Friendly Superintelligent AI: All You Need Is Love

There are a number of ways in which an “intelligence explosion” like this might happen (Good 1965). Perhaps the most plausible would involve a seed AI undergoing recursive self-improvement (Yampolskiy 2016, Chap. 5).

Given that superintelligence is defined as cognitive performance vastly beyond human-level, this is hard to dispute.

Some reviewers have questioned whether AI will be capable of emotion. I see no grounds for skepticism, however. Artificial emotions have long been a theme in AI research (Sloman and Croucher 1981; Picard 1997; Scheutz 2014).

“Perhaps it is no wonder that love has puzzled so many for so long. Part of the confusion is that the word ‘love’ has been affixed to different parts of this larger, dynamic love system” (Fredrickson 2016, 848).

The third theory, love as valuing (Singer 2009), likely concurs as well. Though explaining how would take us too far afield.

Of course, in order for it to do this, it must have sensible views about what our well-being consists in. A religious zealot might sincerely think that she advances my well-being by forcibly converting me. It seems pretty clear, however, that the zealot’s beliefs about well-being are false. (Were it not for her belief in an afterlife, she would probably reject them herself.) There are currently vibrant research programs in psychology and philosophy, which have revealed a good deal about the nature of well-being (for surveys see Snyder and Lopez 2009; Fletcher 2016). A superintelligence, with its superior epistemic position, can be expected to have even better-informed views about well-being.

It has been suggested to me that love egalitarianism might be operationally equivalent to an interest-based consequentialism. If loving someone means being invested in her well-being, then loving equally should require weighing and advancing individual interests equally. Something like this is probably right. But it also seems clear that love comes with deontological constraints. If I killed a healthy person in order to harvest his organs and save five other people, I could not plausibly insist that I nevertheless loved him. This shows why we couldn’t simply instruct the AI to promote human well-being. “Promote” is vague. Does it mean: maximize the sum total? maximize the minimum individual level? maximize the maximum level? satisfice to some threshold?… Love helps to resolve this problem. Loving is not a maximizing procedure, and (as I suggested) comes with side constraints. Though we can’t articulate the procedure for making loving decisions, we clearly follow some such procedure in our daily lives. And we seem to think it’s the right way to do things.

On Velleman’s view, respect for others is the minimum of moral expectation, while love is the maximum of moral supererogation. “[R]espect is a mode of valuation that the very capacity for valuation must pay to instances of itself. My view is that love is a mode of valuation that this capacity may also pay to instances of itself. I regard respect and love as the required minimum and optional maximum responses to one and the same value” (Velleman 1999, 366).

Of course, people sometimes also disagree about what’s involved in loving someone. But, this is not typically disagreement about what love is. As the research surveyed in Sect. 3.3 showed, there is a remarkable degree of consensus on that question (despite appearances). One might object, in a similar spirit, that there is disagreement about what well-being consists in. Since loving involves an investment in well-being, if well-being is as controversial as morality, then my view has the same problem as MM. I deny, however, the antecedent of this conditional. While there certainly are competing theories of well-being, for the most part, there isn’t much disagreement in the literature over what contributes to a person’s well-being (see Fletcher 2016). People tend to agree on which things are good for a person, even if they disagree about why those things are good for her. (E.g., is accomplishment intrinsically good for a person, or only insofar as it contributes to his positive mental states?) When it comes to the practical matter of promoting well-being, however, such disputes may not be of much significance.