We are on the cusp of AI developing traits or adapting in the same way living organisms do through evolution.
Mothwing eyes are intricate patterns that have evolved over millions of years through a process of natural selection. Initially, moths developed cryptic colouration to blend into their environments, evading predators. Over time, some species developed wing scales with microstructures that reduced light reflection, helping them remain inconspicuous. These structures eventually evolved into complex arrays resembling the texture of eyes to deter predators, a phenomenon called “eyespot mimicry.” This natural error-creation adaptation likely startled or confused predators, offering those moths an advantage — precious moments to escape. The gradual development of these eye-like patterns underscores the intricate interplay between environmental pressures and biological responses, resulting in the remarkable diversity of moth wing patterns seen today.
Critically, moths are not and were not conscious in or of the development of eyespot mimicry or any other evolutionary adaptations. They did not think, “Let us moths create a pattern on the wing to confuse the owl.” Evolution is a gradual, unconscious process that occurs over generations through the mechanism of natural selection. Individual organisms do not consciously choose or design their adaptations; rather, random genetic mutations lead to variations in traits within a population. Suppose a particular trait, such as the “eyes” pattern on wings, provides some advantage in terms of survival or reproduction. In that case, individuals possessing that trait are more likely to pass on their genes to the next generation. Over time, these advantageous traits become more prevalent in the population. This process occurs without the organisms having any conscious intent or awareness of the changes in their traits. It results from environmental pressures and the differential survival and reproduction of individuals with different traits.
AI can develop new traits with advantages, but that does not make it conscious.
AI (as of 2023) does not possess consciousness or intent like humans. However, consciousness and the development of new survival traits are unrelated. Indeed, neither is the characteristic of “independent decision-making” linked to survival. Adaption, evolution and the crafting of new advantage does not have any links to higher-order thinking or awareness.
AI operates based on algorithms, data, and programming, and any development of new traits or capabilities at the start (where we are today) will be a direct result of deliberate human design and engineering rather than unconscious adaptation.
Whilst simple algorithms can simulate processes that resemble aspects of evolution, such as genetic algorithms and neural architecture search, to optimise certain parameters or designs. Whereas AlphaGo (Google Deepmind) demonstrated the ability to learn and improve its gameplay through a combination of deep neural networks and reinforcement learning techniques based on data. AlphaZero went further insomuch that it doesn’t even need a data set to start as it is generated through selfplay.
What is happening here?
AlphaGo & AlphaZero demonstrated that processes guided by predefined objectives and criteria set by human programmers or researchers have enabled new traits (moves) to be crafted. These “AI” implementations do not have self-awareness or independent decision-making ability, but that does not prevent the autonomous development of new traits or adaptations in the same way living organisms do through evolution. AlphaGo’s capabilities are still within the bounds of its programmed algorithms and training data. It doesn’t possess a consciousness and is a product of human engineering, designed to excel at a specific task, but as with AlphaZero and AlphaFold, such systems can create new traits and advantages.
Whilst AI systems such as AlphaGo/ AlphaFold showcase the power of AI to adapt and improve within predefined parameters, it highlights that we are on the cusp of AI developing traits or adapting in the same way living organisms do through evolution.
We are on the cusp of AI developing traits or adapting in the same way living organisms do through evolution.
We should not be surprised as these AI systems are trying to mimic how humans and living organisms learn and evolve; therefore, a natural consequence is evolution, the development of traits based on “errors” that provide an advantage.
There is a very fine line between AI systems operating under the guidance of human-defined objectives and constraints and such systems creating errors and adaptations, essentially improvements based on patterns learned from data, just like nature. The observation becomes interesting as data can now be created independently of human activity using selfplay.
The development of new traits or improvements in AI based on “errors” in data will soon mimic the unconscious forces of natural selection, which we will only see once it has been created. Critically, we have to question “how will we know” because it is new and different. This question is one that regulation is not set up to address or can solve, and it is why regulating the AI industry makes no sense.
The development of new traits or improvements in AI based on “errors” in data will soon mimic the unconscious forces of natural selection, which we will only see once it has been created.
Questions for the directors and senior leadership team.
We are fully into automation and the implementation of AI across many of our systems, and indeed, we are using the data to make improvements that we did not see. Have you questioned if this new trait has an advantage, and how have you determined it has an advantage and for whom? Is the advantage for you, your ecosystem, your customer or your AI?
Thank you Scott for seeding this.