The AI Family Tree

Think about how you might teach someone to play chess versus how you might teach them to recognize a face. For chess, you can write down specific rules: "knights move in an L-shape." For recognizing a face, it is much harder to write a rule; you just have to see enough examples to build an intuition. The history of AI is essentially a debate between these two approaches—teaching computers the rules or letting them learn from experience.

The first branch of the family tree is Symbolic AISymbolic AI. This approach dominated from the 1956 Dartmouth Workshop1956 Dartmouth Workshop until the 1990s. Scientists believed that if they could describe the world in precise logical symbols, a machine could "think" by manipulating those symbols. A famous success from this era was Deep BlueDeep Blue, the chess computer that defeated Garry Kasparov. It didn't "learn" chess in the modern sense; it used brute force calculation based on rules humans provided.

The second branch, which dominates today, is the "Connectionist" approach, or neural networks. Instead of giving the computer rules, we give it data and let it discover the patterns. This shift from "logic" to "learning" is what enables modern breakthroughs like ChatGPT.

You might wonder where current technology fits in this lineage. Despite the impressive capabilities of modern systems, every AI that exists today is classified as Narrow AINarrow AI. These systems are brilliant at specific tasks—predicting protein structures or writing code—but they are brittle. They cannot transfer their knowledge to new, unrelated domains. A chess-playing AI cannot play checkers without being reprogrammed.

This stands in contrast to Artificial General Intelligence (AGI)Artificial General Intelligence (AGI), which refers to a hypothetical system that possesses human-level cognitive abilities across all domains. While AGI is the goal for companies like OpenAI and Google DeepMind, it is critical for governance professionals to remember that we are not there yet. We must govern the specific, limited tools we have, rather than the sci-fi entities we imagine.

The path of AI progress has not been a straight line. It has been a cycle of boom and bust. When researchers in the 1960s promised machines that could do any work a man could do within twenty years, they set expectations they could not meet. The resulting disappointment led to funding cuts known as "AI Winters."

To understand this, consider the Lighthill ReportLighthill Report of 1973. It was a devastating critique commissioned by the British government that concluded AI had failed to deliver on its grand promises. It triggered a massive reduction in support for the field. Another winter followed the collapse of the Japanese Fifth GenerationJapanese Fifth Generation computer project in the 1980s.

For a governance leader, these winters are a cautionary tale. They remind us that hype is dangerous. When we overpromise what AI can do—selling it as magic rather than math—we risk a backlash that can stall genuine progress.

Understanding this family tree helps you ask the right questions. If a vendor tries to sell you a "cognitive solution," you can now ask: "Is this a symbolic system following fixed rules, or a machine learning system based on probabilistic patterns?"

If it is symbolic, you govern it by auditing the rules. If it is machine learning, you govern it by auditing the data. By correctly placing a system on the AI family tree, you can choose the right tools to manage it. Now that we have mapped the family tree, we can look deeper into the specific types of intelligence these systems exhibit in our next article, Types of AI Intelligence.