Sentient Ai will probably kill you
I was listening to Hannah Fry (the Queen of geeks on 'The Rest is Science' podcast) get all sceptical about the prospect of sentient AI. She reckons we humans evolved sentience through natural selection to have "theory of mind' needed to model and predict the behaviours of others based on what we think they are thinking. Sounds about right to me.
But she finished up saying AI sentience might be possible to evolve artificially if we put an AI using machine learning in an environment with the right evolutionary pressures. “Should we find a reason to do so.”
[Edit: "Machine learning" is where a machine alters its own code to bettter achieve a programmed goal.]
This got me thinking. With the current global push for dominance in drone warfare this process is more than likely happening , with autonomous combat drones being 'trained' using machine learning to model and predict opponent drone battle tactics in the field.
That means the first sentient AI developed will very likely be a skilled killer with no social skills, nor Asimovian priorities about preserving human life.
Have a nice day
Last edited by
Sputnik;
May 5 @ 5:43pm
Among many sentient animals only few have that prediction behavior like great apes.
As for the combat drones, there's no automatic evolution in them. They're trained by humans in the end.
But I mean a recursive model of reality that includes similar models held by other agents. 'Recursive' since it also models itself,.and so can be self-reflective and thus self-aware.
When a machine is fully 'trained' by a human that's simple 'computer programming'.
'Unsupervised Machine learning' is self directed learning, whereby a machine alters its own code to bettter achieve a programmed goal.
'Supervised machine learning' is a combination of the above.
https://en.wikipedia.org/wiki/Machine_learning
According to the article “US air force denies running simulation in which AI drone ‘killed’ operator” from the web edition of The Guardian, the scenario appears to have been a human‑on‑the‑loop thought experiment. In contrast, the methods currently being considered for cutting‑edge weapons development are based on a human‑in‑the‑loop model. It even seems that manned systems may function as nodes within these networks.
This is viewed as an essential component of network‑centric warfare, where operators—whether on the surface or in the air—are provided with dynamically selected and context‑appropriate decision‑making information.
Regarding LAWS, it seems that many countries are in the process of establishing their own legal and operational self‑regulations. However, in light of the earlier clarification about the accuracy of the information, I would prefer not to cite specific sources or URLs in this post. I hope you will understand.
Sad, that was a cool story.
Regarding laws, my own country Australia is really dragging ts feet here- publishing 10 voluntary "guardrails" in its 2025 Guidance for AI Adoption
https://www.industry.gov.au/publications/voluntary-ai-safety-standard/legal-landscape-ai-australia
On the subject of cognitive evolution, Ostriches have evolved a certain defense strategy, but I'm unsure just how effective it might be against sentient killer drones.