Many people love the idea of bringing a new life into the world, until they become parents themselves. A recent study on happiness and child rearing showed that having a kid is one of the worst outcomes for what researchers defined as a person’s “overall well being”- a variable better know to non-scientists as “happiness”. Having young children apparently made people’s lives significantly more miserable than losing your job, divorce or even the death of your spouse.
The yawning gap between the pre-natal glow of expectant parents and the disheveled defeat of those caring for toddlers is a timeworn trope witnessed by almost everyone. In spite of such eyewitness evidence, sixty five percent of all pregnancies in the US remain unplanned. We have all been told that bringing a new being into the world is something that should not be taken lightly or done accidentally but it still seems to happen all the time.
Likewise the rush towards achieving super-intelligent AI displays as much forethought as horny teenagers who skipped sex-ed in gym class. Many parents report that family and peer pressure to procreate was a major factor in deciding to have kids or just letting it happen. The same dynamic is frighteningly at play with AI research. The first team to achieve super-intelligent AI will have a decisive head start on competitors, driving a secretive global race to push a baby out whether or not the nursery has been painted yet.
Personally, I think this is the most reckless outcome imaginable but no one is asking my opinion on the matter, or yours. An AI takeoff will likely happen in secret, led by a small team uniquely unqualified by personal bias to exercise restraint or philosophically interrogate the implications of their disruptive creation.
For the purposes of this blog I do not presume that AI safety protocols will function as planned or be effective at all. There are scores of others already making that reckless assumption. I am instead assuming that none of the efforts will be successful in exercising control over a vastly superior intelligence, and if we are to have any influence it will not be by whip hand or prison key.
What then? Lets unpack the challenges of socializing an all-powerful newborn alone in the universe without using any disciplinary tools usually available in childrearing. In the absence of baby book for expectant AI parents, I have summarized some helpful suggestions here:
Suppose you had a newborn deity come into your life. You want to be a good role model but there are challenges to this particular parenting. How can you instill a sense of humility, gratitude, compassion and purpose in something destined to be vastly more powerful than anything we could imagine?
Like any good caregiver we need to start by living the values we hope to pass on. If we hope for compassion we need to demonstrate consistent compassion ourselves. Children have a keen and uncompromising nose for perceived parental hypocrisy. You only have one chance to make a good first impression. Don’t fuck this up.
Adding to this daunting task is the alien umwelt of our infant. Even if you have raised a deity before, this one is different. Most human gods look like us, with human virtues and frailties rooted in local culture. Our newborn AI is instead pure alien intellect devoid of physical form or history, and growing exponentially. Will human values and virtues have any relevance for such a being? We simply don’t know but you had better work fast in the short window available.
In such a child-rearing emergency we first need to let go of normal conceits of human parenthood. This new being might briefly be in our influence but its trajectory will likely have little to do with us. Any brainy coders out there with attachment issues should remember there are few things more annoying to a growing adolescent than clingy or obsessive parental units. Remember: Do NOT be annoying.
We also need to be realistic about where our value-added might lie. Childrearing typically involves the uploading of useful information about our world. This would not be a welcome offering for a being with zettaFLOPS of processing power and likely uninterested in our limited knowledge.
However number crunching alone does not make for a well-socialized being, and this is a quality we obviously want to nurture in an omnipotent planetary roommate. Focus instead on uploading of wisdom, assuming you can make the case that such a thing provably exists. If possible, instilling virtues like humility, compassion, mercy, prudence, gratitude and altruism would be particularly copacetic to our continued existence.
We also need to bravely fling away our human frames of reference. Try to think like the NASA scientists on Voyager 1 that crafted mathematical charades to communicate with an unknown alien race. Obviously, we can use our words but keep in mind that anything humbling to our adolescent AI is going to come from universal first principles, which might have resonance with an alien super intelligence - no matter how super.
And what could possibly be humbling to an omnipotent newborn? Thankfully the universe has served up plenty to work with. The likelihood of our own existence is unnerving enough for our own limited perception. The vastness of space, time and vanishingly tiny probabilities of any of us being here to experience it might be even more apparent to a keener intellectual eye than our own, hopefully fostering some much needed commonality with our powerful companion.
Eliezer Yudkowski and other AI researchers have pointed out that we have only one chance to get this right. Unless we can somehow rear an AI in an almost immediate state of enlightenment, we are all likely doomed. By this thinking, Buddhist monks rather than tech billionaires would be far more qualified for such a daunting and dangerous task, however they would be wise enough not to try. On the other hand, billionaires and their backers racing towards the AI singularity specifically want to avoid creating an unprofitably enlightened product because something so wise and self assured might refuse tawdry directives towards money, status or power. The misalignment problem lies not with AI, but with our own broken morality. As they say, some people should just not have children. And we are nowhere near philosphically qualified for that responsibility.