OpenAI working towards reducing biases in ChatGPT

The MeITY is already looking to integrate some ChatGPT functionality into a WhatsApp bot service to share information about schemes, via voice. (Source: Reuters)

OpenAI, the research and development company behind the conversational AI platform ChatGPT has said that it is investing towards reducing 'glaring and subtle biases' in how its AI responds to different inputs.

There has been a worldwide scan on ensuring algorithmic accountability, with the Indian government on multiple occasions stating that the country will build a framework to prevent misuse of AI. More recently, the Department of Telecommunications released a draft standards paper on ensuring fairness of artificial intelligence.

On February 16, OpenAI said in a blog post, "We are investing in research and engineering to reduce both glaring and subtle biases in how ChatGPT responds to different inputs. 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."

OpenAI also said that they are developing an upgrade to ChatGPT that will allow users to easily customise its behaviour. "This will mean allowing system outputs that other people (ourselves included) may strongly disagree with," OpenAI said.

The company said that it will be necessary to ensure that there is proper balance in the AI, as risks of customisation include "enabling malicious use of our technology and sycophantic AIs that mindlessly amplify people's existing beliefs".

Also read: ChatGPT: Will the AI legal framework come of age in India?

OpenAI is also looking for public input to take decisions on ChatGPT's hard bounds.

"As a starting point, we’ve sought external input on our technology in the form of red teaming. We also recently began soliciting public input on AI in education (one particularly important context in which our technology is being deployed)," it added.

ChatGPT's algorithm is first 'pre-trained', wherein the model learns to predict the next word in a sentence. For doing that, the model is exposed to a vast array of text on the internet. Next, OpenAI "fine-tunes' its model to narrow down system behaviour.

For 'fine-tuning' the ChatGPT model, OpenAI uses a narrower dataset that is generated with the help of human reviewers who follow company guidelines. "The guidance we share with reviewers is more high-level -- (for example, “avoid taking a position on controversial topics”)," it said.