Levels of AI

Two Classifications of AI Levels

The term levels of AI has two distinct meanings:

1. Levels of AI intellectual capability — from narrow AI to hypothetical superintelligence (ANI → AGI → ASI)
2. Levels of organizational AI maturity — how deeply a company leverages AI (AI Maturity Model)

Classification #1: AI Intellectual Capability

ANI — Artificial Narrow Intelligence

Status: Existed since the 1950s-60s (early ML), exploded since 2010 (deep learning).

Characteristics:

• Specialized in one task or narrow domain
• Can exceed humans in its specialization (e.g., AlphaGo in Go)
• Cannot learn beyond its domain without retraining
• No true “understanding” — statistical patterns

Examples in 2026:

• LLMs: GPT-5, Claude 4, Gemini 2.5 — advanced language understanding, but still ANI
• Computer Vision: facial recognition, medical diagnosis (e.g., skin cancer)
• Recommendations: Netflix, Spotify, Amazon
• Autonomous vehicles: Tesla Autopilot, Waymo
• Games: AlphaZero (chess, Go, shogi), DeepMind in StarCraft II
• Scientific: AlphaFold (molecular biology), GNoME (materials)

ANI limitations:

• Lack of common-sense reasoning
• Cannot transfer knowledge between domains
• Hallucinations in LLMs (generating false information)
• Requires huge datasets and compute

AGI — Artificial General Intelligence

Status: Hypothetical, does not exist in 2026.

Characteristics:

• Capable of any intellectual task at human level
• Learns new domains without retraining — knowledge transfer
• Possesses common sense, understands context
• Makes decisions in situations never encountered before

2026 debate:

• Optimists (Sam Altman, Ray Kurzweil, Demis Hassabis): AGI by 2030-2040, scaling LLMs + new architectures
• Skeptics (Yann LeCun, Gary Marcus, Emily Bender): AGI requires fundamentally new approaches (world models, symbolic reasoning), not just scaling

AI expert forecasts (AI Impacts Survey 2023):

• Median: 2047 (50% probability)
• 10% chance: by 2030
• 90% chance: by 2080

ASI — Artificial Superintelligence

Status: Hypothetical, likely after AGI.

Characteristics:

• Exceeds human intelligence across all domains — science, creativity, practical wisdom, social skills
• Self-improvement (recursive self-improvement) — AI improves itself
• Potential “singularity event” (Kurzweil)

Ethical debate:

• Alignment problem — how to ensure ASI acts in line with human values?
• Existential risk — debate on risk to humanity (Bostrom, Yudkowsky, Hinton)
• AI safety research — institutes like Anthropic, DeepMind Safety, MIRI

Classification #2: Organizational AI Maturity Levels

The AI Maturity Model is a tool for assessing how mature a company’s use of AI is. Popular models: Gartner, McKinsey AI Maturity, Microsoft AI Maturity Framework.

Level 1: Ad-hoc / Aware

• Isolated, uncoordinated experiments
• No AI strategy
• AI knowledge scattered, mostly in IT
• ROI unmeasurable

Typical signals: “We tried ChatGPT in one sales team, but nothing came of it.”

Level 2: Experimenting / Active

• Pilot projects (PoCs) in 2-3 areas
• Sandboxes and playgrounds
• Beginning to hire data scientists / ML engineers
• Collaboration with external AI vendors

Typical signals: “We have 3 AI pilots — customer service chatbot, review sentiment analysis, invoice automation.”

Level 3: Operationalizing / Operational

• First models in production with monitoring
• Beginnings of MLOps (versioning, deployment, monitoring)
• AI strategy at department level
• Defined roles: ML Engineer, Data Engineer, MLOps Engineer
• AI-related KPIs (e.g., accuracy, response time, ROI)

Typical signals: “Our churn prediction model has been in production for 6 months and reduced churn by 15%.”

Level 4: Transforming / Systemic

• AI embedded in core business processes
• AI strategy at C-Suite level
• Center of Excellence or dedicated AI team
• Robust MLOps + AI governance
• Culture of experimentation, A/B testing

Typical signals: “Each of our products has an AI component. We have an AI Council reporting to the CEO.”

Level 5: Optimizing / Transformational / AI-First

• AI is the core of strategy and culture
• Continuous learning, rapid iteration
• AI partner ecosystem
• Proprietary models, data moat, custom LLMs
• AI Ethics Board, AI Explainability, AI Act compliance

Typical signals: This is how OpenAI, Anthropic, Google/DeepMind, Tesla, DeepL, Shopify (2026) operate.

How to Assess Your Company’s AI Maturity Level

Sample assessment questions:

Strategy (10%):

• Do we have a documented AI strategy with budget and KPIs?
• Does the board understand AI and participate in decisions?

Data (20%):

• Do we have a data lakehouse / warehouse?
• Is data quality measured?
• Compliance (GDPR, AI Act)?

Talent (15%):

• How many ML Engineers / Data Scientists do we have?
• AI training program for other employees?

Technology (20%):

• MLOps pipeline (CI/CD for models)?
• Model monitoring in production?
• Experimentation platform?

Processes (15%):

• Model governance (review, approval, audit trail)?
• AI Ethics guidelines?
• Incident response for AI failures?

Culture (10%):

• Does the entire organization use AI in daily work?
• Experimentation mindset?

Results (10%):

• Concrete case study with measurable ROI?
• AI products/features generating revenue?

Key AI Autonomy Levels (SAE-like)

For operational systems (vehicles, chatbots, agents):

Level	Name	Description	2026 Example
0	No automation	Human does everything	Manual process
1	Assistant	AI suggests, human decides	GitHub Copilot, Grammarly
2	Partial automation	AI executes in narrow scope	Tesla Autopilot, OpenAI Deep Research
3	Conditional automation	AI operates autonomously under conditions	Agentic AI (AutoGPT, Cursor Composer)
4	High automation	AI independent in most situations	Waymo (robotaxi), does not exist for chatbots
5	Full automation	AI independent in all situations	Does not exist in 2026

See Also

• Artificial Intelligence — definition and applications
• Machine Learning — what it is
• AI & Machine Learning Training Hub
• Generative AI and LLM for Business
• AI Maturity Assessment Framework for Enterprises