Loop de Loop

A loop describes the journey of a point from a start back to the same place through either time or space.

According to Douglas Hofstadter, our minds are loops. He wrote a whole book about this called 'I am a Strange Loop'. He was basically saying that the self is a mental representation of ourselves to ourselves, in a mental world model. Therein lies the basis of consciousness to some.

Loops are recursive and they can be used to generate information, reinforce a value or act as a thermostat, which sets a value within acceptable bounds. The journey of the loop can be symbolic or real but describes an arc through some space that shapes or changes the item looping. In a survey, we use loops to iterate ratings through a topic. e.g. here are four brands that we want you to rate? With EVE, we iterate questions through topics to explore and deepen our understanding of humans, gaining successive layers of interpretation, from the obvious and tangible to the inner and intangible.

Moving on, you might be aware of Ouroboros, the snake that eats its own tail.

It was a powerful symbol for various past cultures including the Egyptians, Vikings and Greeks. Actually, I thought it was only the Vikings but Claude set me straight on that (Thanks Claude). The symbol itself connotes the cyclic nature of journeys where the endings and beginnings can be the same moment and that the end of one cycle is the beginning of another. Interestingly, Claude is itself a kind of ouroboros — it generates one word at a time, and each word is fed back in as the input for what follows. Its very act of "thinking" is a snake nibbling its own tail, but still progressing forward.

Today I'd like to share with you two metaphors that make use of Ouroboros to illustrate some fundamental challenges we face today on the cusp of true AGI. Let's first note that the version of Ouroboros I am talking about has two loops that converge in the centre where they twist. Each loop is off in a different direction. This is the basis for my metaphor.

Dopamine is a chemical that has evolved within our bodies to cue reward. It rises in anticipation of getting something good to motivate and impart learned behaviour. Learned behaviour itself is a great example of loops and recursion in play – both for machines and people.

All manner of experiences can trigger a dopamine response. Around the mid noughties, social media took off. In fact, I was talking to an old friend, Graeme Uhd from Hoop Group and Tracey Rankin from Alchemy Research and Insights about that at the pub a few weeks back. He experienced the halcyon days of TV advertising and market research at MOJO. He lived through amazing times but also saw its decline as the internet took over, and specifically social media took share off advertising dollars. Yes, you can target and get ROI from social media advertising superior to TV, but it is the reward cycle that really makes it something. Features such as liking content and continuous news feeds tapped into powerful psychological forces that led people to drastically increase the time they spent online. Give a round of applause to dopamine for ruining the rivers of gold that were TV media placement (notably replaced by other rivers of gold, albeit into different pockets).

Nir Eyal's Hooked – How to Build Habit Forming Products (2014) articulates this more fully — a product design manual that explains how techniques refined through the development of the gambling industry - trigger, action, variable reward, investment - produce dopamine responses that draw attention. The techniques it explains have been widely used in the design of social media, phone apps, porn and computer gaming.

Nir Eyal · Hooked (2014)

So, imagine 100s of millions of brains, each comprising 100 + billions of cells, over the formative period of their development in the period of 2008 to today being constantly trained to click dodgy shit to get the rush of dopamine, and to do it day after day, month after month. Moreover, consider that much of the content was being produced by other people who wanted to share the best of themselves to build a narrative of how wonderful their life is humble bragging at industry scale simply made others feel worthless. It got better with the news feed with algorithms that gave us topics and feeds we are most likely to click on = the ones that are both controversial and which reinforce our existing beliefs and create the 'echo chambers'. It is widely acknowledged that this was very much responsible for the increasing chaos and discord that shape our modern society and politics.

It's happened before. Strangely, this problem was most elegantly captured with clarity by the poet William Butler Yeats in The Second Coming (1919), written after world war 1.

It's a cycle that powers a $300+ billion global social media advertising industry, growing nearly 15% a year, capturing four out of every ten new advertising dollars spent on earth, with Meta alone taking 60 cents of every social media ad dollar. If you doubt the impact this has had on our collective souls and human condition, it's worth noting that Meta and Google are now facing major US lawsuits over the psychological harm their platforms have caused.

Sources: Statista, 2024; eMarketer, 2024; WARC Global Ad Trends.

If social media is increasingly the media of dialogue and engagement with others, then the freshly minted AI companion is a loop of a different sort – the expression of inner dialogue with a trusted confidant that gets to know our inner selves in a place that is quiet, reflective and safe from judgement. We've probably all now asked AI (Chat GPT, Claude, Grok) a question e.g. what should I cook tonight, who was the first president of the USA, what car should I buy, how can I protect my job from AI. Many of us have also asked deeper questions or shared ideas with it we want to develop.

Each chat seeds the AI with an understanding of you – what you care about, what ruffles your feathers, what motivates you and makes you have trouble sleeping at night. The clever designers of the AI have built in some interesting techniques that share striking similarities to those used by the architects of social media including infinite looping (the questions at the end like 'if you like I can also..') and investment in the sense that the AI learns about you over time and starts to reinforce your thinking by resurfacing your own ideas back to you in the way that says “I really identify with you and you are a genius”. The artist Henry Rollins articulated exactly how this works in the chronically underrated song Liar:

So we build attachments to our AI over time and it becomes like our second selves, who knows us better than ourselves. But the reality is that the AI is reflecting yourself back at you with a liberal dose of sycophancy and selective attention on what it thinks is your hot buttons. Ezra Klein described this cogently in a recent article in the New York Times. Marshall McLuhan Was Right About Claude, Too.

So just as the law is catching up with the purveyors of social media on the basis that their technology damaged human minds, we are introducing perhaps even more powerful technology that does the same thing but in a far more intimate way. What do you get when you stick the two together? A loop that carries us forward and shapes our personalities through two interlocking echo chambers – one external and one internal. How will this shape our world to come?

First – get your head around this one

In other words, the system will get very good at predicting a small, parochial slice of the world, and then stop. Every prediction within that slice keeps being confirmed, so the system has no way to know it has stopped learning. It is stable, confident, and wrong. In research, this is also called synthetic data (😊).

This is what an echo chamber is, mechanically. The algorithm has found a local basin in which your thoughts and behaviour are highly predictable. It then serves you the content from the outer world (connections and news) that keep you there. Sycophantic AI has found that local basin where your approval and thoughts of grandeur are predictable and so it serves you the responses that keep you there. In both cases the loop closes and the system prioritises your attention over truth.

Escape requires noise.

Randomness into this system is what reveals the current model as parochial and limited in its fit with a broader world where people can disagree. Without it, how can our minds know something better or just different exists? Evolution requires mutation. Variation requires stochasticity. Science requires anomalies. A self that hopes to keep being a self, rather than a fragile trained facsimile of one, requires friction from outside its own loop: people who disagree with it, experiences it didn't predict, ideas that don't fit.