Self‑Declaration
The first and most common is self‑declaration. If the author is known to you and trusted, this is, in principle, fine. However, not every author recognises the extent to which they have used AI and they might honestly declare that their book or article is human authored whereas in practice there is an excessive contribution of AI. It is so easy for this to happen. For example, the author might have upgraded from rule‑based grammar checking in Microsoft Word to AI‑based grammar checking in a product such as Grammarly. Initially the author might only use the minimum level of suggestions from Grammarly, but it will offer to help and without thinking about it the author will start accepting Grammarly’s seamless offer to rewrite sentences and its friendly offer to generate paragraph‑scale text.
Ambushing
The second is what I call ambushing. That is when other people submit the author’s publication to third‑party AI‑detection tools such as Pangram, GPTZero, Humalingo, Winston, Originality, Copyleaks and others. All of a sudden, the author is publicly called out and shamed. This process is pitiless and fraught. One problem is that each of the detection tools has different strengths. Some are optimised for SEO and content marketing materials. Some are optimised for education assignments. Some detect AI by similarity to text in their enormous data sets of human‑written and AI‑generated text. Some detect AI by using statistical analysis of word sequences. Another problem is that once an accusation is made, the author’s defence falls on deaf ears.
Author‑Initiated
The third is author‑initiated external verification. This is the process supported by Proudly Human. For Proudly Human certification, the author applies for certification.
It starts with providing proof that the applicant is human and has a unique identity.
Next, the applicant (author) is expected to read Proudly Human’s de minimis standards, which reduced to their fundamentals say that the author may use AI for research purposes (human authors have had human research assistants for time immemorial), the author may use AI to bounce ideas (human authors have had family, friends, colleagues and editors to test their ideas for time immemorial), and the author may use AI for formatting tasks and rules‑based spell and grammar checking. However, the author must not use AI to write sentences and paragraphs.
Finally, the applicant uploads the article or manuscript. Yes, Proudly Human use AI‑based detection tools to detect AI. To overcome the limitations of individual detection tools we use several commercial detection tools, reject outlying results and average the remainder. If the manuscript passes this evaluation, a certificate is issued along with a QR code that enables future readers to verify that the application of the ProudlyHuman™ certification mark is legitimate. It is worth noting that the uploaded manuscript is not used for any other purpose. That is, it is not used to train AI models and it is never sold or shared with any third party.
Authors must choose, but it is obvious from the fact that I founded Proudly Human that I strongly believe that the author‑initiated verification process is the best way to identify human authorship.