Approaches to Artificial Intelligence: Five Tribes & Four Levels
Explore the diverse landscape of artificial intelligence, from the five main algorithmic approaches ("tribes") that have shaped machine learning to the four levels of AI intelligence. Understand the different strategies and classifications used in developing intelligent systems.
Approaches to Artificial Intelligence: Five Tribes and Four Levels
Artificial intelligence (AI) is a broad field encompassing many approaches to building intelligent systems. This section explores different perspectives on AI development—the "five tribes"—and the four levels of AI based on machine behavior.
The Five Tribes of Machine Learning
Five main algorithmic techniques or "tribes" have historically shaped the development of machine learning:
- Symbolists: Focus on symbolic reasoning and rule-based systems. These AI systems use symbols and rules to represent knowledge and reason about it. They work well in situations where the rules are known and clearly defined. They are often used for expert systems and other applications requiring logical deduction and decision-making.
- Connectionists: Build artificial neural networks modeled after the human brain. These AI systems use artificial neural networks to learn from data, which are composed of interconnected nodes that process information in parallel. They excel at tasks involving pattern recognition, image and speech processing, and other complex tasks requiring pattern recognition.
- Evolutionaries: Use evolutionary algorithms (inspired by natural selection) to find optimal solutions. These AI systems use evolutionary algorithms that mimic natural selection and genetic mutations to improve their ability to solve problems over time. They are suited to optimization problems and problems where the solution is complex and not easily described using logic.
- Bayesians: Employ probabilistic methods and Bayesian inference to deal with uncertainty. These AI systems use Bayesian methods to model uncertainty and update their beliefs based on new evidence. They’re very effective when dealing with incomplete or uncertain information.
- Analogizers: Use kernel machines to identify patterns and solve problems by analogy. These AI systems identify patterns and use similarity measures to determine appropriate solutions. They are often used in recommendation systems and other applications that must work with a large amount of data that can be used to draw conclusions.
Four Levels of Machine Intelligence
Based on their capabilities, AI systems are broadly classified into four levels of intelligence:
- Reactive Machines: These are the most basic AI systems, reacting to inputs without memory or learning capabilities. Deep Blue is a classic example.
- Limited Memory: These machines use data from the past to inform their current decisions, but that data is not stored persistently. They typically don’t “remember” past interactions in the same way that humans do. Self-driving cars are an example of limited memory AI.
- Theory of Mind: This represents a higher level of intelligence where machines can understand and respond to human emotions and intentions (not yet achieved). Understanding human thoughts and feelings is something that is still being developed.
- Self-Awareness: The highest level, where machines possess self-awareness and consciousness (also not yet achieved). This is the most challenging aspect of AI and represents a significant leap beyond current technology.