We covered learning in AI in this post. After learning is complete, there is the reasoning, decision making and planning stage. In short, after an AI model extracts information from data and learns, it applies this knowledge for reasoning. Then it decides a course of action, which is the decision making part. And finally, it plans the consecutive actions, which is the planning part. Now lets explain each a little bit more:

We can classify reasoning into three categories:

Symbolic Reasoning

Probabilistic Reasoning

Neural or Learned Reasoning

Let’s explain each a little more:

Symbolic reasoning, which is also called as Logic-Based or Rule-based reasoning, is the oldest method. It operates by using techniques such as true-false statements, if-then logic, first order logic such as variables and quantifiers, knowledge graphs such as structured facts and relationships, logic solvers such as Prolog. It is the most straightforward, easily understood method and works well under well defined domains. Its disadvantages are that it cannot handle uncertainty or noise well and it is hard to scale (in AI, scaling means using more computation resources and more data, which, enables better performance of AI for a wider range of tasks, at least up to a point, after which the additional benefits start to decrease (as in law of diminishing returns) and therefore not practical)

Probabilistic Reasoning, which is uncertainty based reasoning. It uses mathematical, especially probability based methods such as Bayesian Networks, Markov Decision Processes, Hidden Markov Models, Probabilistic Interference. This Method is good at modeling real worl uncertainties, and widely used in applications in robotics, vision, natural language processing. The disadvantages of this method are that it needs a enough quantities of data and it needs to know prior probabilities. It can also get complex and computationally intensive.

Neural reasoning, is based on large amounts of data and neural networks, is the most flexible and can produce the best results under complex and ambiguous real life data. It naturally works well with deep learning. Since it is a flexible method, it has good scalability. Its disadvantage is that its results can be ambiguous and hard to interpret or even understand therefore accuracy is less guaranteed than the methods above.

And finally, planning is building a series of actions towards a goal. Therefore it is inherent part of, a subset of decision making. Reasoning and decision making continuously work during execution of planned actions.

By: A. Tuter

——————————————————

Terms of use:

Copying or republishing of any type of content from this website is not allowed without written permission from us. We make dated records of our posts, keep originals of our images.

The content in this website may contain errors, inaccuracies or be incomplete. User assumes all liability as a result of usage of content in our site. Also see our Terms page for more details.