OpenAI published a blog, “How should AI systems behave, and who should decide?”, where they explain how ChatGPT works and that they are going to allow more user customisation, and more input from the public in the decision-making process.
This comes after the Microsoft-backed startup’s chatbot was claimed to be politically biassed, and has an left-libertarian inclination. OpenAI said that the users are rightly worried about the biases in the design and the impact that it creates, and therefore want to work to mitigate political and other biases within ChatGPT, while also accommodating diverse views.
Outlining the role of the reviewers, OpenAI said that for fine-tuning the model to be as less biassed as possible, reviewers will have an on-going relationship. Citing their guidelines, the company stated that to address the biases, reviewers should not favour any political group, and the biases that emerge even then, will be considered bugs.
Explaining the building blocks for their future systems, OpenAI wants to improve the default behaviour of ChatGPT. “In some cases ChatGPT currently refuses outputs that it shouldn’t, and in some cases, it doesn’t refuse when it should. We believe that improvement in both respects is possible.”
The company said that though there would be some bounds on the behaviour of the system, they will allow it to generate content that many people would strongly disagree with, including OpenAI. The company expresses that striking the right balance here would be challenging, and many decisions on the defaults and hard bounds should be made collectively.
You can apply for OpenAI’s Researcher Access Program here.
Microsoft also recently said that user feedback was helping the company improve their search engine, Bing, before a wider rollout. They were able to correct that their AI chatbot was open to being “provoked” by users to generate responses it did not intend.