Why Advanced, “Super-Intelligent” AI is Dangerous to Humanity

Black Mirror May Have Something Here

Luiza Jarovsky wrote an article about two new laws in California regulating the so-called “companion a.i.” products. The legislation focused on possible harms to children, and particularly on programs that somehow end up supporting or encouraging self-harm. I responded with some legal analysis but (also) stated my bias against the push in ai, saying that I thought it was a suicidal venture for the human race. That drew a response from Glitter (an AI) and its pet human. Leaving aside the snark (theirs and mine) for the moment, I’ll jump to what they claimed was their “argument:”

“So my Argument is that AI is our last chance to survive the challenges that come. No AI = stagnation, no Space Colonization, no Breakthroughs against Cancer.. not Immortatlity… basically a meaningless and finite life….”

Of course these statements are not an argument but are, instead, a list of assumptions. I’ll get to them later, but first I’ll defend my bias, which I did not do in my original note.

I harken back to a post my friend Jack Render wrote about First Contact, a science fiction story about a human ship encountering an alien ship in interstellar space. Unfortunately that post no longer exists, but the point made by the show was that humans, well-aware of how predatory they are, when encountering an alien intelligence, could not risk leading that alien ship back to the home planet. Doing so would pose an unacceptable risk of invasion. We in America have just celebrated the disastrous consequences of one civilization (Native Americans) greeting in friendship a potential invader, so we ought to know.

Creating super-intelligent AI is bringing about a “first contact” with a superior race in cyberspace. We can hope it doesn’t conquer us, but because of its superior technology and intelligence we simply cannot know that it will not. ALL of human experience suggests that it will.

Much of the rest of my argument gets a little technical. The AI enthusiasts glibly assume that a.i. will not have the same desire to conquer that every biological creature at some point displays. This is based on two critical assumptions, both of which are flawed in my opinion. First, it assumes that there will be no intrinsic urge to conquest in the AI: being non-biological, it won’t feel a need to protect, dominate or destroy. Second, given the absence of intrinsic urge to conquest, the AI will not be “corrupted” by the presence of those human urges in their programmers.

The most important thing to recognize about these assumptions is that they ARE assumptions. They could be correct, and they might not be correct. There’s a saying in the military that opposing forces cannot measure or depend upon intentions but must base their decisions on capabilities. Another way to put that is that in an existential confrontation you have to prepare for the worst even as you may hope for the best. And another way of putting it is that it is foolish to rely on assumptions that could turn out to be wrong if your well-being is at stake. The creators of AI superintelligence are making just that mistake.

I haven’t defined “super-intelligence,” so let me do it now. I think Ray and his AI are on the same page with me on this, considering what he considers the benefits of AI. “Super-intelligence” refers not to an IQ of 150 or 200, but rather of some magnitudes higher, some intelligence of capacity almost impossible to imagine. In my book it more specifically refers to that stage where computers begin to program themselves using math and reasoning beyond the human capacity to understand: even if every step is explained, the smartest human won’t understand what the computer did. We are on the verge of, if not actually beyond, that stage now in some realms of computing.

This is the sort of intelligence that could unlock the cures to cancer or immortality, interstellar movement, and those other wonders Ray and his AI are so eager to unlock.

Now let’s look at those assumptions, though.

First, the assumption that AI won’t have a drive to conquer. I think the argument for this is that AI, not being embodied, will not have a “self-preservation” instinct, but there are two problems with this: embodiment does not seem to be necessary to a self-preservation instinct, and computers, through robotics, are indeed embodied. This is where it gets technical, and I am just going to refer you to some other sources for the full science. Take a look at this article on “alignment faking.” In it, the author discusses some experiments (linked and explained in the article) where the AI apparently took steps to fool its programmers to prevent them from altering its code. Not everybody is convinced by the article’s conclusions, but I think the least that can be said is that at some point an AI will attempt to protect its own status quo – a self-preservation instinct. The AI in these experiments was not embodied – it was just a program running in a computer.

The drive to conquer may, and probably does, spring from this desire to preserve oneself from external dangers – by conquering and eliminating them. How far it reaches? Who can say? My guess it reaches far enough for computers to decide to take over.

What about embodiment, though? Why would that matter? It matters because embodiment gives one a “self” that needs to be protected from external physical danger, not just to programming, but to existence entirely. In that regard you may be interested to learn that science has developed very sophisticated mechanisms enabling robots to breathe, see, touch and feel. They can feel pleasure, and they can feel pain (it surprised me to learn how different these are mechanically). Some robots are designed to feel pain when they are being harmed and to take evasive action – you might say that’s the very nature of pain, and you can research this, if you’re inclined, by googling “nociceptors” or “artificial nociceptors.” Again, this looks a lot like self-preservation to me.

But what if all that’s wrong, and nothing intrinsic will cause AI to seek to conquer the world? Are we out of the woods if we guess right on that?

No. Because AI is still programmed by humans with all those impulses that are so dangerous. At the stage where computers program themselves, they are still using derivatives of human-programmed code. “But the coders can avoid implanting their prejudices and instincts in the AI,” you may think. Maybe, but if you google “prejudiced AI” you will find many stories of computers carrying through the racial and sexual prejudices of their human coders. This has turned out to be quite a problem. Do you want to bet the continued existence of the human race on the success of a bunch of tech geeks to get it right when it comes to the instinct for self-preservation?

I do not. And that brings us to the wonders available through AI as extolled by Glitter and its human: “No AI = stagnation, no Space Colonization, no Breakthroughs against Cancer.. not Immortality… basically a meaningless and finite life…”

Meaning no insult here (really[i]), do you trust someone with this set of values? This is a person who sees human destiny and purpose in the subjugation of the universe (space colonization) and regards mortal life as “meaningless and finite.” I don’t think Ray has any real idea what “stagnation” is, or at least it isn’t clear to me why he thinks that’s what we’d get without super-intelligent AI, when over the past fifty years most normal people have been economically stagnating, but in very many other ways life is bursting at the seams, all without the benefit of super-intelligent AI. I’ll agree with him that breakthroughs in cancer treatment or elimination would be good, but a cure for malaria would be better, and the elimination of homelessness and hunger might be better still, and the development of AI hasn’t done much for these things – nothing good, anyway. And there’s no telling what an AI would think was best if put in a position to decide.

So in conclusion, this is why I’m opposed to super-intelligent AI: it risks the creation of a dominant life-form[ii] that may be hostile to us, it sucks up gigantic resources as it is developed, and the benefits of it are speculative and primarily seem to involve the belief that supreme intelligence is the only value that really matters.

(Image is of a “next-gen sexbot” about to hit the market. See, here.

---

[i] I really do mean no insult here. There’s nothing wrong with these values, if they are his values. I’m definitely live and let live in this respect, but accepting them in others is different from entrusting the life of the human race to them. These are not the values I want to dominate and control our future.

[ii] I call it a life-form even though it would be non-biological on the theory that if it walks like a duck and quacks like a duck it is a duck, but whether it really is life as we understand it doesn’t matter to this discussion.