ChatGPT has been a high point of discussion on both sides of the camp. On the one hand, it can help companies generate written content and grow their businesses. On the other, it threatens the integrity of search engine results and the jobs of human writers. Fortunately, AI Safety is here to save the day with a new watermark technique. The problem is that it’s not flawless.
Meet Scott Aaronson, a computer scientist that OpenAI hired to work on AI Safe and Alignment. The goal of this division is to deal with the fears that so many of us have about AI. While it’s meant to benefit humans, it should not threaten us in any way. That means not replacing human writers or beating our SEO content for search engine ranking.
In other words, search engines should still show human content at the top and not simply AI works. But what if the AI content is actually better and more informative? I’ll get to my own thoughts on that in a moment.
Back to Scott, he’s found a way to include a watermark in ChatGPT content that marks it as produced by AI. It’s not a logo or image you’ll be able to see. Instead, the watermark will consist of random patterns, code, or words that look natural, but actually encrypt it as a mark for AI content. In that way, we can more easily detect AI with detection and decryption tools.
I suspect that the watermark will also help Google and other search engines detect AI content to prioritize human work; which brings me back to my thought. If I was involved in work with Google or search engine infrastructure, I would have made it so that we can see both human and AI content in different panels.
So, if we searched for a topic, there could be an identifier showing if a human or AI wrote it. Even better, you can switch between tabs to see human and AI content separately. In this way, we can still support human writing without completely disregarding AI.
Scott says that they’re still testing the ChatGPT watermark, but there are some issues. For one, there are ways the encryption can be overcome or removed. That means that, in its current state, we might not be able to tell if humans or AI wrote the articles. We’ll have to wait and see how things progress.
