In the age of emergent generative AI anyone’s default assumption should be that any text or image they encounter is generated by a computer. It requires proof to accept the author of something is a human being. In other words, a Reverse Turing Test needs to be applied. The Turing Test is a thought experiment on whether a computer’s responses might convince someone they are a human’s, and the Reverse Turing Test is meant to see if a response can no longer to be assumed to be from a computer.

Completing the Reverse Turing Test at this tme requires two things in my opinion:

1. A resource by the author that can be seen independently from any artefact under scrutiny. This is what a/this Reverse Turing Test page is for, or in this case what this entire website is for. I have been writing in this space for over 2 decades. It is hard to maintain something fake consistently over such a time. My blog is my avatar, it attests to me being a person. Also this site links to other places on the web where I interact or share material, which are proven to be by me because those places and this one link to each other in a specific way.
2. Whenever a text or image is shared it should contain meaningful references to resources and experiences outside of that text or image. One thing I notice is that generated texts are always self-contained and not entangled in other actions, expressions, tangents or artefacts. There cannot be a reference to the world of an author in a generated text because there is no author with such a world. Every human and each of their expressions however is embedded in a network of meaning. While no generated artefact is. (And yes, if you notice that SEO and corporate texts would fail a reverse Turing test in this way, that is not a coincidence.) Therefore I will aim in each of my expressions to show such links to a broader meaningful context than just the expression itself. For instance by placing it in a stream or timeline from which can be seen how an expression is an output of an ongoing process, by explaining its genesis in terms of previous expressions, i.e. linking to previous things, or in terms of what other people’s expressions mean as source to my expressions (e.g. how I associated from what someone else wrote, or how something is part of a distributed ongoing conversation).

A third future element might be vouching for someone being a human being. I could link e.g. to Peter and attest he’s a real person (and back it up by showing how both he and I have over time documentend various meetings and interactions). This is akin to how in the ’00s there was a network of people mutually certifying server certificates, or how in apps like Threema the trust level of a connection can be increased by meeting in person.

Of course all of this is an effort, especially the third one. But the first two build on the fact that anyone by living creates traces of themselves, and the internet is a place where one can leave much longer persisting traces more easily. It is a bit of effort to make the link to your own past traces more explicit than you perhaps normally would, but it is more costly to need to fake such a rich context if you want to convince someone a generated text is human authored. Unless you already have such a rich context, because you indeed are human. This is akin to surveillance: I don’t use encryption to avoid it per se, but I do use it to increase the overall costs of surveillance. If it is more costly it by necessity will be deployed less generally and more focused.

There is one thing Reverse Turing doesn’t cover: a real person sharing generated things as their own. Although currently I think the second point still would help here too, unless someone adds context and references to a generated artefact, there is maybe much less need: if someone claims generated texts as their own, they are also responsible for those texts, and cannot afterwards renounce those words as not theirs to avoid responsibility. If you claim them you also own the consequences. Generated or not.

The Reverse Turing Test notion I adopted from Maggie Appleton’s talk on the Web as Dark Forest.

This page is to attest that I, Ton Zijlstra, am a human sharing writings, images and other artefacts on the internet. Where I share something online I will reference this site and show context. When I do use generated artefacts I will label them as such clearly and put them in the public domain.