Intelligence – The Swiss Definition

In 2006, two AI researchers in the Swiss Alps — a PhD student and his supervisor— grew frustrated with the chaotic array of AI definitions. They decided a thorough inventory was needed. The PhD student undertook the task, meticulously reviewing hundreds of intelligence definitions and selecting 71. The sources spanned dictionaries, encyclopedias, studies, textbooks, and essays, covering fields like psychology, education, computer science, and philosophy.

With Swiss-like diligence, the PhD candidate identified three recurring key attributes:

“Intelligence:

• is a property that an individual agent has as it interacts with its environment or environments.
• is related to the agent’s ability to succeed or profit with respect to some goal or objective.
• depends on how able the agent is to adapt to different objectives and environments.”

The two researchers distilled these 71 definitions into one based on the three attributes:

“Intelligence measures an agent’s ability to achieve goals in a wide range of environments.” (Legg, S., & Hutter, M. (2007). A collection of definitions of intelligence. Frontiers in Artificial Intelligence and Applications, 157, 17. ArXiv-Download)

Let’s unpack this. The definition identifies the agent (who), specifies the environments (where), clarifies the goal (what for), and highlights what intelligence measures: the agent’s ability (how much).

Intelligence, in essence, is the means to achieve goals.

Visualize an agent starting at a point, aiming for a goal—from status quo to target state. “Achieve” denotes the journey; intelligence enables its completion through effective actions. Thus, “achieve” means “doing what’s needed to reach the goal.”

This “Swiss definition” offers three insights:

First, a goal is a state within a system. Altering the system can make it more attainable. Instead of asking, “What must I do to reach B?” it is more intelligent to ask, “What needs to happen within the system for B to occur?” This can save effort and prevent setbacks.

Second, intelligence is relative, shaped by environments, goals, and agents. Anyone can be intelligent in their domain—kitchen, factory, or ministry. This explains why we are in awe of other people's achievements in unexpected situations.

Third, intelligence is universal. Its abstract terms—agents, goals, environments, achieve—apply to all scenarios, helping to understand even complex topics in artificial intelligence and its future descendants.

Another lesson from this Swiss endeavor:

It’s worth starting with a clear, practical definition. Shane Legg, the doctoral student who co-crafted this definition, became a leading AI researcher and co-founder of DeepMind, applying his insights as Chief AGI Scientist at Google DeepMind to advance AGI and ASI. His PhD supervisor, Marcus Hutter, a professor at the Australian National University in Canberra, co-authored the AIXI model and collaborates with Google DeepMind, significantly advancing AGI and ASI research.

Obviously clear, rigorous definitions can launch real-world breakthroughs.

PS: Even now, the ‘Swiss definition’ stands unchallenged — perhaps because few have bothered to repeat that exhaustive inventory. Yet François Chollet having scrutinized the ‘Swiss definition,’ devised an even more detailed and formalized definition of intelligence.

Sources:

Chollet, F. (2019). On the measure of intelligence. arXiv preprint arXiv:1911.01547. ArXiv-Download.

Legg, S., & Hutter, M. (2007). A collection of definitions of intelligence. Frontiers in Artificial Intelligence and Applications, 157, 17. ArXiv-Download