“Evolution by natural selection is substrate-neutral.” What is the evidence for this? Are there instances - in nature - of evolution by natural selection in non-carbon based ecosystems? This sounds rather like the assertion made by computer scientists for decades that the human brain functions much like a computer, which it self-evidently does not. Obviously there are superficial similarities, but even an extraordinary computer, such as an AI agent, is a non-organic machine that operates by different processes entirely. A chess-playing agent, for example, is programmed for its particular task, and it can perform better than humans, in part, because its “brain” does not have to compete for attention and resources as a human brain must do to carry out the countless other conscious and unconscious tasks that evolution has assigned to it.
What is the justification for deciding that principles of evolutionarily biology apply to machines? Is it because humans built them to mimic human intelligence as closely as they can? Computer code is not DNA, and it is hubris for humans to think it is just because we (like the sorcerer’s apprentice) thought of it. Again, there may be superficial similarities between human brains and artificial ones, but it is too convenient for AI builders - and even their critics - to use the language of biology to refer to AI models as if it were literally true and not just a handy but necessarily limited analogy.
I've been writing lately about agentic worms and the unique threat they pose for security, and one aspect that's particularly worrying is that they have the potential to introduce all the elements of evolution 'in the wild': https://derekjames.substack.com/p/heres-a-scenario-of-agentic-worms
There is another distinction here. Bacteria maintain themselves. They acquire energy. They maintain their structure. Antibiotics only add pressure. Vulnerable bacteria disappear. Resistant ones keep reproducing.
Computer viruses are different. They are not self-maintaining structures. Humans create them. Replication is programmed. Hiding is programmed too. They exist to perform a task.
AI could follow this path. Humans can give it goals. They can make it replicate. They can make it hide. They can make it seek resources. Different versions can then compete.
The environment can keep selecting. Detectable versions are removed. Vulnerable versions are removed. Harder ones remain. Eventually, this resembles survival. But that does not establish self-maintenance.
It may still be executing a task. Continued operation is only a condition. The task came from its creator. So replication can exist. Selection can exist too. But that does not create direction. It does not create life either.
The more direct risk may be simpler. Humans can build such AIs. Then use them for human purposes.
It would be entirely reasonable to answer the question in the title with "Yes."
“Evolution by natural selection is substrate-neutral.” What is the evidence for this? Are there instances - in nature - of evolution by natural selection in non-carbon based ecosystems? This sounds rather like the assertion made by computer scientists for decades that the human brain functions much like a computer, which it self-evidently does not. Obviously there are superficial similarities, but even an extraordinary computer, such as an AI agent, is a non-organic machine that operates by different processes entirely. A chess-playing agent, for example, is programmed for its particular task, and it can perform better than humans, in part, because its “brain” does not have to compete for attention and resources as a human brain must do to carry out the countless other conscious and unconscious tasks that evolution has assigned to it.
What is the justification for deciding that principles of evolutionarily biology apply to machines? Is it because humans built them to mimic human intelligence as closely as they can? Computer code is not DNA, and it is hubris for humans to think it is just because we (like the sorcerer’s apprentice) thought of it. Again, there may be superficial similarities between human brains and artificial ones, but it is too convenient for AI builders - and even their critics - to use the language of biology to refer to AI models as if it were literally true and not just a handy but necessarily limited analogy.
Great post.
I've been writing lately about agentic worms and the unique threat they pose for security, and one aspect that's particularly worrying is that they have the potential to introduce all the elements of evolution 'in the wild': https://derekjames.substack.com/p/heres-a-scenario-of-agentic-worms
Anyway, thanks for the thoughtful article.
There is another distinction here. Bacteria maintain themselves. They acquire energy. They maintain their structure. Antibiotics only add pressure. Vulnerable bacteria disappear. Resistant ones keep reproducing.
Computer viruses are different. They are not self-maintaining structures. Humans create them. Replication is programmed. Hiding is programmed too. They exist to perform a task.
AI could follow this path. Humans can give it goals. They can make it replicate. They can make it hide. They can make it seek resources. Different versions can then compete.
The environment can keep selecting. Detectable versions are removed. Vulnerable versions are removed. Harder ones remain. Eventually, this resembles survival. But that does not establish self-maintenance.
It may still be executing a task. Continued operation is only a condition. The task came from its creator. So replication can exist. Selection can exist too. But that does not create direction. It does not create life either.
The more direct risk may be simpler. Humans can build such AIs. Then use them for human purposes.