There is no fundamental reason why AI censorship won’t be able to understand the symbol of a piece of blank paper, and time is running out to save us from tyranny.
This past weekend, an unprecedented protest broke out in many cities in China. The people wanted to end their government’s notorious zero-COVID policy, the most draconian in the world, causing many non-COVID related deaths. Earlier this year in September, 27 died in a shuttle bus transferring to quarantine camps. Most recently on November 24, 10 died from a fire in a building that was physically locked, while the firefighters were stuck outside the gate.
Triggered by the unimaginable nightmare of burning to death behind locked doors, the Chinese went to the street and protested against the injustice. They are tired of the unscientific and opaque decision-making process for COVID lockdowns, and more importantly, the decaying freedom of speech. Having had deadly experiences from the cultural revolution and the 1989 movement, the Chinese often shun themselves away from politics. But today, if you live in China, you can talk about neither Xi nor the COVID policy, and according to the latest policy, not even hit the wrong like-button.
As people bypass censorship by inventing replacement words and innuendos in their writings, the state is also using an increasing level of artificial intelligence (AI) to detect politically sensitive information. In the latest events, the protesters started using a piece of blank paper to symbolize their censored voice. But is blankness AI-proof?
Prof. Eric Xing, a well-known AI researcher, President of the Mohamed bin Zayed AI University in UAE and Professor of Computer Science at CMU, asked in his tweet whether the latest AI technology can understand the meaning of blankness.
As an AI researcher myself, I would have had a more conservative prediction two years ago. But now, I have to say there is no fundamental reason why AI won’t be able to understand these symbols—perhaps not right now—yet it will simply be a matter of time for the Chinese government and their Wechat / Weibo affiliates to ship large language models in their systems, if they have not already done so.
Large language models are state-of-the-art language understanding machines that came out of AI research in the past few years. Thanks to an exponential growth of model scale, the performance of these models has increased dramatically. Now large language models have been widely deployed in big techs for chatbots, question answering, and even writing mathematical proofs. They are now used to generate novel compositions that humans have never seen.
Many people not in the AI field may have an outdated view of what AI can do. ReCaptcha, as an example, is still widely used on websites to distinguish humans from machines. But the truth is, the AI algorithm behind ReCaptcha can solve the puzzles before you even start typing the first character or clicking on the first motorcycle. That was the state-of-the-art from almost 10 years ago. You would be wrong if you think your screenshot today can bypass the censor by having distracting doodles on it.
Some people compared this ongoing protest with the 1989 movement. But the suppression of freedom of the press was nothing close. 33 years ago, many local newspapers still reported the evolution of protests. Today, the Chinese government is equipped with the strongest censorship machine, from newspapers to social media and all the way to private chats. No videos of protest scenes are allowed to be transmitted anywhere. China even managed to get Apple to modify their AirDrop functionality for iPhones sold in China so that people cannot disseminate anonymous messages through AirDrop.
No way of communication will be safe. AI will likely recognize the innuendos you wrote, the voice messages you sent, the screenshots you took, and the videos you shot. And equally likely, AI will also recognize the things you didn’t write or didn’t say. AI might even dig up conspiratorial meanings out of normal sentences.
Chinese people are so good at tolerating tyranny. They have repeated their history for over 2200 years since the Qin dynasty, waiting for dynasties to naturally come to an end, one after another. But everything will soon be different—AI will probably blow away all peaceful protests in their infancy; AI will probably make the most unbreakable dictatorship in human history.
Time is running out for the Chinese people, and all of us.
That’s very true, AI is being wielded as a tool for control and oppression, and we may be closer to 1984 than we think.