to the activation function input in layer $$i + 1$$.

Furthermore, the von Neumann architecture is based on the idea of storing programs in the same type of memory that data resides. Early processor designs partitioned the two segments to hardware prohibit accidental program corruption.

Turing Complete

Some of the more contemporary bidirectional LSTM networks and recursive networks and other networks derived from recurrent networks are Turing Complete. They can theoretically learn an arbitrary algorithm, and much research is being done along that path.

Cognition and Fuzzy Logic

Such is one of the possible development paths toward the simulation of cognition and true depth of conversation in domain specific natural language applications. This is true conversation with comprehension when listening and expression of ideas when talking or typing back, far beyond chatbots with no real competency.

Cognition is necessary to many of the more creative capabilities of the human mind, and it is still not clear that deep networks will be the more optimal and pragmatic solution to that challenge over rules based systems, especially one with probabilistic rule handling, which is what fuzzy logic containers do.

More information is provided at How important is fuzzy logic for Artificial Intelligence?.