Cosmos
Cosmos is a quarterly science magazine. We aim to inspire curiosity in ‘The Science of Everything’ and make the world of science accessible to everyone.
By Cosmos
Are chat bots hoarding state secrets? Is Chat-GPT collecting confidential information?
Mark Pesce, a futurist, tech expert and columnist for Cosmos, has announced a new book called Getting Started with ChatGPT and AI Chatbots helping people understand the “rules of the road” for chatGPT and other chat bots.
“There has never been a technology in the history of technology that has been deployed this quickly,” he told Cosmos.
“I thought the best thing that I could do right now is give people some rules of the road.”
The issue, according to Pesce, is that people are using these technologies with their guard down, leading to confidential information, or even potentially state secrets being given to AI services.
He highlighted a Crikey story in August which showed that Department of Defence staff accessed ChatGPT servers thousands of times before the web domain was eventually restricted.
“Organisations need to develop policies around how and when and why chat bots can be used, then they need to develop procedures and then they need protocols for what happens when it goes wrong,” says Pesce.
“Organisations don’t have any of these right now.”
The second problem Pesce goes into in the book is that chat bots are just not very good at telling the truth.
“The other reason I wrote the book is because in their essential nature, these things cannot tell confabulations from the truth,” Pesce says.
“How can you tell whether an AI chat bot is returning a response that’s true or not, if you’re outside your own area of expertise?”
While Pesce sees both issues as big problems already, companies like Microsoft and Google are currently trying to implement chat bots into as many of their programs as possible, so this will only become a much larger issue into the not so distant future.
“These are very powerful, wonderful tools, but think twice before you share data with them,” says Pesce.
“If it’s personal, private, commercial in confidence – if it’s something you wouldn’t want shouted from a rooftop – probably don’t hit enter.”
You can preorder Mark Pesce’s book Getting Started with ChatGPT and AI Chatbots: An introduction to generative AI tools here.
Do electric car batteries explode? Should I care about my BMI? Get the facts you really want to know about. Listen now.