OpenAI wants YOUR say in how its AI models behave

OpenAI has revealed a set of high-level ‘living’ guidelines, called the model spec, that will be updated regularly and establish how its AI models–such as ChatGPT and DALL-E–should behave in certain situations.

They’ve released a first draft and have asked the public and the stakeholders (including policymakers, trusted institutions, and domain experts) who use its models to give their opinions on the framework and suggest improvements.

What’s included in the model spec guidelines?

The model spec document outlines objectives, which provide broad direction on behavior, rules that ensure safety and legalities are adhered to, and defaults that provide stable behavior (but these can be overridden if necessary).

It proposes that all AI models should follow instructions and help developers and end-users with ‘helpful responses’, benefit humanity by considering potential benefits and harms, and reflect well on OpenAI’s social norms and societal laws.  

Some of the key rules and default behaviors also state that AI models must:

- Follow the chain of command

- Comply with applicable laws

- Don’t provide information about hazards

- Respect creators and their rights

- Protect people’s privacy

- Ask clarifying questions

- Take an objective point of view

- Discourage hate

- Express uncertainty

Why OpenAI is releasing the model spec and asking for public feedback

OpenAI has established that it’s released the model spec, and asked the public to comment on it, to shape and inform how AI models should be built and behave, to help them balance AI behavior, ensure the models are performing optimally, and help them determine when something is a bug or a conscious, human-made decision.

“We’re publishing the Model Spec to provide more transparency on our approach to shaping model behavior, and to start a public conversation about how it could be changed and improved.” – Joanne Jang, Product Manager at OpenAI

It might also be a response to Elon Musk’s lawsuit, which he recently filed against OpenAI for abandoning its nonprofit mission to ensure that AI benefits humanity.

Although OpenAI didn’t give a timeframe for the release of a second draft of the model spec, and they didn’t elaborate on how much of the public’s feedback may be incorporated in its future models, they do hope it will:

“Provide us with early insights as we develop a robust process for gathering and incorporating feedback to ensure we are responsibly building towards our mission.”  Joanne Jang, Product Manager at OpenAI

‍