Chaser putting up paywall to thwart artificial intelligence training
Australian satirical publication The Chaser is putting its online content behind a paywall because of fears its archives are being used to train artificial intelligence models that will one day replace its writers.
Late last year, Chaser editor Cam Smith announced the comedy outlet would change its decade-long approach of posting free online content and begin restricting access to users who signed up with a free account.
Paywalled content is commonplace in digital media, an industry that has increasingly moved from an ad-supported model to subscription-based over the past decade. The Chaser’s decision, however, cited another reason for the decision: the emergence of publicly available large language model artificial intelligence products.
“What’s really put the nail in the coffin of The Free Chaser is a little thing called [OpenAI’s artificial intelligence product] GPT,” Smith wrote.
Get Crikey FREE to your inbox every weekday morning with the Crikey Worm.
With a new generation of artificial intelligence products sparking a new wave of fears that people’s jobs will be automated, Smith argued that satirists, too, are at risk of replacement: “Even now, [ChatGPT] is a more competent writer of satire than most of the people we’ve worked with. That includes myself, and I’ve been writing news satire for 16 years.”
Artificial intelligence models like GPT-3 ingest huge sets of data to create connections that are then used to generate content in response to a user’s prompt. GPT-3 was primarily trained on a dataset called Common Crawl that comprises more than 3 billion web pages scraped from the internet — which may already include The Chaser’s work (Crikey was unable to confirm whether their catalogue has been captured due to the dataset being temporarily unable to be searched).
Smith, who studied computing at university, told Crikey that he “absolutely” believes that artificial intelligence can produce good satire.
“Topical comedy is really just a numbers game, as in you throw 100 writers at the daily news and maybe five will come back with the same joke, because there are patterns and rules for writing comedy that they will all follow,” he said.
“Two things computers are very good at is repeating rules and generating results on an inhumanly large scale, so they’re a natural fit for this stuff.”
Artificial intelligence expert Dr Alan D Thompson agreed that deep learning artificial models like GPT-3 are capable of acts of creativity like satire.
“What they do is like when a person goes to an art gallery or sees thousands of paintings. When given a prompt, it conceptualises what something might sound or look like and create something like it,” he said.
Thompson shared his experiences with Leta — an artificial intelligence product that feeds a GPT-3 text model into a synthetic avatar to create a life-like but artificial person — where the program came up with jokes involving clever wordplay and expressed emotion. He said Leta even coined a word — “bushjoy” — to explain the excitement and pleasure that comes from being in the wild.
One of the obstacles to artificial intelligence language models is that the information consumed to create connections used to answer prompts is based on static datasets and not real-time information. ChatGPT does not know about the insurrection in Brazil on January 8, for example, but it does know about the January 6 US insurrection since its database was captured after the event.
The lack of up-to-date knowledge is a challenge for artificial intelligence-created topical humour. But Smith believes that it is only a matter of time until this changes. He even went as far as emailing Google to find out more about an internal language model that was rumoured to have “mastered” humour. (Google never got back to him.)
Thompson, on the other hand, said the computing power required to make the connections in the datasets is so enormous that real-time processing is still far off. He also takes a more optimistic view about what humour-producing artificial intelligence could do to satirists.
“Artificial intelligence can amplify and augment us. It’s not going to recreate a Chaser article word for word. But it is a new paradigm,” he said.
The Chaser has received 40,000 free sign-ups since the paywall announcement and the publication hopes that they’ll also be able to convince those users to pay in the future.
Smith said that the satire site has an even bigger audience out there but they’re hard to reach because of the vagaries of social media platforms and their algorithms — another challenge afflicting the digital media industry. To prove his point, Smith revealed at the end of the post announcing the change that “half” of the text had been written by artificial intelligence.
“If we’re being honest it was more like one-third was written by AI once you take into account the editing and tweaking we did, but that doesn’t sound as good as ‘half the article was written by AI’,” Smith confided.
This content was originally published here.