4 Important Concepts to Master Artificial Intelligence
Introduction
Artificial Intelligence (AI) is a sub-field of computer science that focuses on designing intelligent systems that can act rationally, take decisions, and solve problems (similar to human beings). In this article, we discuss four important concepts that one needs to learn in order to master artificial intelligence.
1. Probability
Most real-world problems do not have deterministic solutions and the solutions are dependent on the particular situation that one is in. To model such problems, it is necessary to understand the pros and cons of each event and the probability of each event materializing.
For example, consider the case that you are planning to buy non-refundable tickets to an open-air concert. For you to be able to fully enjoy the concert, you want the weather on the day of the concert to be sunny. To make an educated decision, you have to take into account the probability of the weather being sunny, cloudy, or rainy and then evaluate your ‘overall experience’ for each one of the weather conditions.
Without a good knowledge of probability, it is difficult to fully grasp the complete power of artificial intelligence.
2. Search and Exploration
Search is one of the key focus areas in AI. Search is essential for exploring the desired solution space to find good (and even optimal) solutions to the problem under investigation. Some of the simplest examples of search are Breadth First Search (BFS) and Depth First Search (DFS) in a tree. Similarly, for finding paths in a graph, the most basic approaches are Dijkstra’s and Floyd’s algorithm.
More informed search approaches include A* and hill-climbing. In a hill-climbing approach, the gradient of the function is used to determine the direction of ascent. Even if hill-climbing approaches do not yield the globally optimal solution, they definitely provide a locally optimal solution. A more sophisticated and advanced approach is to formulate problems as constraint satisfaction problems.
3. Planning and Reasoning under Uncertainty
Planning refers to the task of coming up with a sequence of events toward achieving a goal. One of the most important decisions while designing a planning algorithm is representation of states, goals, and actions. Additionally, sometimes complete knowledge of the exploration space is unavailable (e.g., MARS rover cannot know the terrain of Mars priori) and/or only limited or partial observations can be made (e.g., one is able to only decode part of the messages exchanged in a covert communication). In such scenarios, AI algorithms need to make decisions under the realm of uncertainty. Knowledge of probability is critical for solving problems in such situations as one needs to take into account the chance of certain events happening.
Some important concepts to learn in this regard are Bayesian networks, Markov Decision Processes (MDP), and Partially Observable Markov Decision Processes (POMDP).
4. Statistical and Reinforcement Learning
With copious amounts of data becoming available in a variety of different fields, learning based approaches are becoming increasing popular in the field of artificial intelligence. Both unsupervised and supervised algorithms are being widely used in AI and it is necessary to have good understanding of these approaches to successfully design good solutions to real-world problems. At the highest level, this can be viewed as the sweet spot where machine learning and AI intersect.
For example, it is necessary to grasp important concepts such as decision trees, maximum likelihood estimation, regression models, kernels and K-nearest neighbors. After learning these statistical approaches, one can then take a deeper dive into reinforcement learning, an approach that bases its next decisions or actions based on the rewards (both immediate and cumulative) that have been received.
