# Explainers - AI rules and terms

> Plain-language explanations of the rules, standards and technical terms behind AI in organisations, each linking to its official source.

- Web page: https://www.alesanalytics.com/explainers
- Markdown: https://www.alesanalytics.com/explainers.md
- Publisher: Ales Analytics

## New to this? Start with these five

1. [Private deployment](https://www.alesanalytics.com/explainers/private-deployment.md)
2. [Model gateways](https://www.alesanalytics.com/explainers/model-gateway.md)
3. [Grounded answers](https://www.alesanalytics.com/explainers/grounded-answers.md)
4. [Baseline](https://www.alesanalytics.com/explainers/baseline.md)
5. [The EU AI Act](https://www.alesanalytics.com/explainers/eu-ai-act.md)

## Rules and standards

- [The EU AI Act](https://www.alesanalytics.com/explainers/eu-ai-act.md): The EU's AI Act in plain language: what it bans, what it calls high-risk, what it asks of general-purpose models, and the deadlines as amended in 2026.
- [ISO/IEC 42001](https://www.alesanalytics.com/explainers/iso-42001.md): What ISO/IEC 42001 is, what an AI management system covers, what certification does and does not mean, and how it relates to ISO 27001 and ISO 23894.
- [The NIST AI Risk Management Framework](https://www.alesanalytics.com/explainers/nist-ai-rmf.md): What the NIST AI RMF is, how its four functions work, what the generative-AI profile adds, and why organisations outside the US use it.
- [Data protection law and AI](https://www.alesanalytics.com/explainers/data-protection-and-ai.md): How data protection law applies to AI wherever you operate: lawful basis, minimisation, automated decisions, deletion, and residency by architecture.
- [Model risk management](https://www.alesanalytics.com/explainers/model-risk-management.md): What model risk management means, where it comes from, what an inventory and independent validation are, and why the same structure suits AI systems.
- [AI rules around the world](https://www.alesanalytics.com/explainers/ai-rules-around-the-world.md): US state laws, the UK approach, Singapore's framework, Japan's law, the OECD principles and EU product rules that touch AI, each linked to the original.

## Ideas and terms

- [Baseline](https://www.alesanalytics.com/explainers/baseline.md): What a baseline is, why value cannot be shown without one, what to record, and how to capture one when the AI is already partly in use.
- [Control group](https://www.alesanalytics.com/explainers/control-group.md): How a control group works, when a randomised comparison is possible, what to do when it is not, and why a small permanent comparison is worth keeping.
- [The AI profit and loss view](https://www.alesanalytics.com/explainers/ai-profit-and-loss.md): A one-page way to report what an AI system earned, saved and helped avoid, net of everything it cost, in the terms a finance team already uses.
- [Total cost of ownership](https://www.alesanalytics.com/explainers/total-cost-of-ownership.md): The cost layers around an AI system: model, data work, integration, checking, governance, change and infrastructure, and why the licence is a small part.
- [AI inventory](https://www.alesanalytics.com/explainers/ai-inventory.md): What an AI inventory records, why every framework asks for one first, how to find AI hidden in software you already use, and how to keep it current.
- [Language models and open weights](https://www.alesanalytics.com/explainers/language-models-and-open-weights.md): What a language model is, the difference between renting one as a service and running one you can download, and why open weights matter for control.
- [Reasoning models](https://www.alesanalytics.com/explainers/reasoning-models.md): What a reasoning model does differently, how it is trained, what the extra thinking buys and costs, and why its written working is not proof.
- [Multimodal AI](https://www.alesanalytics.com/explainers/multimodal-ai.md): How a model reads scanned documents, charts and recordings directly, what it gains over converting them to text first, and where it still sees things that are not there.
- [Private deployment](https://www.alesanalytics.com/explainers/private-deployment.md): What running AI on your own servers or cloud account means, what stays inside, what disconnected means, and what it gives you under data protection law.
- [Model gateways](https://www.alesanalytics.com/explainers/model-gateway.md): What a model gateway is, why routing, policy and cost control end up in one place, and how it makes changing model a setting rather than a project.
- [Owning or renting the hardware](https://www.alesanalytics.com/explainers/own-or-rent-infrastructure.md): What decides between buying accelerators and renting capacity: how steady the load is, residency, what you already own, and costs the hardware quote omits.
- [Tokens and inference](https://www.alesanalytics.com/explainers/tokens-and-inference.md): What a token is, what inference means, why longer documents and more people at once change the hardware you need, and how that feeds into sizing.
- [Grounded answers](https://www.alesanalytics.com/explainers/grounded-answers.md): How an assistant finds the right passage in your documents before answering, why the citation matters, and what makes a grounded answer trustworthy.
- [AI agents](https://www.alesanalytics.com/explainers/ai-agents.md): The difference between an assistant and an agent, what an agent is allowed to touch, and why the boundary and the audit trail are the product.
- [Human sign-off](https://www.alesanalytics.com/explainers/human-in-the-loop.md): What human oversight means in practice for AI systems: the checkpoint, who approves, what they see, and how it is recorded.
- [Access and identities for AI](https://www.alesanalytics.com/explainers/machine-identities.md): Why an AI assistant or agent needs its own identity, how its permissions are set and checked, and why most AI security work is access control.
- [Tuning and routing](https://www.alesanalytics.com/explainers/fine-tuning-and-routing.md): Tuning a model on your own examples, what routing does, and why sending each question to the right model makes systems faster, cheaper and easier to audit.
- [Evaluation sets](https://www.alesanalytics.com/explainers/evaluation-sets.md): How a test set of real questions with reviewed answers is built, why your own people write it, and how it turns quality into a number tracked over time.
- [System of record](https://www.alesanalytics.com/explainers/system-of-record.md): What a system of record is, why the same fact often lives in several systems, and how naming one authority per fact is what makes an AI answer trustworthy.

## Contact

solutions@alesanalytics.com, or https://www.alesanalytics.com/contact.

- Every page as Markdown: https://www.alesanalytics.com/llms.txt