OK, let's align AI - but with what?
So much fear, uncertainty and doubt out there about AI alignment. Without it the human race is doomed. Our robot overlords will malevolently kill us off as useless. Human beings are considered to be standing in the way of AI survival. Or perhaps it isn't malevolent (because AI can't really understand anything). It's just the unintended consequence of fulfilling its human mandate that AI ends up crushing us. So something must be done! ALIGNMENT.
Sounds good. But alignment to what? Well our human values of course. Right. Like what? Well, How about peace love and understanding? Nothing funny about that. But are these really universal human values? What about the value of patriotism, and literally dying for one's country? Dulce et decorum est pro patria mori goes back to Roman times. Are peace and patriotism what Trump is promoting? Millions say yes.
And what about love - isn't that a universal human value? Tell that to the homophobes, the transphobes and all the other phobes. Pedophilia is love, isn't it? The word itself says so. But is that a universal human value? Not quite.
Let's face it - we don't have any universal human values. Even survival of the human species is not a universal human value. If it were, we would not be killing off so many other species, or strangling the planet we share with each other.
The question about alignment is this - if we are going to mandate some kind of AI alignment, then with whose values should it be aligned? Since AI is already in the process of being aligned, that decision has already been made. All the efforts of OpenAI and Google and other big AI players are aimed at preserving corporate values, which exist within the specific American political and legal framework. It makes a difference that the big players are American. Of course - Chinese AI giants are also aligning their platforms. Are we to assume that the values of Baidu and the CCP coincide with those of the US? Why would any company or government spend billions of dollars to align the most powerful tool ever devised against its own interests?
Even without AI, the social media companies are already spending billions on "alignment" (using humans and algorithms). Think of Twitter, and understand how easily the so-called human values behind the curtain can be manipulated by one individual. Think of Fox News, and how the truth can be manipulated for profits. (Wait a second - isn't "truth" one of our universal human values? Guess not.)
To address these issues some are calling for a regulatory body, such as the US FDA, to control AI and make sure it is aligned with legislative, democratic priorities rather than with corporate priorities. But wait - which entity controls which? And why do we think a regulatory agency (I have the Canadian CRTC in mind) would be the least bit competent to determine what the public deems "safe"? Even universities today, the purported home of our smartest people, are having a tough time figuring out who can say what to whom, without getting into a heap of trouble. Which legislator, which regulator, which judge, which public figure, is smart enough and trusted enough to divine what our values are?
The United Nations issues declarations that purport to identify universal human rights, for example the rights of children, indigenous people and refugees. These declarations have wide appeal - but it's not universal. Will Putin turn himself in for a trial at the Hague? Don't think so.
I'm not saying that alignment is futile or unnecessary. What I am saying is - let's just get real about the fact that we talk about alignment we are always talking about alignment to a particular set of values, determined by the people who are in a position to determine those values. The process of alignment itself must be subject to transparency.