Predictive Processing in a Dead and Dark Internet World
In 2023, anthropologist Maggie Appleton published an article titled “The Expanding Dark Forest and Generative AI”, in which Maggie discusses the problem posed by the deluge of content produced by various AI models. This unmitigated flood, which by some estimates accounts for more than 49% of Internet traffic, is making the world wide web an evermore dead and dark forest.
Referencing Yancy Strickler’s article “Dark Forest Theory of the Internet”, Appleton describes the necessity for organic human interaction to exist almost entirely within niche and remote corners of the Internet, lest one be assaulted by bots, targeted ads, click/ragebait, and scammers of all shapes and sizes. The Internet has become a virtual Amazonian rainforest where the forest floor is fraught with danger and safety is behind gatekept communities.
The Turing Test is Dead, and We Have Killed It
Alan Turing proposed his namesake test that has served as a baseline for computers to suitably pass as human, in which an observer cannot differentiate between a computer and a person in a double-blind interview. As AI has become more advanced, not only has it become harder to pick a computer out of a human line-up, it has become difficult to pick a human out of a computer line-up. This, as Appleton notes, is the “Reverse Turing Test” and we are beginning to fail it at an alarming rate. We are beginning to call into question the “realness” of everything we see online. So how do we survive and thrive in this new habitat?
Create Against the Machine
Appleton proposed five solutions to this “realness” problem:
- Triangulate Objective Reality
- Be Original, Critical, and Sophisticated
- Develop Creative Language Quirks, Dialects, Memes, and Jargon
- Consider Institutional Verification
- Show Up In Meatspace
Of these solutions, numbers three and five are perhaps the most practical. Until we reach the technological singularity of Advanced AI, after which we may have more pressing issues, it is unlikely automated content can compete with memes, slang, and all of the subtleties that go with them. This is perhaps even more true in person where body language, facial expressions, and other forms of nonverbal communication develop over several millennia confirms that you are indeed human and are interacting with other humans. For one reason or another, we have a developed sense of the uncanny that will probably server as a new age Turing Test (at least until Replicants are a thing).
My least favorite, and perhaps Appletons by her own description, is number four — verification. While there are some cases in which Verification is appropriate, such as job search sites, financial institutions, etc., numerous issues arise from attempting widespread implementation. Anonymity offers a level of protection for anyone who wanders through the forest. I remember the common adage was “Don’t use your real info on the net” in the late 90’s and early 2000’s. Of course, that anonymity can be abused by malicious actors for nefarious purposes. But if we had to, what could be done?
Well there are a few positions one could reasonably take. From a laissez faire perspective, we could collectively accept that anonymity is a thing we possess that can be used or abused, like a car, bottle of wine, or tray of cookies. Rather than try to litigate best practices from the top-down, we push from the bottom-up to impose an onus of responsibility on the end user to prove they are legit. Not necessarily posting utility bills and drivers licenses for the world to see, but in their actions, language (borrowing from number three), and the “posting-a-crumbled-sheet-of-paper-with-three-fingers-and-todays-date” method.
At the end of the spectrum, the usage of federated identity could be a reasonable middle ground. As opposed to verifying with every single website you want to visit, these sites would rely on a federated identity structure similar to what is already in place that allows you to use your Facebook or Gmail account to authenticate. Of course, not many people will want to link their Facebook and LinkedIn accounts to their Reddit and underwater basket weaving forum accounts. New third-party federated identity providers will need to provide a means of anonymizing their accounts, likely using a hash or salt-based system that provides blind authentication. Alternatively, sites could opt for individual authentication or none at all. Ultimately, it is likely that individual sites will independently choose whether they want to rely on Verification of humanity.
The Internet and Its Consequences…
Now I must admit to a bias of being somewhat of a cynic. I give Appleton sincere credit for being an apparent optimist regarding the inevitability of intellectual goalposts shifting higher as the mundane and unoriginal become norms of generative AI. While I agree with the underlying premise, I am unsure if we will collectively rise to the challenge. There is a growing body of evidence regarding the impact social media, our diets, lifestyles, and environment on intelligence. As such, while there will always be individuals who become trailblazers, creators, and originality, I am not sure about the rest of us. It is my profound hope we are merely in a gully, and there is an upward trend in our future. If we are destined for such a trend, it will likely be a result of our collective adoption of the other methods put forth as well as solutions to the myriad of problems we face in the many other aspects of life beside this single pane.
Triangulating Reality vs. Predictive Processing
Finally, option number one is what spurred me to write this article as I came across Appleton’s work about the same time as another subject — Predictive Processing Theory. Truthfully, I owe my exposure to both topics to two excellent videos by Youtuber Kyle Hill. While reading into both topics beyond the videos, I started thinking about how the theory of predictive Processing in neuroscience represents a potential problem with the first solution from Appleton’s article: Triangulating Objective Reality.
The World As We (Sort Of?) Know It
To briefly summarize, Predictive Processing Theory suggests that we live in a hallucination. Our minds do not create reality from our senses, rather it has an ongoing working “model” of the world around us that goes more or less unchanged until something breaks with its myriad assumptions. As such, we do not actually live in a “real” world, in fact we are closer to the “Brain in a Jar” idea. The implications of this theory are many, and it is not without its criticisms. Yet, it may help explain why polarization between groups can sometimes be irreconcilable.
If there is a fundamental disconnect between reality and how humans see the world, then triangulating objective reality becomes a vote rather than an observation. This is further complicated with specificity. For example, if you go outside and look up and then look down, and I told you the sky was blue and the grass was green, I would wager there's a very high chance you’d agree with me. Now let us think back to a certain dress from a few years ago that split the world into two camps: gold and white vs. black and blue. Both sides insisted the other was wrong. Now can we agree on what qualifies left-wing and right-wing politics?
The point being: as specificity increases, so too does the difficulty in triangulating objective reality. So too is it increasingly difficult to find common ground.
But is it essential to agree on what makes someone a Protestant and another a Baptist? Maybe. It depends on what needs to be agreed upon in order for a disagreement to matter. I would describe this as a “reality ladder”. Whatever rung you’re on, all of the preceding ones need to be in place to have good footing. What is the point in debating the differences between Nihilism and Absurdism if neither can agree whether the sky is green or the grass is blue.
Perhaps the debate is proof enough of one’s humanity. If you take the same stand regardless of how I prompt your answer, you might be human after all. Or maybe if I ask a slightly different way, I can prove you’re an autogenerated entity.. or maybe I am? What is important to understand is as the forest floor continues to merge into an amorphous blob of boring and soulless content, the areas where we can all find “common ground” are beginning to disappear.
