FIELD GUIDE // AI → AGI → SUPERINTELLIGENCE

What is SI?

Superintelligence is a hypothetical intelligence that would substantially exceed the best human cognitive abilities across virtually all relevant domains. Here, SI means superintelligence—not the International System of Units.

What would that mean in practice? How does it differ from today's AI? And which questions are still open?

Read the full explanation of superintelligence →

AI → AGI → ASI / SI
Telemetry // Dispatches

Latest News & Opinion

All Articles →
§ 1 — Definitions

One field. Different levels of capability.

AI

Artificial Intelligence

Computer systems that perform tasks associated with intelligence: recognising patterns, generating language, planning and making decisions.

AI already exists. Some systems are specialised; general-purpose models work across a wider range of tasks. Being extremely capable at one activity does not establish general intelligence.

See how AI differs from superintelligence.

SI

Superintelligence

A hypothetical intelligence far beyond the best humans across virtually all relevant cognitive domains.

The threshold is broad superiority, not a single impressive skill. In AI discussions, artificial superintelligence (ASI) usually names the same idea while making its machine origin explicit.

Neither the arrival of SI nor its timing is established.

Illustration of two cartoon robots above a human finger touching a robotic finger, representing humans and the AI systems they build.
Image: “AI Humans and Robots” by Bovee and Thill, via Wikimedia Commons, licensed under CC BY 2.0.
§ 2 — The map

AI → AGI → SI → CASI

A way to distinguish breadth and performance—not a guaranteed development path.

01 // The broad category

Artificial intelligence

Specialist and general-purpose systems

A chess engine and a language model are both AI, but their ranges of ability differ. Useful work in several fields does not automatically mean human-level generality.

ESTABLISHED Uneven capabilities
02 // General capability

AGI

Artificial General Intelligence

Broad, flexible intellectual competence, often defined with reference to human-level ability. What counts as sufficient breadth and reliability remains disputed.

CONTESTED Definition & threshold
03 // Beyond human level

Superintelligence

SI / Artificial Superintelligence

Substantially better than the best humans across virtually all relevant cognitive domains. No broadly accepted demonstration establishes this general capability.

HYPOTHETICAL Pathways uncertain
04 // A speculative extension

CASI

Conscious Artificial Superintelligence

A superintelligence that would also have subjective experience. Whether machines can be conscious at all, and how anyone could tell, are open questions. Read the primer or compare CASI vs. ASI.

SPECULATIVE Consciousness unresolved
§ 3 — Explore the questions

A little clarity goes a long way.

More capable AI could help with science and engineering, but capability does not settle questions of control, access or accountability. Start with the distinctions; then look at the mechanisms and safeguards.

  1. Elon Musk’s $25 Trillion Superintelligence Output Forecast

    Musk predicts software-based digital superintelligence will expand global GDP by 20% to 30%, doubling white-collar productivity.

  2. What is ASI and What is CASI? Core Definitions

    Foundational definitions distinguishing superhuman cognitive capability (ASI) from subjective machine consciousness (CASI).

  3. CASI vs. ASI: Comparative Analysis

    A head-to-head comparison of Conscious Artificial Superintelligence (CASI) and standard ASI across architecture, detectability, and safety risks.

  4. What Is Superintelligence?

    A clear explanation of superintelligence: what SI means, how it differs from AI and AGI, and why its capabilities, risks and timelines remain uncertain.

  5. Definitions of Superintelligence, Compared

    Six influential definitions, from I. J. Good (1965) to Google DeepMind’s Levels of AGI (2023), side by side: what they agree on and where they differ.

  6. What Is Artificial Superintelligence (ASI)?

    What ASI means, why it overlaps with superintelligence, how it differs from AGI, and the open questions about development, safety and governance.

  7. What Is AGI?

    Artificial general intelligence explained: breadth, transfer, human-level capability, disputed definitions and its relationship to superintelligence.

  8. AI vs Superintelligence: What's the Difference?

    Compare AI, AGI and superintelligence by breadth, capability and evidence. Includes a plain-English table and examples of what the labels do—and do not—mean.

  9. What Is an Intelligence Explosion?

    The intelligence-explosion hypothesis explained: recursive self-improvement, hard and soft takeoff, possible bottlenecks and why rapid progress is not guaranteed.

  10. What Is AI Alignment?

    AI alignment explained through intended goals, value disagreements, reward hacking, corrigibility and oversight, with examples and limits of current approaches.

  11. What Is AI Safety?

    A practical guide to AI safety: reliability, robustness, misuse, security, alignment and evaluation, from existing systems to possible future risks.

  12. The Orthogonality Thesis

    Why high intelligence doesn’t guarantee good intentions: Nick Bostrom’s thesis that capability and goals vary independently, and what it means for alignment.

  13. The Capability vs. Control Problem

    Why AI capabilities are advancing faster than our ability to verify and steer them, and why control research must keep pace with capability scaling.

  14. Instrumental Convergence

    Why almost any capable AI, whatever its main task, might pursue the same sub-goals: self-preservation, goal integrity, cognitive enhancement and resources.

§ 4 — Common questions

The short answers.

What is superintelligence?

Superintelligence is a hypothetical intelligence substantially beyond the best human cognitive abilities across virtually all relevant domains. In AI discussions, it usually means a machine system. Read the full explanation.

What is the difference between AI and superintelligence?

AI is the broad category, covering systems with many different abilities. Superintelligence is a hypothetical, much higher level of capability within that category. Compare AI, AGI and SI.

Is ChatGPT a superintelligence?

No broadly accepted demonstration establishes that ChatGPT is superintelligent in the general sense used here. Producing useful answers or outperforming people on selected tasks is not evidence of superiority across virtually all cognitive domains.

Does superintelligence exist today?

There is no established, broadly accepted demonstration of artificial superintelligence. The concept remains hypothetical; impressive performance in a particular area is a different claim.

What is AGI?

Artificial general intelligence means broad, flexible intellectual capability across many domains, often understood as human-level. Definitions and the threshold for meeting them are disputed. Explore the AGI question.

Could AGI become superintelligent?

Possibly, if capabilities improved far beyond human level across a broad range of domains. AI-assisted research and recursive self-improvement are proposed routes, but neither sustained gains nor a rapid transition is guaranteed.

What is an intelligence explosion?

An intelligence explosion is a hypothetical feedback process in which better AI becomes better at producing further AI improvements, leading to rapid capability growth. Learn about the hypothesis and its possible limits.

Is superintelligence dangerous?

A hypothetical highly capable system could create serious risks through misuse or behaviour that conflicts with intended goals. The outcomes are uncertain, not inevitably harmful. Safety work asks how to evaluate and reduce those risks.

When could superintelligence arrive?

There is no reliable, agreed date. Predictions depend on contested assumptions about research progress, resources, capability thresholds and bottlenecks. Superintelligence should not be treated as inevitable.

What is AI alignment?

AI alignment is the problem of making a system's behaviour reliably reflect intended goals, values and constraints. It is about achieving the real intention, not just optimising an incomplete metric. See examples of the alignment problem.

§ 5 — The vocabulary

A glossary, not a guessing game.

Clear definitions of 78 terms, from agents and alignment to weights and zero-shot learning. Read alphabetically or search for the concept you need.

Open the searchable AI glossary →

Start with large language models, frontier models or recursive self-improvement.

Imprint / Legal notice

Michael Moritz, Neubaustraße 5, 6063 Rum, Austria · Moritz.m1@gmx.at
Purpose of this site: an educational field guide to AI, AGI and superintelligence.