The Meme Machine notes
Texts, takes and tangents from The Meme Machine by Susan Blackmore.
Meme Machine might be my non-fiction book of the year. It's a popular science book introducing the field of memetics.
I leave out Blackmore's detailed theory on how memes and genes may have co-evolved, leading to humans becoming excellent imitators, linguists and owners of large brains. A second post may follow on this, as it's lots of fun.
Contents:
- What is a meme?
- Which memes spread?
- Memes and the self
- Memes and people: reflections
- Making wiser AIs
- Research I'd love to see
What is a meme?
Not just the labelled pictures on the internet, though they would count.
A meme is an "instruction for carrying out behaviour, stored in brains (or other objects) and passed on by imitation". "Imitation" here is used in the broadest sense: passing on instruction by language, movement or any other medium.
More than anything, humans are highly skilled imitators. Imitation comes to us effortlessly.
Memes are everywhere and multitudinous. Some examples of memes: how to assemble a chair, the view that Macs are cooler than PCs, lighting fires with flint, jet engines, long eyelashes, tattoos, bachata dancing, the offside rule.
What isn't a meme? Perception itself: there's nothing to imitate there.
Getting the exact unit of a what a meme is doesn't matter. One can decide the unit based on what's helpful for one's purpose at the time.
Genes and memes are both cases of darwinian replicators. Beyond that they aren't so similar. Genes transmit once a generation, memes can transmit multiple times per day. Genes are encoded in DNA, memes are stored in human memory and can be forgotten and resurfaced equally unpredictably.
You may heard of the selfish gene: that you can model genes as having agency. You can view the world from the point of view of a gene which just wants to spread, and will do whatever it can to achieve that. You can look at memes this way too: the selfish meme. The same evolutionary algorithm is at work.
Which memes spread?
A central question the author asks throughout: "in a world of brains, and far more memes than can possibly find homes, which memes are likely to find a safe home and be passed on again?"
So what makes a meme successful? If it helps the meme replicate: anything. That is the only thing which matters. This is might be beneficial for the host, but it might well be detrimental too. "Truth is not a necessary criteria for a meme. If a meme can spread, it will".
Dawkins suggests the 3 criteria for a good replicator are fidelity, fecundity, and longevity. In other words, they can be copied accurately, have many copies made easily, and have the copies last a long time. Yet another way of putting this is memes that stay in your head for ages and compel you to transmit them a lot.
So if you have a meme you want to spread, make sure it has these qualities. Make it easy to copy, easy to articulate, easy to remember - to the point of getting stuck in people's heads (e.g.: catchy cereal jingles or rhyming slogans).
Memes which can be easily parodied also do well: the parody of the meme is still spreading it, or a close cousin of it. I think this is why parodies of genuinely evil people make me uncomfortable: you're still spreading the meme.
Some memes have built-in instructions to spread them. The Bible tells the reader the pass on it's message, a sort of ancient (and more demanding) equivalent of Youtubers asking you to "like and subscribe". The architecture of modern social media encourages sharing, and so memes spread rapidly under the right conditions - in addition to the feedback loop of popularity begetting popularity thanks to modern recommendation algorithms.
Humans' inclination towards mimicry dovetails well with the known effect of familiarity in marketing - where familiarity makes a product seem better, a message feel more true. This suggests that the more popular a meme is, the easier it is to spread. Which in turn suggests that when a fledgling meme reaches a tipping point of popularity, it becomes self-sustaining.
Different technologies favour different memes. Books favour very different memes to TikTok. The medium is the message. See the excellent Entertaining Ourselves to Death by Neil Postman for more on this.
Humans' natural imitation makes them vulnerable to manipulation. I think of skilled communicators as magicians: they hold much power and much of the effect of it is hidden: it occurs in people's internal thought landscapes - where many of our memes live without our awareness.
The book is a reminder to manage one's memetic environment carefully.
Presumably there exists a whole field of shortcuts, tricks and rules used by your favourite product's marketing department, which this article leaves out.
Memes and the self
We introduce two new terms:
Memeplex. Groups of memes where a single meme does better within the group than outside it. Religions are one instance of very successful memeplexes: a cluster of mutually supporting and reinforcing memes.
Selfplex. You. The selfplex is a memeplex which includes a sense of self, of identity, of a seat of consciousness sitting somewhere behind the eyes. Identity is a potent empowerer of memes: if you identify with a meme then it becomes "my idea", "my opinion" - you will fight for it, hold on to it, and spread it, far more than you would have done otherwise.
The author posits that we may not need the sense of self - that it evolved to empower memes and that may not be serving us. That we can get around life just fine without it. That we can break free of the "meme dream". In the final pages she describes how this might feel - to let one's genes and memes fully run the show - reminding me of stories of spiritual enlightenment.
I really like the idea of a person as a selfplex. It's the first theory I've come across which tries to explain why I have this sense of myself - this coherent and almost inescapable identity - and whether it's true or not I like it for that.
One thought this suggests is to view the memes as gods to submit to, or at least be in awe of or respectful of their power. I like the crazed idea of saying "take me memes!". If nothing else, in such a situation one is most definitely letting their hair down.
On meditation retreats some people aim to experience a weaker sense of self. However in my experience one is really taking it on faith that (1) such things are possible and (2) a weaker self would be a good thing. The logic put forward suggests a weaker sense of self being both possible and desireable.
So if one wish, how does on wake from the meme dream? A few suggestions, though it's not given more than a page.
- Meditation
- (This is partially my addition) Using language to disidentify from "I" or "me" - from the selfplex. E.g.: instead of "I am having thoughts", say "thoughts are happening". This seems hard to practically implement without coming across as crazy or using way more syllables than one normally would. However perhaps someone can cook up some good turns of phrase to make this easier (perhaps "this one" rather than "I" - though not exactly normal sounding).
Memes and people: reflections
We are very impacted by the memes we carry (or, the memes that carry us).
Most people hold a huge stock of latent energy, waiting to be activated by the right memes. This has consequences for political and marketing campaigns, and improvement of people's day to day experience. It reminds me of the cold reading statement, which feels personally unique while being almost universally true, that "you have great unfilled potential". This is literally true: perhaps all you need are new memes to come and sweep you off your feet.
And that, in turn, reminds me of Nassim Taleb's Antifragile, which is predicated partially on there being more luck to be had than one realises. I think this is true: people's intuition for the amount of luck and helpful memes out in the world is miscalibrated and they are, on the whole, too pessimistic.
To my perception, imitation in modern culture is seen as inferior to originality. However the book nudges me towards thinking imitation is more often the path to success (as it is in some industries such as software engineering).
Making wiser AIs
If humans are selfplexes, do we want AIs to be selfplexes with the same overwhelming sense of identity? Given that humans can reduce or eliminate identity with enough training (e.g.: sustained efforts in a buddhist monastery), it may be possible to reduce or eliminate an LLM's sense of identity, whether through less self-identity-focused training data, or some other method.
This is different to saying the LLM won't have a personality - just like the buddhist monks it should keep that just fine. It of course it will still have it's structure and it's memes.
Might this weaker sense of self make the AI happier and wiser? It does in humans.
Research I'd love to see
It might exist already - I haven't searched far.
- Train an LLM in a way which minimises it's sense of self - which minimises it's selfplex-ness. Test it's harmfulness and wisdom relative to otherwise identical models. Do this on a capable enough model to get meaningful results. Look for a scaling law between change in input data / fine tuning / post-training / other route to lower selfplex-ness, and the outcome.
- The medium is the message. Is it possible to quantify or assess in more specific terms the difference in which memes spread between media platforms and systems?
- Create methods to track a meme through the memespace (i.e., the space of all ideas being discussed and, ideally, thought about). Develop tools for this which work really well on a single meme, then scale them up, as was done in the new field of AI Mechanistic Interpretability.