In the course of the past week, I am willing to bet that you have been asked by a machine to show that you are human. One website or two is bound to have ordered you to “prove that you are not a robot”, making you decipher letters and numbers, or choose all the pictures in a grid that feature traffic lights. Some tests can be frustrating when you simply cannot make out the squiggly text on the screen, or fail to correctly identify all traffic lights. But even as the blood of your humanity boils with rage, you must persevere with the test or the machine will not accept that you are human.

The irony of having to demonstrate your humanity to a machine is nothing short of ludicrous. And yet, these tests to verify our humanness serve a critical purpose. They are designed expressly to authenticate that we are indeed humans, as opposed to “bots”. Bad actors with nefarious intent deploy bots – essentially automated programs – to scrape data, enter information into websites, or click on pay-per-click advertisements to generate revenue.

To avert such unwelcome exploitation, computer researchers in the 1990s devised these Captcha (Completely Automated Public Turing test to tell Computers and Humans Apart) tests to block bots from entry. A host of other innovations including reCaptcha have since been developed to identify human users.

And therein lies the rub. While the bar for us to prove that we are humans is high, the threshold we set for artificial intelligence (AI) to reveal its presence and involvement is virtually non-existent. Consider ChatGPT, the latest AI wunderkind that people can’t stop talking about. So compellingly human and breathtakingly persuasive is the output it generates that many have been fooled into thinking they have read the work of a talented scribe. Trained on massive amounts of text authored by humans to begin with, ChatGPT successfully mimics human writers with minimal telltale signs as to its artificiality.

Greater transparency is surely vital if such AI-generated text, a significant proportion riddled with inaccuracies or falsehoods, is poised to flood our information ecosystem. Tech news site CNET was found to have posted a series of articles authored by an AI bot that spouted patently erroneous information. One explainer on compounding interest contained at least five significant inaccuracies. Gallingly, CNET had initially hidden the fact that such articles were authored by AI as they bore the nondescript byline “CNET Money Staff ”.

Northwestern University used ChatGPT to fabricate 50 medical research-paper abstracts and tested them on a group of medical researchers. They incorrectly identified 32 per cent of the faux abstracts as being real, indicating that even experts can be misled. These troubling developments signal an urgent need for greater transparency in AI. In the face of deceptively persuasive text and disturbingly realistic images and videos – all AI-generated – the exhortation that consumers must sharpen their digital literacy and critical discernment rings hollow.

Instead, innovation and regulation around the labelling and watermarking of AI-generated content should be advanced so that AI is forced to reveal its hand. OpenAI, the company behind ChatGPT, has actually developed a watermark to indicate the authorship of text it generates by embedding a statistical pattern into the text generator’s choice of words and punctuation marks. Although purportedly not yet in use, such watermarks will aid in the detection of AI-generated text and help combat misinformation and plagiarism.

Such efforts bode well for industry self-regulation and a greater push for responsible AI, but the efforts must be rigorously tested and implemented at scale. On the regulatory front, the Cyberspace Administration of China passed legislation on AI-generated content that took effect on Jan 10. Users are prohibited from using generative AI for activities that undermine public interest or threaten national security, while service providers must audit AI-generated content manually or through technical methods. The authority has stressed the importance of watermarks to identify AI content, and ruled that watermarks must not be deleted, manipulated or concealed. 

One may argue that such heavy-handed regulation is possible only because of China’s real-name verification system for Internet and AI-software users. But with many Chinese companies at the forefront of AI innovation, the methods they develop to manage the profusion and diffusion of AI-generated content may prove instructive. These methods will include techniques for labelling and tagging AI-generated content of all genres, constructing a feature database for identifying illegal and fallacious information, and stringent requirements for AI-generated content service providers to periodically review their algorithms and conduct safety assessments of content-generation tools.

Just as humans are subjected to Captcha tests every day to certify our humanity to machines, so too must we demand that bots and other AI tools reveal their presence to us humans. It is perhaps the greatest contradiction of our digital age that although humans forged these digital tools and infrastructures, we have unwittingly subjugated ourselves to the bots that we made.

The most alarming doomsday scenario in science fiction lore is that humans will one day kowtow to robots. Spoiler alert – that day has long arrived and we humans have yielded perhaps all too willingly.