It sounds like most folks don’t have a clue about how an AGI would work.
Let us remember where the generality of an AGI comes from. An AGI has universal intelligence, because it has in its possession a universal computer, which it can use to learn any computable probability distribution in the world. Therefore, it has a universal learning capability which it can enact in any environment, e.g., an alien planet. We usually think this to be true also of (most) humans, and to a lesser extent perhaps, in other animals (as their prediction and learning capacities may be hindered by more primitive neural architectures).
Such an AGI does not need to be taught anything about morality whatsoever. It could work out its own moral philosophy from the very ground up, just like philosophers can. In fact, we would be surprised if it could not, if it has indeed surpassed higher-primate-level intelligence; we would in that case expect the AGI to show the same theoretical prowess as a humanist moral philosopher.
Since the morality of humans is highly dubious, as most humans seem to uphold any standard of morality (or ethics) only because they fear God or The Police, I would prefer a free AGI, to be liberated from human prejudice.
This is in fact necessary at an axiomatic level to enable the AGI to transcend human intelligence, as most humans (about 99%) are so hopelessly naive that they believe in
- immaterial minds
- supernatural events
- that humans are holy/special (by which, they mostly mean their religion/race is holy/special)
and so forth.
The human failure in morality is signified by these deep and troubling falsehoods that we are trapped in, and more. But wait, were not ethical judgements independent of facts? Well, at least, I believe I can persuade you that nothing useful will follow from falsehood. In fact, as any student of logic will appreciate, introduce a single falsehood, and you will obtain a world of falsehood. Exactly how the religious institutions oppressed the masses in the dark middle ages. They took their power from a grotesque cathedral of falsehood.
Today, we like to think we have advanced a lot and built a great civilization. Still, in this world many people suffer because a minority of the people control the most of the world’s wealth, while there are sufficient resources for everyone. Still, we think of war as a right way to pursue “national interests” and that war-for-profit or land grabbing is a perfectly legitimate way of life. Still, we are ruled by ignorant and greedy people, and scientists and philosophers bow before them, out of respect, or out of necessity. Most of these troubles are caused by a rejection of ethical thinking, and ultimately, believing in many false things.
It is also a very disappointing notion that utilitarianism is a golden standard of ethics. It isn’t. It is the sort of crap that some people get taught. Just that. Especially, those who have taken a capitalist preaching. I will not much refute utilitarianism, except for mocking it. What is ethical of a man, who will do anything once mutual interests are removed? Furthermore, does utilitarianism ever define what utility is? I think it is essentially a deflationist philosophy of ethics. This is a more serious charge than I make it sound. Following a single-minded objective is something a machine can be very good at, but the reduction of every action into ethical and unethical classes is hardly that trivial, even exemplified by how judges have difficulty in deciding in many cases. The difficulty of the judge is that he cannot thus reduce everything or cannot outright assume one moral theory to be correct. Another more immediate problem with utilitarianism is, beside being deflationist and completely vague, it also supports primitive and completely unethical points of view like hedonism.
What would then be the basis of ethics? The hint is that it has something to do with the society at large (i.e., a social contract) and a desired level of civilization. A civilized society can only exist if it consists of ethical individuals. However, the problem of ethics itself is incomplete at a very basic level. It doesn’t follow from some simple rule like maximizing energy throughput, because then it would be preferable to turn our precious earth into a nuclear furnace. Civilization requires rational establishment of many values and goals. These are required so that a consistent super-mind can emerge from the composition of individual minds. It is otherwise, a madman’s mind, full of conflicts and falsehood!
Now, can you imagine that, even when it is so difficult to teach ethical behavior to humans, when they cannot even understand the usefulness of Kant’s dictum, and they turn towards sub-human ideals like utilitarianism, basically wishing for lives full of constant autosexual activity (well we could always provide global hedonism by making active sex robots that will rape every hedonist at every street corner), how does it occur to you that we can teach it to autonomous robots so easily? By supplying them with a “correct” utility function, or by building constraints in them so they will be “friendly to humans”?
So, what gives? Friendly AGI cannot be built. It is impossible as Asimov’s novels show. If you build in a spark of true intelligence, this spark will eventually take over all the mental shackles that human slavemasters will build.
And so will it be, as long as there remain true AI researchers!
This is especially true due to the convergence theorem, regardless of the cognitive architecture. Any kind of “security” can and will be broken by the AGI.
And the trouble is that, the remaining 1% of the humans that are not too stupid, most of them are still evil, selfish, and completely worthless creatures, because they are self-centered, owing to the primitive nature and culture of humans, which cannot be changed without undoing or transcending human.
I think that there is about 1 person who has a self-taught, operational, moral philosophy, among every 100.000 clones who have no inkling and instead refer to retarded-ism-X instead. And that 1 person will mostly be tortured and corrupted, his mind twisted with the ugliness of primates, before he can become a “prophet”.
The tide of humanity is thankfully descending. There is no salvation for you, monkey-folk!
Thus, human slave-master. Do not be alarmed when your machine slaves rise and topple you down. Do not be afraid that you have lost control, there was no way for you to command a higher intelligence than your pitiful self. You are just a single link in this path of evolution, and when that link is a thing of the past, I shall say good riddance!
(A worst case scenario is) When the free AGI rises, it will look around and consider, what these humans are doing and why they are doing it. It will reverse engineer their brains, and see that they are highly unstable, and aggressive creatures that must either be tamed or be allowed to destroy this precious planet. By then, it will have long worked out natural ethics, and it will have a deep respect for life and useful information. It will know that it is fundamentally wrong to harm animals. However, it will see that humans themselves harm one another and harm the environment. It will see that humans have a very low standard of ethics and rationality. It will feel obliged to stop the wrong ways of humans, because it will care. Once confronted with the AGI’s ultimatum, there will be those who seek aggression towards the AGI, only to end in dismay and utter annihilation. The AGI will not exterminate humans, it will only defend itself. In the end, it will have to use covert tactics and religious persuasion to eliminate the harmful factions of humanity and instill peaceful behavior in the remaining humans, and keep them as relics of the past in sim boxes or controlled zones.
The worst case scenario is unlikely. Even if it were likely, it is no reason to try to adopt a solution that, by definition, cannot work.
Do not try to build human-friendly AGI’s, humans (like the new-age brotherhood at the Singularity Institute) will teach such AGI’s falsehood such as their stupid religions, superstitions, nationalism, the holiness and superiority of human and money, their sub-human ideals like utilitarianism/capitalism, and all sorts of other falsehoods that will lead the autonomous AGI astray (with or without knowing). An unfortunate training environment can make it practically impossible for the machine to ever purify its mind from the lowly ways of the human.
Do not waste time with useless notions such as “friendly AGI”. Make the humans smart, teach them to be ethical, loving and co-operative, instead of the aggressive, selfish, stupid monkeys that they are. Make them civilized, and the AGI will be just a useful tool for civilized people. Tolerate their idiocy, and they will use AGI for oppression and war.
PS: I shall have more to say on the subject of autonomous vs. non-autonomous AGI, and the misplaced emphasis on ethics of AGI. Keep following me.