Original article was published by Sreshta Putchala on Artificial Intelligence on Medium
Trying to define “intelligence” is hazardous, but we can broadly consider it to solve problems. There are two approaches to solve — the first is to solve problems using first principles and logic. The second is to harness knowledge gleaned from experience (data) or from other agents.
There are two approaches for AI — a cognitive and an engineering approach. The cognitive approach seeks to understand how intelligent behavior arises. In the engineering approach, the goal is to construct smart machines. Al provides a computational platform for these ideas. In both, the ideas manifest themselves as computer programs. These programs draw ideas from many disciplines — computer science, philosophy, psychology, economics, mathematics, logic, and operations research.
We often refer to an autonomous program that senses its environment and acts independently in a goal-directed manner as an agent. The agent chooses various actions subjected to the constraints to reach the desired state that satisfies the agent’s goals. This approach is the first-principles approach where an agent solves a problem by reasoning about actions, exploring combinations, and choosing the ones that lead to the goal. Reinforcement learning extensively uses these approaches. However, there is a danger of the combinatorial explosion that the agent has to contend with. We then refine and modify the search-based policy to include heuristic knowledge, and then we move towards ways to deploy more explicit forms of domain-specific knowledge. These will consist of logic and reasoning, memory structures and the exploitation of experience, deeper knowledge in models and ontology, and the relation between language and knowledge.
Humans use language to communicate and represent knowledge, for example, in an article or a blog. Formal reasoning and argumentation are also often done using language. There are various aspects of language, including text processing (which gained prominence with online information). For example, in a chatbot, a computer, to interact meaningfully, it will need to access the meaning of what is said. The specific formalism used is not the important thing. The idea of sentences is expressed in a language used to model the concepts in the underlying domain.
AI draws upon the strong mathematical and philosophical base of logical reasoning. It looks at ways to structure knowledge to connect.