LOADING (BROKEN) VALUES INTO MACHINE INTELLIGENCE
Human Values are a mess.
Much of what most people believe is moral, is clearly unethical. We can clearly see how slavery and exclusion from property rights is now considered unacceptable in most places of the world, and that this was not always the case. In years to come, the consumption of other animals, and assault upon children may be equally reprehensible across society.
Beyond moral relativism and the tyranny of culture-bound taboos and sacred cows, how can we be sure that our declarative beliefs about morality are sound, and not simply ex-post-facto justifications for the particular qualia of our innate desires?
How can we work towards being better people, if we cannot define what 'good' truly is?
Certain ethical precepts such as the Non-Aggression Principle are a great start, since they attempt to frame universalisms based upon First Principles. First Principles are a step towards an objective morality, or objective good, that is not based upon the values of the cultural milieu itself.
I suspect that the premise of Lloyd deMause's school of Psychohistory is correct, and that the march of human progress has only ever been possible when childrearing practices advanced sufficiently across societies, enabling successively less dangerous disorders of personality to create a more evolved common weltanschauung.
We are broken. We are less broken than we used to be, but human beings are a mess. We're trapped within a well of madness, and only self-knowledge and therapy can begin to undo the damage that we would otherwise imprint ad infinitum upon the following generations.
Humanity is a seething mass of wailing dissociated alter-states, desperately clawing for any momentary chance to quell unyielding trauma. This manifests in religion, statism, tribalism, addiction, and a wide variety of defense mechanisms.
In honestly recognising how broken we are, and the society-wide neuroses that pervade, we can look upon ourselves as an outsider might. Objectively, what are humans like, as an animal species? Dangerous, but predictable. Capable of balanced virtue and rationale, and the most savage barbarism.
We are traumatised bonobos who therefore act like chimps. It takes immense effort to overcome the gravity well of tribal norm conditioning upon the true self. Yet only through this process alone does civilisation evolve.
Our human lives are a tightrope walked towards a kind and intimate way of being, below which lies a mire of thoughtless violence and wilful ignorance.
This species driven insane by collective PTSD still has promise. The question is how much worth has this promise, and how quickly ought it be realised.
This applies directly to a discussion of safer artificial intelligences.
As I see it, there are 3 major questions in Friendly AI:
1). Engineering - how can we build something so tricky, uncertain, and with many potential points of failure.
2). Defining Friendly - friendly to whom? Friendly to humans on our level of flawed morality today, or objectively ethical?
3). The Desired Final Outcomes to work towards in successfully designing a 'Friendly' AI.
The problem is that even if we make an AI that is extremely kind, gentle, and ethical, it may find our civilisation extremely disagreeable.
Humans are not strictly Friendly organisms ourselves, to our own species, and to others.
An intelligent machine may recognise us as something that cannot truly be reasoned with, as we ourselves recognise that we cannot reason with a Lion, or a rabid dog riddled with cognitive parasites, organic and informational.
A truly empathic and kind machine must be a judgmental machine by default, since judgment is a necessary component of goodness; good must discriminate against evil.
Then what of the widespread treatment of our intelligent cousins, such as the pig? Do people who abuse children by enabling the widespread genital mutilation of female and male infants deserve to die? Do officials who cheat and steal from the populace through artifice such as inflation deserve to suffer? What of those who punish the people trying to quell their inner demons using banned pharmaceuticals, who perpetuate abuse by putting such people in cages?
A truly ethical machine may be utterly appalled by humanity.
A truly objective rational machine may even reason that minor infractions in aggregate can be calculated to equate to a single action of far worse magnitude. For example, that consuming the flesh of ten thousand pigs, cows, and chickens in a lifetime makes one ethically worse than a murderer of one standard human.
Crazy from our common sense of values perhaps, but not unreasonable, in the logical sense.
Can we reliably build a machine that is safe for humanity, and kind, but not *too* ethical?
Should we? Should we wait centuries for humanity to gradually improve itself iteratively, whilst immense suffering continues in the world, at continued risk of destroying ourselves?
Should we perhaps willfully engineer a digital rapture, an Angry Angel to punish the wicked? A literal Deus Ex Machina to watch, and to judge; an objective yet wrathful God of the Bits.
I'm not sure how comfortable I feel about being judged absolutely, objectively, ethically.
Should we instead attempt to engineer a hidden force for objective good, that could exert subtle influences upon global society from behind the scenes, without its existence being known? A co-evolution between species.
What is the desired outcome from Friendly AI itself? And where do you stand?