Image: © piter2121/Stock.adobe.com
ChatGPT has now reached 100m active monthly users. But OpenAI is worried a single, monolithic AI system will increase bias.
OpenAI said it is working on an upgrade that will allow users to customise ChatGPT to suit their own needs when using the viral AI tool.
ChatGPT, which Microsoft is investing $10bn in, has rapidly grown to become one of the most popular topics in the AI sector, drawing the attention of the public, tech giants and various industries.
First launched last November, the advanced chatbot quickly grew into one of the most popular pieces of software ever released. One study estimated that ChatGPT reached 100m monthly active users last month, Reuters reports.
However, OpenAI revealed in a blogpost yesterday (16 February) that it is concerned about the “biases in the design and impact of AI systems” such as ChatGPT.
“There will therefore always be some bounds on system behaviour. The challenge is defining what those bounds are,” OpenAI wrote.
“If we try to make all of these determinations on our own, or if we try to develop a single, monolithic AI system, we will be failing in the commitment we make in our charter to ‘avoid undue concentration of power’.”
To that end, the San Francisco-based start-up is now working on an upgrade to ChatGPT that will allow users to customise the AI’s behaviours and define its values “within broad bounds” and “up to limits defined by society”.
“This will mean allowing system outputs that other people (ourselves included) may strongly disagree with,” the company said.
“Striking the right balance here will be challenging–taking customisation to the extreme would risk enabling malicious uses of our technology and sycophantic AIs that mindlessly amplify people’s existing beliefs.”
Rebecca Wettemann, CEO of analyst firm Valoir, said that ChatGPT on its own “can’t identify right from wrong or truth from fiction” and that it only reflects the training it’s been given.
“The ready availability and usability of ChatGPT highlights the importance of understanding the bias in AI models and their training,” she said.
Last December – at around the time ChatGPT crossed 1m users – a Q&A site for programmers issued a temporary ban on answers created by the chatbot.
The site, Stack Overflow, claimed the number of correct answers created by ChatGPT is “too low” and that posting answers made by the AI could be “substantially harmful” to the site and its users.
10 things you need to know direct to your inbox every weekday. Sign up for the Daily Brief, Silicon Republic’s digest of essential sci-tech news.
