China’s version of ChatGPT could mimic ‘memory hole’ from George Orwell’s 1984. The age of Party-approved propaganda spread through AI might be just around the corner.
In the end we shall make thoughtcrime literally impossible, because there will be no words in which to express it,” said Syme to Winston in George Orwell’s Nineteen Eighty-Four. In the novel, Winston Smith is in-charge of adjusting the past to the new narratives offered by the fictitious Party that rules over the imaginary state of Oceania. The truth is tightly controlled in Oceania.
Unlike Orwell’s Winston, who works for the Ministry of Truth, OpenAI’s ChatGPT trained on machine-learning dataset is likely to create trouble for the Chinese Communist Party (CCP) rather than solve the problem of ever-shifting narratives.
The ChatGPT chatbot, created by Microsoft-backed artificial intelligence company OpenAI, has created a buzz – and panic – around the potential of AI to shake up industries that rely on creating text-based content. In 2019, Microsoft invested $1 billion in OpenAI, which put the company ahead of its Silicon Valley rival, Google.
Just like the rest of the world, China is interested in capturing the potential of ChatGPT which can create uncensored content. The ChatGPT programme isn’t available in China because a foreign phone number is required to access the programme, but some software developers have found a way around it.
Beijing’s initial response to ChatGPT was to ban access to the service, which WeChat’s mini programme facilitated. That doesn’t mean the government and the companies have decided against developing the Chinese version of ChatGPT completely.
China’s internet giant Baidu has struggled with past iterations of a chatbot called PLATO-XL, which couldn’t answer basic questions such as when is Jack Ma’s birthday.
Baidu described PLATO as the world’s first ‘dialogue generation model’ trained on ‘11 billion parameters’, which gave the programme a very limited set of capabilities responding to queries in single sentences.
After PLATO’s failure, Baidu started another project called Ernie Enhanced Representation Through Knowledge Integration (Chinese name Wenxin Yiyan) in 2019, a ChatGPT style bot that will likely be launched in March. Ernie is being trained on 260 billion parameters as compared to GPT3 trained on 175 billion parameters, though the former was trained on a smaller dataset.
Baidu has started collaborating with technology companies to develop Ernie for real-world use cases. Taiwan’s telecom company FarEasTone announced a partnership with Baidu to access Ernie through Baidu’s Smart Cloud.
And yet, creating a Chinese version of ChatGPT poses new challenges for the censorship regime in China.