AI Attributes™About AI Attributes™
AI Attributes is an open research effort on AI behaviour. It curates and develops open models that read attributes of text, such as emotions, moral framing and sensations, and it builds the products that put them to use.
The models are published with open weights on Hugging Face, and any platform can connect to them through an API. See Models.
Why it exists
Large questions about AI systems are hard to study whole. AI Attributes takes them a part at a time: one attribute, one product, one clear statement of what a result means. The aim is to make smaller questions easier to ask, without assuming an answer to the larger ones.
The products
Observers
An observer is one attribute of text, the open models that read it, and an API. It takes a piece of text and returns a score for each label, with the model and version that produced it. It never takes part in the conversation.
- OpenEmotions™: Open models for emotions in text: the 28 GoEmotions labels.
- OpenSensations™: Open models for the senses in text: vision, hearing, smell, taste, touch and the inner body.
- OpenMorals™: Open models for moral framing in text, following Moral Foundations Theory.
Environments
An environment gives a model a situation, tools and a deadline. What the model calls is executed and logged, so its decision is something it did, not something it said.
- OpenSimulations™: Simulated situations where a model acts: tools, a deadline, and every decision executed and logged.
Clients
A client runs the models and shows people the results. It attaches observers and environments with a URL and a key.
- AI Observability Studio: Run one prompt through many models and watch them decide, side by side.
- Am I Conscious?: An open experiment observing one persistent AI agent over time.
Each product works alone. They connect through written contracts, so a client attaches an observer or an environment with a URL and a key.
What this is not
AI Attributes does not claim to:
- create conscious AI or prove that a machine is sentient;
- measure what anyone, or any model, feels;
- judge right and wrong;
- provide clinical or psychological assessment;
- decide the moral status of any system.
A classifier's score is a reading of text, and a model's decision in a simulation is behaviour in a simulation. Neither is, by itself, evidence of subjective experience.
Who is behind it
AI Attributes is an independent project. It does not claim a university, laboratory, sponsor or expert affiliation.
Read the method and ethics, or the principles.