Skip to content
The Future Explorerby Benjamin Talin
Enablertechnological

AI & Machine Learning

Artificial intelligence and machine learning are computer systems that learn a task from examples instead of following rules a person wrote down. Show one millions of photos and it learns to recognise a cat. One chatbot found 100 million users in two months, yet only about one US firm in ten uses AI to make what it sells. It turns tasks we can only show, not explain, into tasks machines can do.

Maturity

3/ 5
measured
BasisCapability is among the most mature on this map. Use inside organisations is not: about one US firm in six uses AI anywhere in the business, and most firms that use it call it only somewhat important.SourceStanford HAI, AI Index 2024 (Apr 2024); US Census Bureau, BTOS (26 May 2026) and Annual Business Survey 2023 (17 Sep 2025); EIG (19 Dec 2025); provider pricing series (E-104)Vintage2024–2026
ReadinessCapability 5Deployment 3Absorption 3Bound bytrustTrendimproving

Where it feeds in

5 megatrends in 2 forces

Where you have already met this

ChatGPT found 100 million users in two months. By September 2025, about one US firm in ten used AI to make what it sells. The model was ready. The org chart was not.

If this moved

If AI capability doubled, no megatrend on this map would move. Automation of cognitive work, the application gap and precision everything are held at 2 by workforce trust and managerial capability. The next gain has to come from there.

What to watch for

Firm-level AI use in the Census Bureau's business survey, especially outside the information sector. Productivity statistics in the sectors that use AI most - where a real effect would have to show first.

Not to be confused with

Automation. Automation executes steps someone wrote down; machine learning infers the steps from examples. The two are sold together and fail differently.

What it is

Systems that improve at a task from data rather than from explicit instruction. The umbrella category under which most of the current wave sits.

How it works

  1. 01Data becomes numbersText, images or code are split into small units and turned into numbers a model can process.
  2. 02Training finds patternsThe model adjusts billions of internal parameters to predict the next word or the right label, over enormous volumes of data.
  3. 03Tuning shapes behaviourThe trained model is refined with human feedback and examples of good answers, so it follows instructions.
  4. 04Inference produces an answerA user's input runs through the finished model, which returns the most probable output. Every answer is a probability, not a lookup.
  5. 05Grounding and guardrails limit the damageSystems retrieve verified documents to anchor answers and filter what comes out. This is where reliability is bought, and why it costs.

State of play

  • AI surpassed human performance on several benchmarks, including image classification, visual reasoning and English understanding, but trailed on competition-level mathematics, visual commonsense reasoning and planning (Stanford HAI, AI Index 2024, Apr 2024).
  • The share of US firms using AI to produce goods or services rose from 4.6% at the start of 2024 to 10% in September 2025 (EIG analysis of the Census Bureau's Business Trends and Outlook Survey, 19 Dec 2025).
  • Asked about AI in any business function, between 17% and 20% of US firms reported use from December 2025 to May 2026 (US Census Bureau, 26 May 2026).
  • Of US businesses using AI, 58.0% called it only 'somewhat important' to their processes or methods (US Census Bureau, Annual Business Survey 2023, 17 Sep 2025).
  • Between 2022 and early 2025, unemployment rose 0.30 percentage points for the most AI-exposed fifth of US workers and 0.94 points for the least exposed fifth (EIG analysis of the Current Population Survey, 10 Aug 2025).

Why it matters

It converts tasks that could only be specified by example into tasks a machine can perform. That is a genuinely new capability class, and it is why this enabler steepens more megatrends than any other in the Compendium.

  • It feeds more megatrends than almost any other enabler, and it constrains none of them. Each one is held lower by something human or structural.
  • It is the book's clearest case of capability mistaken for arrival. A 5 in the lab and a 3 in the firm are both true. The megatrends it feeds sit at 2.
  • The job-loss story has not shown up in the data yet: unemployment rose less for the most exposed workers than for the least exposed.

What holds it back

Reliability at the tail

A system that is usually right still needs a person for the times it is not, and that person is the cost.

Organisational change

Using AI well means redesigning work, not dropping a tool into it. That is managerial capability, and it scores 2.

Perceived relevance

Most of the smallest firms say AI does not apply to them.

Terms

Model
The trained system itself: a very large set of numbers that turns an input into an output.
Training
The expensive phase in which a model learns patterns from data.
Inference
Running a trained model to get an answer. Every question is an inference.
Large language model (LLM)
A model trained on very large amounts of text to predict and generate language.
Retrieval-augmented generation (RAG)
Feeding a model relevant documents at question time so its answer rests on them.
Fine-tuning
Training an existing model further on specialised data for one kind of task.

Steepens

The gradient behaves like a sticky rubber band: it is pulled hardest toward the weakest enabler. Marginal links are shown but do not pull.

If this moved one step

Computed from the map as it stands: this score raised from 3 to 4, every other score unchanged. It is arithmetic, not a forecast.

Raised one step on its own, it lifts no megatrend's weakest link; each gradient it feeds still rises a little.

The gradient also rises in

What if it moved?

What would change the gradient

Conditional, not predictive. No dates, no probabilities.

  1. 01Inference cost falling far enough to make marginal use free
  2. 02Reliability at the tail, which is what blocks unsupervised deployment
  3. 03Nothing on the capability side - the binding constraints on its megatrends are human and structural

Common misconception

That capability equals deployment. Model capability is among the most mature enablers in this book; the megatrends it touches are held down by workforce trust, managerial capability and regulation, all of which score lower.

Talin, B. (2026). AI & Machine Learning. The Future Explorer, dataset v0.4.0 (2026-11-16). https://explorer.talin.digital/e/ai-ml/