Italy bans ChatGPT (over privacy concerns)

As the first major action by a nation to limit the spread and use of generative AI, Italy's government has taken the step to formally ban ChatGPT use not only by government employees, but by all Italians.

As reported by the BBC, "the Italian data-protection authority said there were privacy concerns relating to the model, which was created by US start-up OpenAI and is backed by Microsoft. The regulator said it would ban and investigate OpenAI 'with immediate effect'."

While I believe this concern is rooted in a misunderstanding as to how ChatGPT operates - it is a pre-trained AI, that doesn't integrate or learn from the prompts and content entered into it - given that OpenAI does broadly review this injected data for improving the AI's responses means there is enough of a concern for a regulator to want to explore it further.

Certainly I would not advise entering content that is private, confidential or classified into ChatGPT, but except in very specific cases, there's little to no privacy risk of your data being reused or somehow repurposed in nefarious ways.

In contrast the Singaporean government has built a tool using ChatGPT's API to give 90,000 public servants a 'Pair' in Microsoft Word and other applications they can use to accelerate writing tasks. The government has a formal agreement with OpenAI over not using any data prompts in future AI training.

What Italy's decision does herald is that nations should begin considering where their line is for AIs. While most of the current generation of large language models are pre-trained, meaning prompts from humans don't become part of their knowledge base, the next generation may include more capability for continuous finetuning, where information can continually be ingested by these AIs to keep improving their performance.

Specific finetuning is available now for certain AIs, such as OpenAI's GPT3 and AI21's Jurassic, which allows an organisation to finetune the AI to 'weight it' towards delivering better results for their knowledge set or specific goals. 

In government terms, this could mean training an AI on all of Australia's legislation to make it better able to review and write new laws, or on all the public/owned research ona given topic to support policy development processes.

It makes sense for governments to proactively understand the current and projected trajectory of AI (particularly generative AI) and set some policy lines to guide the response if they occur.

This would help industry develop within a safe envelope rather than exploring avenues which governments believe would create problems for society.