AI Governance Hub
Machine learning
Machine learning develops models from data. A model might learn patterns in labeled examples to classify a message or estimate a number. Testing on examples outside its training data helps assess whether it generalizes to new cases.
AI explained · Updated 15 September 2026
What is artificial intelligence (AI)?
Artificial intelligence, usually shortened to AI, describes systems that infer how to produce an output from inputs. That output might be a prediction, a recommendation, a decision or newly generated content. AI appears in many kinds of software, from image recognition to assistants that draft text.
Understanding the task matters more than the label. Ask what the system receives, what it produces and what a person needs to check before relying on it.
AI describes a family of approaches
The OECD definition focuses on a system's ability to infer outputs and recognizes different degrees of autonomy and adaptation. AI is not limited to chatbots, and a system does not need to keep learning after deployment to be considered AI. Read the OECD explanation of AI and non-AI systems.
Machine learning, generative AI and rules
Generative AI
Generative models produce content such as text, images, audio or video in response to inputs. A request to draft an email is different from asking a classifier to label an existing email. Both can involve AI.
Rules and automation
Some AI approaches use knowledge and reasoning rules. However, a fixed formula or a simple “if this, then that” workflow is not automatically AI. A tool can automate useful work without training a model or generating content.
The categories overlap: a generative system can use machine learning, and an application can combine a model with ordinary rules. See Google's introduction to machine learning for model types and examples, and the OECD discussion for knowledge-based approaches.
Illustrative workflows
Three examples: input, output and review
1. Drafting a support response
A language model receives a customer question and selected support articles, then proposes a reply. The reviewer checks the policy, quoted facts and whether the draft includes information that should not be shared. Giving the system permission to send the reply adds a separate operational decision. The agent approval planner helps plan that boundary.
2. Predicting demand
A forecasting model receives historical sales and relevant business inputs, then estimates future demand. A team compares forecasts with actual results and checks performance after a product or market change. A useful review records the data period, error measure, business assumptions and person who can adjust the plan.
3. Creating and editing a video
A generative video model can create footage from a prompt; an editor can trim and crop footage that already exists. Those are different tasks. Check the source rights, captions and final video before sharing. Our video-to-Shorts guide demonstrates an edit with supplied captions; it does not claim to generate footage or transcribe speech.
AI Learning · Sources checked 15 September 2026
AI for science and everyday life
Explore Google's work across health, weather and disaster resilience, learning, and economic opportunity. These external resources are published by Google and Google DeepMind. AIGH curates the links; access and features are managed by each provider.
Health: AlphaGenome Atlas
Google DeepMind's Atlas predicts the molecular effects of all 9 billion possible single-letter changes in the human genome. Researchers can explore its predictions through a website portal without writing code. Predictions support research and require scientific validation.
Weather resilience: WeatherNext 3
Google's global weather AI produces hourly forecasts and higher-resolution predictions. Its experimental research dataset requires an access request; historical and real-time data have different usage terms.
Economic opportunity: AI & Economy ATLAS
Google's open report studies how people use its AI products across work and everyday tasks. The first report analyzes 15 million deidentified interactions. It describes usage in Google's products, rather than measuring the entire AI economy.
Learning: AI across languages
Google reports that its language technologies reach languages spoken by 7 billion people. Explore its work on translation and inclusive speech technology, and check language and feature availability in the specific Google product you plan to use.
Read James Manyika's overview on Google's blog. These are educational links to third-party resources. AIGH does not provide these models or claim a Google partnership.
What should you check before using AI?
Use these practical questions when evaluating a tool or a proposed workflow. The answers should relate to the actual product and account settings you intend to use.
- Data: What information leaves your device, who receives it and how long is it kept?
- Reliability: How will you detect incorrect, incomplete or inappropriate output?
- Impact: Could an error affect someone's access, money, safety or opportunity?
- Control: Who reviews the output, limits permissions and stops the workflow?
- Evidence: What tests, source references and decisions will you retain?
Our AI governance guide turns those questions into an example review process. NIST's voluntary AI Risk Management Framework provides a broader basis for managing risks throughout AI development and use.
Choosing an AI hub: start with the task
An AI hub can be a learning resource, a model directory or a collection of practical tools. Before creating an account or buying access, choose one outcome and check the complete path to it: the input you provide, the work the tool performs, the review you need to make and the file or decision you receive.
Learn the basics
Use the examples above to distinguish generating content, making predictions and editing an existing file. Match the capability to your task before comparing products.
Review AI at work
Start with the AI governance workflow, then try a Free AI Review of a project export. Check the free score and summary before considering a paid report.
Create a video
For existing footage, use the online video editor or the free Shorts sample. Review the sample and export limits before choosing a Creator pass for your own footage.
A clear product page should explain what is free, what requires payment and what happens to your data. A demonstration can help you check the output; it does not establish that every workflow or device will behave the same way.
What can you try on AI Governance Hub?
Our site includes governance tools, educational puzzles and creator utilities. Their names do not mean that every feature uses an AI model. Choose by the result you need:
Assess an AI inventory: the governance assessment preview analyzes an export using the published rule-based methodology. It prioritizes review; it does not test your deployed model.
Practise reasoning: the IQ Challenge offers educational reasoning puzzles and a free starting summary. It is not a standardized IQ test or a measure of an AI model's intelligence.
Edit a short video: try a free video-to-Shorts demo. This browser editor uses visual changes and imported captions. Your own footage requires a Creator pass; the editor does not automatically understand the story.
Browse the complete tool directory for local checklists, evidence tools and available starting points. Each product page explains its inputs, output and access terms before you choose a workflow.