Welcome

Artificial intelligence is a field that's developing fast! In response, policy makers from around the world are creating guidelines, rules and regulations to keep up.

Finding accurate and up-to-date information about regulatory changes can be difficult but crucial to share best practices, ensure interoperability and promote adherence to local laws and regulations. That's why we've created AIR ️💬.

AIR ️💬 is a Q&A search engine designed to provide relevant and high quality information about AI policies from around the world. Using this tool, your AI policy questions will be answered, accompanied by relevant analyses by the OECD's AI Observatory!

How it works (and doesn't)

AIR ️💬 uses Large Language Models (AI algorithms that work with text) to pinpoint sections of policy documents that are relevant to your question. Rather than presenting you with the specific policy section verbatim, AIR ️💬 has been designed to summarize the information in a digestible format, so that the response you receive more naturally fits with the question you've posed.

It's helpful to keep in mind that AIR ️💬 is entirely restricted to our database (see “Available Sources” below). These sources are from the OECD.AI Database (containing national AI policies) and AI-related reports from the OECD iLibrary. If the answer to your question is not contained in these policy documents, the model won't be able to respond.

Since we restrict the model to information found in the documentation, it has a hard time with questions that require more generalized knowledge. Therefore, if you ask the model for information about AI policies in Asia, the model won't necessarily show you Japanese policy documentation. To overcome this limitation, it's best to be as specific as possible in your question, referencing the particular country you're looking for information on.

For more information about the tool's strengths and limitations, please see our website here.

Risks

We have done our best to make sure that the AI algorithms are only taking information from what is available in the OECD AI Observatory's Database; but, of course, Large Language Models (LLMs) are prone to fabrication. This means LLMs can make things up and present this made up information as if it were real, making it seem as if the information was found in a policy document. We therefore advise you to check the sources provided by the model to validate that the answer is in fact true. If you'd like to know exactly which documents the model can reference in its response, please see below.

Recommended usage

AIR ️💬 can only answer specific types of questions, for example:

• Questions about policy documents that are currently in the OECD AI Observatory's database
• Questions that are posed in English and target English language documents;
• Questions for which the answer can be found in the text (i.e. the thinking has already been done by the author) these AI models are not able to write their own research report combining information across policy documents and analyzing them itself).

If your question is outside the scope of the recommended use, the model has been instructed not to answer.

We are looking to create a tool that is as inclusive as possible. While currently the tool only works with English language questions and documents we will continue assessing AIR ️💬's capacity to perform as intended for users with different levels of fluency in English and plan to expand the functionality to ensure accessibility and impact across countries and user groups.