Intelligence

As explained in Deep Learning with Python, Third Edition

The capacity to understand, learn, and adapt to tasks and environments, discussed here as the broad property AI aims to approximate.

Intelligence is the ability to use past experience and innate prior knowledge to handle novel, unexpected situations. It works by finding precedents, forming analogies, and applying abstract concepts rather than merely repeating learned responses. section 19.3.1 section 19.3

Making the unfamiliar recognizable

A person seeing a jet plane for the first time might call it a large, loud metal bird that does not flap its wings. A car becomes a horseless carriage; electricity becomes water flowing through a pipe. These analogies connect unfamiliar situations to reusable concepts. section 19.3.1

Figure 19.7 A kaleidoscope produces rich (yet repetitive) patterns from just a few beads of colored glass.

The kaleidoscope image represents intelligence as recombining a relatively small set of meaning-bearing abstractions into seemingly endless patterns. section 19.3.1

The kaleidoscope hypothesis treats novelty as patterned recombination: the world appears endlessly complex, yet its parts share similarities and structural relationships. Intelligence depends on recognizing and exploiting those relationships. section 19.3.1

A practical target for intelligent systems is broader adaptation rather than narrow task completion:

• Measure how many sufficiently unfamiliar tasks a system can handle from a fixed amount of experience and prior knowledge. section 19.3.4
• Test tasks that were not programmed or anticipated, preventing a system from succeeding through task-specific shortcuts. section 19.3.4
• Use ARC-AGI-style grid tasks, where examples demonstrate a transformation and the system constructs an output for a new input. section 19.3.4

Figure 19.9 An ARC-AGI task: the nature of the task is demonstrated by a couple of input-output pair examples. Provided with a new input, you must construct the corresponding output.

The ARC-AGI figure makes the adaptation test concrete: infer the rule from input-output examples, then apply it to an unseen input. section 19.3.4

Two proposed ingredients are abstraction acquisition and abstraction recombination. Their interaction determines which abstractions a system can manipulate and how effectively it can combine them; search and symbols are identified as possible complements to parametric abstractions learned through gradient descent. section 19.4

How Deep Learning with Python, Third Edition covers it

How deep it goes

Defines intelligence through analogy, generalization, and abstraction recombination, then points toward search and symbols without detailing implementation.

Examples

Uses analogies, the kaleidoscope, and ARC-AGI input-output grid tasks to illustrate recognition, recombination, and adaptation.

What it assumes

Assumes awareness of deep learning while defining intelligence from past experience and innate prior knowledge.

What you can do after

Evaluate intelligence by measuring performance on unanticipated tasks relative to fixed experience and prior knowledge.

How it connects

Connects intelligence to deep learning, analogy, abstraction, generalization, search, and symbols.

liveBookmanning.com

Read Deep Learning with Python, Third Edition on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Sutskever's ListIntelligence is the capacity to achieve goals or perform cognitive tasks, considered here as a property that behavioral metrics may approximate without establishing its inner nature.
• 

  liveBookmanning.com
  Inside AIIntelligence is the cognitive capacity to make decisions, solve problems, and respond appropriately to situations.
• 

  liveBookmanning.com
  Deep Learning with R, Third EditionIntelligence is the ability to use available information and past experience to produce successful behavior in uncertain and changing situations.
• 

  liveBookmanning.com
  How Large Language Models WorkIntelligence is the broad and disputed capacity to perform cognitive tasks such as reasoning, learning, and adapting to situations.
• 

  liveBookmanning.com
  The Art of AI Product DevelopmentIntelligence is the capability of an AI system to infer, generate, decide, or perform tasks that require adaptive reasoning.

Compare how 6 books explain intelligence →

Related concepts

• augmented intelligenceused by · Augmented intelligence is the use of computational systems to extend human capabilities while supporting rather than replacing human expertise or judgment.
• abstractionhas part · Abstraction is a simplified representation that exposes relevant properties or behavior while omitting lower-level details unnecessary for a given purpose.
• psychometricsmeasured by · Psychometrics is the field concerned with the theory and methods of measuring psychological attributes, abilities, and behavioral constructs.

This explainer was generated by Manning's knowledge graph from the text of Deep Learning with Python, Third Edition. The section links point to the parts of the book it draws on.