The arrival of AI chatbots marks a historical dividing line after which online material can鈥檛 be completely trusted to be human-created, but how will people look back on this change? While some are urgently working to archive 鈥渦ncontaminated鈥� data from the pre-AI era, others say it is the AI outputs themselves that we need to record, so future historians can study how chatbots have evolved.
, an entrepreneur and former chief technology officer at both The New York Times and The Wall Street Journal, says he sees AI as a risk to information such as news stories that form part of the historical record. 鈥淚鈥檝e been thinking about this 鈥榙igital archaeology鈥� problem since ChatGPT launched, and it鈥檚 becoming more urgent every month,鈥� says Pant. 鈥淩ight now, there鈥檚 no reliable way to distinguish human-authored content from AI-generated material at scale. This isn鈥檛 just an academic problem, it鈥檚 affecting everything from journalism to legal discovery to scientific research.鈥�
For at cybersecurity firm Cloudflare, information produced before the end of 2022, when ChatGPT launched, is akin to low-background steel. This metal, smelted before the Trinity nuclear bomb test on 16 July 1945, is prized for use in delicate scientific and medical instruments because it doesn鈥檛 contain faint radioactive contamination from the atomic weapon era that creates noise in readings.
Advertisement
Graham-Cumming has created a website called to archive sources of data that haven鈥檛 been contaminated by AI, such as a full download of Wikipedia from August 2022. Studies have already shown that Wikipedia today shows signs of huge AI input.
鈥淭here鈥檚 a point at which we we did everything ourselves, and then at some point we started to get augmented significantly by these chat systems,鈥� he says. 鈥淪o the idea was to say 鈥� you can see it as contamination, or you can see it as a sort of a vault 鈥� you know, humans, we got to here. And then after this point, we got extra help.鈥�
runs the Wayback Machine at the Internet Archive, a project that has been archiving the public internet since 1996, says he is sceptical about the efficacy of any new efforts to archive data, given the Internet Archive stores up to 160 terabytes of new information every day.
Rather then preserving the pre-AI internet, Graham wants to start creating archives of AI output for future researchers and historians. He has a plan to start asking 1000 topical questions a day of chatbots and storing their responses. And because it is such a massive task, he will even be using AI to do it: AI recording the changing output of AI, for the curiosity of future humans.
鈥淵ou ask it a specific question and then you get an answer,鈥� says Graham. 鈥淎nd then tomorrow you ask it the same question and you鈥檙e probably going to get a slightly different answer.鈥�
Graham-Cumming is quick to point out that he isn鈥檛 anti-AI, and that preserving human-created information can actually benefit AI models. That is because low-quality AI output that gets fed back into training new models can have a detrimental effect, leading to what it is known as 鈥�model collapse鈥�. Avoiding this is a worthwhile endeavour, he says.
鈥淎t some point, one of these AIs is going to think of something we humans haven鈥檛 thought of. It鈥檚 going to prove a mathematical theorem, it鈥檚 going to do something significantly new. And I鈥檓 not sure I鈥檇 call that contamination,鈥� says Graham-Cumming.
