A reference card for definitions of intelligence.
This entry attempts to summarize the definitions of intelligence that I am familiar with. It is the result of trying to organize my thoughts on the subject. When reading multiple sources from various fields that address the concept of intelligence, perspectives can differ. Sometimes, it even seems as though they are discussing entirely different topics.
Disclosure: Some of the links below are affiliate links. This means that if you click through and make a purchase, I may earn a small commission at no extra cost to you.
1. Chollet: Intelligence as learning efficiency
Definition: Intelligence is the efficiency with which a learning system turns experience and priors into general skill at previously unseen tasks
Proponent: François Chollet
Proposed metrics: Skill Acquisition Efficiency (SAE)
Perspective: Emphasizes generalization and the ability to learn new tasks quickly from limited data.
1.1. References
- Chollet: proposes a benchmark for intelligence, the ARC challenge
2. Hutter: Universal Artificial Intelligence
Definition: Intelligence measures the ability of an agent to achieve goals in a wide range of environments
Proponent: Marcus Hutter
Proposed metrics: Universal Intelligence Measure (UIM); a formal mathematical definition summing over all computable reward functions weighted by their simplicity.
Perspective: Provides a formal, mathematical framework based on algorithmic information theory to define and measure intelligence.
3. Schmidhuber: Compression progress
Definition: Intelligence is the ability to generate novel, interesting actions or data by maximizing future expected reward or compression progress
Proponent: Jürgen Schmidhuber
Proposed metrics: Measures based on Compression Progress; improvement in predictive models driven by intrinsic motivation and curiosity
Perspective: Highlights the role of intrinsic motivation, curiosity, and data compression in discovering new patterns and behaviors.
3.1. References
- Schmidhuber on consciousness: intelligence [and conciousness] is a byproduct of data compression
- AI Blog: Schmidhuber's AI Blog
4. Mainstream AI: Task-specific metrics
Definition: Intelligence is often operationally defined by performance on specific tasks or benchmarks without a unified overarching definition
Proponent: Various
Proposed metrics: Task-specific metrics such as Accuracy, F1 Score, BLEU Score, etc., depending on the task
Perspective: Focuses on empirical performance in narrow domains, emphasizing results on established benchmarks over a holistic definition of intelligence
4.1. References
- Russel, Norvig: Artificial Intelligence: a modern approach
5. Psychometrics: Intelligence Quotient (IQ)
Definition: Intelligence is the capacity for reasoning, problem-solving, planning, abstract thinking, understanding complex ideas, and learning from experience
Proponent: Psychometrics
Proposed metrics: Intelligence Quotient (IQ); standardized tests measuring various cognitive abilities
Perspective: Emphasizes quantifiable cognitive abilities and compares individuals to population norms, focusing on human intelligence assessment
6. Bostrom: Superintelligence
Definition: Superintelligence refers to an intellect that greatly surpasses the cognitive performance of humans in virtually all domains of interest
Proponent: Nick Bostrom
Proposed metrics: No specific metric; concept explores potential capabilities beyond human levels
Perspective: Focuses on the implications, risks, and ethics associated with creating intelligences that far exceed human cognitive abilities
6.1. References
- Superintelligence: Paths, Dangers, and Strategies
7. Ethology: Animal intelligence
Definition: Intelligence is the ability of an animal to adapt to its environment, learn from experiences, solve problems, and use tools
Proponent: Ethology
Proposed metrics: Behavioral tests assessing problem-solving, tool use, social learning, and communication abilities
Perspective: Studies intelligence across different species, emphasizing evolutionary adaptations and ecological contexts
7.1. References
- Wikipedia: Animal cognition
8. Human-like AI: Artificial General Intelligence (AGI)
Definition: Intelligence is the ability of an artificial agent to understand, learn, and apply knowledge in a general, human-like way across a wide range of tasks and domains
Proponent: Various
Proposed metrics: No standardized metric yet; Turing Test, general AI benchmarks, and evaluations across diverse tasks are used
Perspective: Aims to develop machines with general cognitive abilities comparable to humans, capable of understanding and reasoning across various contexts
8.1. References
- Wikipedia: Artificial General Intelligence
