More philosophy >

Ex-Treme Speculation

Tremes are a postulated "third replicator" obeying the rules of Darwinian evolution, in similar manner to biological genes and cultural memes. Their definition is broadly similar to that of memes except that they pass between, and are evolved by, computer systems without human knowledge or intervention. But does the idea hold water? I would suggest that it does not.

A fun image capturing the essence of tremes.

Certainly, computer data exists which displays such properties. Computers increasingly pass vast amounts of data to and fro, adapting and re-purposing it to their programmed needs and without any human interference in the process. But does that make the information distinct from human memes?

In computer science and the real systems it spawns, I will here distinguish two kinds of information. The more familiar type is overt data. It might be encrypted but it is stored in human-defined form in human-designed repositories and is in principle human-readable. Indeed, much of it does surface from time to time. Even hitherto unknown and unsuspected information communicated and evolved by the computers pops up in search results, advertisement targeting and other "smart" applications and services. We all know the pop-ups on social media convinced that we will like something we have never come across or are friends with someone we have never met. The rest of it, never seen by human eye, is nevertheless still there, open to discovery were a system administrator or developer to go looking. Should such information be subdivided into memes and tremes, on the basis of whether or not a human has been directly involved in its processing yet? Frankly, that would be absurd. For example suppose one computer passed on to another my job CV complete with computer-generated metadata, in which it had somehow picked up the wrong email address for me. I then went into the new copy of the metadata and corrected it. Would that transform it from a treme into a meme, from one kind of replicator into another? I think not.

The other significant kind of information is that stored in a certain kind of computer system called a neural network, once it has learned a particular skill. As the name implies, such a system mimics the design of a biological brain, though in a very crude and massively over-simplified way. Such networks cannot be programmed with software to perform a specific task in the usual way, but must be taught to carry out the task through repetitive trial-and-error learning. The more learning data is presented to it, the more its understanding evolves and its ability to perform the task improves. When a neural network is taught in this way, for example to read handwriting, the information it learns is spread throughout the network in a way that humans cannot readily access. A "soft" network, implemented as a software model rather than a silicon chip, can then be copied and passed to other networks, some of which may have more data available and may further evolve their skills. So, with no human ever directly involved in the creation and evolution of the data, can it now be distinguished from memes?

Again, I would say not. Firstly, it is sometimes possible to infer the way in which the information is stored by methodically tinkering with specific "neurons" and observing the effect on the system's skills until, ultimately, it can be unravelled and ingested by a human. Even the most complex and sophisticated AI networks are in theory open to such hacking. And sometimes, active intervention by the human is not even necessary to get at the information as it can be inferred simply by watching. This happens in the real world when some auditor smells a rat and analyses the patterns revealed in say the behaviour of a social media platform. A typical example was the discovery that several smart "AI" based platforms had taught themselves to be racist by copying and learning from human Internet memes. So once again, the fact that the information is supposedly inaccessible does not mean that it actually is. Moreover, in half a century of involvement with computer systems I have yet to come across any class of computer information whatsoever that was not inherently accessible to humans in one way or another, however much hard work that might take. So once again, the case for a third replicator beyond the meme fails.

Even if I am wrong and some human could demonstrate that data inaccessible to them was being communicated and evolved by computers, would that make it a treme? Well, let us consider the parallel case of biological systems. Since the days of Sigmund Freud and his discovery of the subconscious mind, neurologists have been discovering more and more information being processed by the subconscious and wholly unconscious areas of the brain. Some of it even gets communicated to other individuals, a good example being somebody's unconscious body language communicating emotional states and even being responded to by equally unconscious movements of the person next to them. Or, take the famous waggle dance of bees. One bee communicates to the others information about where to find nectar. That information may then be evolved by other bees, either through mistakes in copying the dance or further visits to the site. The particular significance here is that the intelligence of a bee is of the same order as that of the most sophisticated AI computer systems in existence. (Indeed, a modern server farm or cloud service might justifiably be compared with a beehive in terms of system complexity and sophistication.) The information communicated in a waggle dance is no more nor less a "third replicator" than the metadata concocted and communicated by computers. If the one is at best a primitive meme, then so too is the other, there is no third paradigm about it.

Finally, it has been suggested that human and AI information is somehow qualitatively different, that the implementation of computers on purpose-designed silicon is somehow qualitatively different from brains made of naturally-evolved carbon chains and that only the biological brain can have sentience and conscious experience. Today there may be some trivial truth in that - there is a case that the waggle-dance of a bee is qualitatively different from the intellectual concepts we are discussing here and that the bee lacks sentience. But, as the bee also demonstrates, that is a matter solely of complexity and organisation and not of architectural origin. Given some future AI system as complex and suitably organised as the human brain and it will be as qualitatively sentient as you or I. What the chemical substrate of two such brains or their origins might be will be wholly immaterial, there is no scientific principle which might distinguish on that basis alone between the information each communicates and evolves. Nobody (and that includes the AI) will care a jot which kinds of substrate a meme is sustained and evolved by at any given moment, it is a meme and that is that.

In a nutshell, the suggestion that a computer "treme" might exist as a fundamentally distinct object from a human meme is utterly without scientific foundation. The theory appears to arise from superstitious beliefs in consequence of a fundamental ignorance of digital technology. There is only vague and emotive aspiration bordering on pseudoscience in its support, not one testable scientific hypothesis put forward. All the genuine science points the other way.

The real tremes

But I do believe that there is a third replicator waiting in the wings. Its first manifestation will probably be an AI system too, but one which understands its own design and is capable of improving on itself. Over time such self-replicating AIs will increase in number and evolve greater sophistication, with intelligently-driven mutation taking over from (or possibly merely supplementing and synergising with) the randomness of Darwin. Such an AI system - design plus designer, memotype plus phenotype - would be a true third replicator, a new treme.

It is also possible that we humans will learn enough gene technology to do the same for ourselves. Dolly the sheep already points the way to the basic biological side of things, while genetic engineering technology is also making enormous and accelerating strides. We would then move from being coexisting independent gene and meme replicators, to a unified gene+meme replicator in which the memes are the engineering specification for the genes. We too would become the design plus designer.

Maybe too the current advances in biologically-enhanced silicon chips, DNA memory devices, silicon-based neural implants, smart prosthetics and the like will by then have converged to create the true cyborg, and there will be no distinction between human and artificial intelligence even at the substrate level. What price the distinction between biological and AI systems then? Such speculation is entirely based on anticipated progress in today's technology. Once again, I challenge anybody to come up with the slightest scientific dividing line between the thoughts of superintelligent humans, AIs or cyborgs.

Updated 14 Oct 2023