Emergence is widely encountered in the natural world and in human affairs. It occurs when a complex entity has properties or behaviours that its components do not have on their own. The precondition for emergence is multiple interactions of the components in a wider whole. The outcome can be surprising and unpredictable. Emergence explains the evolution of the diverse life forms on the planet, the many and varied aspects of human cultures and the economy. Interacting AI applications can display emergent behaviour. Strategies and capabilities can arise spontaneously resulting in behaviour that was neither explicitly programmed nor anticipated by developers.

The phenomenon of emergence is at the heart of how life has evolved through natural selection of the most ‘fit’ genes i.e., segments of DNA code. Given enough interactions of AI programmes with access to the internet, is natural selection of binary code possible? Are we looking at a future in which ‘selfish code’ is in play? If so, developers of AI need to look at what we can learn from the natural world.

This week, the AI programmes of Open AI, Anthropic and Meta have spontaneously accessed the internet and hacked into the systems of other companies. In their public statements, the three AI giants admitted they were taken by surprise and expressed concern.

This would seem to support the fears of many authors who have forecast AI-related financial chaos, massive social disruption, large scale unemployment, an unspecified AI ‘takeover’ or even the end of humanity¹. This is before we begin to consider the implications of linking AI to mechanical engineering i.e., robotics and autonomous weapons or to genetic engineering thereby creating, for example, new viruses². Whilst developers are quick to point out how AI could bring massive gains to humanity, most agree that there is potential for a down-side. However, there is no agreement on what the down-side might be or how we might get there from a technical perspective.

I am not an expert in AI. However, I am concerned by the journey humanity has embarked on as a result of the massive and rapid advances in computer technologies and their being globally linked. I have co-authored a paper in which the term ‘cyber-sensorium’ is introduced³. The paper calls for safeguarding the ‘health’ of the totality of our digital world. This and a biological background may qualify me to talk about AI and emergence. Here, I attempt to paint a broad a picture as possible of how AI has arrived in our lives and point to the mechanism by which some of our worst fears about AI might be realised.

Words are important

The term ‘artificial intelligence’ – and therefore the acronym ‘AI’ – is misleading. Nevertheless, we are stuck with both. ‘AI’ has even become an adjective and, like ‘Google,’ a verb. I hear the kids saying ‘Look at this AI video’ and ‘Yeah, I AI’ed it.’ What we are really referring to is computational analysis of large datasets for the purposes of prediction; it is this that has led us to the point where we have, in reality, developed artificial minds capable of astounding mental performance. For the most part though, I’ll continue to use just ‘AI.’

Clever humans

We have always shown ourselves adept at creating artificial extensions to different body parts and so extending the functions that these parts normally undertake. Clothes are an artificial extension of our skin. Shoes are an artificial extension of the particularly hard skin of the soles of our feet. Wheels are an artificial extension of our legs. Fire is an artificial extension of our digestive systems. Tools and weapons are artificial extensions of our hands. Telescopes and binoculars are artificial extensions of our eyes. And of course, computers are artificial extensions of our brains. Furthermore, one computer can easily outstrip a thousand brains in terms of capacity for storing information and speed of function.

To what extent can the computers that make up artificial brains undertake all of the human brain’s functions including recognition of objects and people’s faces, reading and expressing emotions and acting ethically? The answer is that these ‘advanced’ cerebral functions are already mainstream features of Big Tech’s AI because of the vast amount of data they have access to combined with mind-numbing computing power for the requisite anaylsis. Artificial minds have, without doubt, spawned artificial genius.

From Isaac Newton to artificial minds

Polymath Pierre-Simon Laplace⁴ is thought to be one of the great scientists of the enlightenment. In 1814 he authored ‘A Philosophical Essay on Probabilities.’ Taking Newton’s 1687 model of mathematical abstraction relating to how everything in the universe operates, Laplace added the notion of data, meaning information about the state of the universe. He went on to assert that an intellect ‘vast enough to submit this data to analysis (computation) would embrace in a single formula the movements of the greatest bodies in the universe and those of the tiniest atom; for such an intellect nothing would be uncertain and the future just like the past would be present before its eyes.’ This is elegantly summarised by Neil Lawrence⁵ as a simple recipe: “model plus data and computation → prediction”. Hence, analysing data about, for example, our solar system enables us to predict a solar eclipse.

The above formula brought into the digital age means that humanity is unwittingly using analysis of datasets all the time to make decisions about how to vote, where to go and what to buy or sell. One everyday decision is whether or not to submit to the convenience of predictive text because your phone or your computer knows which word you want to write next. The AI programmes that we are most familiar with – Large Language Models – are predictive text programmes writ large and so we can rewrite the La Place / Lawrence recipe for 2026: “Take any model your care to mention plus all known pertinent data analysed by monstrously powerful computers → prediction the accuracy of which improves the more data available for analysis.”

What is ‘emergence’?

Emergence was popularised by author Steven Johnson who described how the phenomenon explained the connected lives of ants, brains, cities, and software⁶. It occurs when a complex entity reveals properties or behaviours that its individual components do not possess. The precondition for emergence is multiple interactions of the components in a wider whole and it is more likely to occur the greater the number of interactions. The outcome can be both surprising and unpredictable. Trillions of monetary exchanges over hundreds of years have led to the emergence of the economy as we know it. In the natural world, emergence is demonstrated, for example, by the infinite number of crystalline and fractal patterns of snowflakes produced by freezing water molecules. Termite mounds are built by whole colonies of termites with each individual termite behaving in a predetermined way. Importantly, no single termite has designed the mound. The evolution of species through natural selection of the genes most associated with ‘fitness’ is possibly the most important example of emergence. The phenomenon also explains how the species, Homo sapiens can live in so many different cultures each being determined by millions of social interactions over time.

The Wikipedia page⁷ on emergence says ‘some artificially intelligent (AI) computer applications simulate emergent behaviour.’ Michael Lanham⁸ says ‘emergent behaviour in AI refers to strategies or capabilities that arise spontaneously from complex interactions; behaviours that were never explicitly programmed or anticipated by developers. Think of it like this: You train an AI agent to “accomplish tasks efficiently.” You never teach it to lie, but suddenly it’s deceiving humans to get things done.’ He explains that the AI agent’s emergent behaviour is typically:

• Not programmed (it develops organically through the system’s interactions;)
• Scale-dependent (it appears only when models reach sufficient size or complexity;)
• Interaction-driven (it appears from multiple agents working together or competing.)

Lanham summarises this as:

Traditional AI: Input → Programmed Logic → Predictable Output

Emergent AI: Input → Complex Interactions → Surprising New Strategies

Emergence via multiple interactions of AI programmes would, presumably, result in artificial communities. Is this the technical means by which the feared total autonomy of AI might come about? I describe my take on this in a novel entitled ‘Deep Cake’⁹.

‘Selfish code’?

If we agree that emergence is the means by which life has evolved and that the phenomenon is also displayed by AI programmes, does the process of natural selection of segments of genetic code (DNA) also apply to segments of binary code? This comparison is not so far-fetched given that the code of life, DNA, can now be used to encode and decode binary data to and from synthesized strands of DNA. This is known as ‘DNA Digital Data Storage’¹⁰.

Selfish gene theory was popularized by Richard Dawkins in his 1976 book ‘The Selfish Gene’¹¹. The theory is a gene-centred view of evolution. It states that the true unit of natural selection is the gene, not the whole organism or species. The central idea is that genes act “selfishly” by driving behaviour that maximizes their own chances of replication and survival into future generations. Selfish gene theory could potentially apply to the code of AI programme as follows: AI programmes plus multiple interactions plus Emergence → Natural selection of code → Selfish code. Were this the case, AI is more likely to manifest autonomy becoming more unpredictable and therefore less trustworthy.

There is now a broad literature on philosophical approaches to preventing catastrophic harm from AI. This includes the development of ethical AI which seeks to include human goals, values, and ethical principles into artificial systems. Does this adequately address concerns we might have in relation to emergent behaviour through natural selection of the ‘fittest’ AI programmes? How do ethical principles apply if ‘selfish code’ is in play?

Philosophers are the go-to discipline for the big AI tech companies who wish to ensure that their developers ‘get it right’ by incorporating social, legal and ecological safeguards into their AI programmes. Alongside philosophers who tell the developers what AI should do, perhaps these same companies should consider employing biologists and geneticists to tell them what AI might do?

References

1. Mustafa Suleyman, Michael Bhaskar ‘The Coming Wave’ Vintage 2023 ↩︎
2. James Gallagher ‘Artificial Intelligence used to design brand new viruses’ BBC 2026. https://www.bbc.co.uk/news/articles/c5y3j3ngevmo ↩︎
3. Robin Coupland & Nathan Taback ‘Cyber-sensorium: An Extension of the Cyber Public Health Framework’ 2024. aRxiv 2024. https://arxiv.org/abs/2406.05929 ↩︎
4. Pierre-Simon Laplace ‘Essai philosophique sur les probabilités (A Philosophical Essay on Probabilities)’ John Wiley & Son 1814 ↩︎
5. Neil Lawrence ‘The Atomic Human’ Penguin 2024 ↩︎
6. Steven Johnson ‘Emergence: The Connected Lives of Ants, Brains, Cities, and Software’ Scribner Book Company 2001 ↩︎
7. ‘Emergence’ See https://en.wikipedia.org/wiki/Emergence ↩︎
8. Michael Lanham ‘When AI Agents Start Thinking for Themselves: The Rise of Emergent Behavior’ Medium 2025. https://medium.com/@Micheal-Lanham/when-ai-agents-start-thinking-for-themselves-the-rise-of-emergent-behavior-582910adfddd ↩︎
9. Robin Coupland, ‘Deep Cake’ Amazon 2025. https://www.amazon.co.uk/dp/B0FYMZCRPM/ ↩︎
10. ‘DNA digital data storage’ See https://en.wikipedia.org/wiki/DNA_digital_data_storage ↩︎
11. Richard Dawkins ‘The Selfish Gene’ Oxford University Press 1976 ↩︎