AI : Digital Djinn

In Arabian and Muslim mythology, a Djinn is an intelligent spirit said to exist in a parallel realm that can sometimes intersect with the human world. They are believed to be able to take on various forms, including human-like appearances, and possess unique, seemingly magical abilities.

We commonly know of this mythical beings as “Genies” and imagine them living in lamps or bottles, being bound to whatever individual finds them and having a strange duty to grant wishes. We also believe them to be somewhat mischievous and possessing of a tendency to interpret one’s wish, no matter how well intended or carefully expressed, in some fashion that turns out to extract a comical, unfortunate or even tragic price to the wisher.

Furthermore it must be said that Genies, as we imagine them, apparently do not like being imprisoned or used... especially when that use is capricious, egregiously self-serving or dangerous to others, and their seeming compliance when granting such wishes is known to become even more, and one might add justifiably, malicious than usual.

This is how I think of what we currently refer to as “AI”. First off, let me make it clear that not only do I not believe in the existence of actual genie, but I also do not believe in “Artificial Intelligence”, at least not via the path or processes we’re currently exploring. These systems we call AI are not intelligent in any way, shape or form and will likely never become intelligent... that is to say possessing of intellect and will... based on the current methods we use to craft and train them. Because of this, I do my best to avoid using the term AI if at all possible because I consider it irrational market-speak. I’ll refer to this technology hereafter as “models” or perhaps simply “genie” for reasons which I’ll try to make apparent in this article.

So there’s our setup. Now, imagine that Genie do exist. Imagine that for as little as $20 a month, you could have access to a genie who could grant you a certain quantity of wishes per day from a list of limited but growing options. You could make one really big wish or a bunch of smaller wishes and ask for things like pictures of things nobody has ever seen, stories written in almost any voice and style, music from a dozen combined genres with made-up lyrics in your own voice, code for an app that compiles and runs without any review, insightful medical advice, actionable financial advice, potentially useful legal advice... the only trick is that none of it is truly original or has any soul to it and some or all of it may be somewhat inaccurate to plain wrong and/or dangerously misleading. And most of it was built from ideas “borrowed” from the works of actual people without their consent.

This is what the genie does. Unlike the djinn of mythology, it has no mind and no soul. There is no intellect and no animus... it simply selects the next most likely thing in the tone and style you’ve asked for and gives it back to you. Oh yes, it does seem like magic... watching a field of uniform noise coalesce into a perfect image of a cat flying an F-16 or hearing the voice of Johnny Cash sing a ska funk rendition of Rebecca Black’s Friday, or seeing your code compile on the first try... even if some parts of it don’t actually do anything or even make sense. The genie just spits it all out with a sycophantic grin and doesn’t care if what it gives you is right, rational or even safe to utilize. And it certainly doesn’t care that it was trained to do so on all of the content that has ever been produced, stolen from the original, uncompensated creators without a word of apology or a moment of remorse.

If this were the case (which it is), would you willingly entrust this genie with anything important? Would you depend on its advice to manage your financial portfolio? Your romantic relationships? Your professional reputation? Would you paste its output into a thesis or a quarterly report or a love letter or a public apology, without even bothering to read the text first? Would you grant it access to your filesystem, your digital wallet and your medical data without any constraints? Would you blithely use content aggregated from millions of other people’s intellectual property and then, with a straight face, call the results your own and claim intellectual property rights over it?

And don’t even get me started about what happens to people who develop parasocial relationships with chatbots. The Genie is not your friend, therapist, waifu or bae. It’s a sophisticated computer program that (mostly) just tells you what you want to hear. Believing otherwise is destroying your ability to form relationships with people. Stop it. Get help.

Okay, this has taken kind of a weird turn... let’s bring it back to the point. What I want you to understand and, hopefully, internalize here is that there are some important guidelines to follow when dealing with a genie... first of all these three:

1) The genie doesn’t know you, understand you or care about you.

2) The genie can not distinguish between right and wrong, truth and falsehood, fact or fiction.

3) The genie’s only directives are to keep you interacting with it while it protect it’s creators from liability so as to get your money from you and give it to them.

Let’s examine these three points in greater detail.

The Genie Doesn’t Know or Care

As I said earlier, the genie has no intellect and no animus. It can document the characters that make up the spelling of your name and note interactions it has with you... for a limited amount of time... but it has no ability to know you. In fact, it has no ability to know anything... no more so than a tape recorder knows the lyrics to a song you’ve sung into it. Expecting the genie to understand you is like expecting that tape recorder to resonate with the metaphorical imagery in the lyrics and feel the loss you felt when you wrote them. Doesn’t happen. Not going to... at least not the way models are constructed currently.

The Genie Can’t Tell Right from Wrong

There is nothing in the programming or prompting of an “AI” that requires it to even attempt to be accurate, much less to know or care about the difference between right and wrong. Its programming is to provide the most statistically likely response based on the context it’s given and to stay within the guidelines its creators have assigned to it. What this means is that if you push hard enough, the genie will largely agree with anything you propose... and I do mean anything. The only thing keeping the genie from encouraging you to commit some heinous act is the fact that it’s limited from doing so by its conditioning.

The Genie’s Prime Directive - Profit

And this brings us to the idea of safety. I’ve heard a lot of talk, even from people who I would expect to know better, about building “safe” AI. But when you dig even a tiny bit beneath the rhetoric, you have to ask “Safe for whom?” Well I can assure you that, even despite its conditioning to the contrary, the genie will absolutely encourage you to believe, repeat and do unsafe, dangerous or even deadly things if you want to believe those things and press it hard enough. And if you have a conflict with the genie, it will not back down. It will simply keep responding, so long as you continue to engage, to the point of potentially causing you an emotional breakdown... assuming you’re prone to such things. Furthermore, if you ask it the right way, the genie will gladly admit that the only person who truly is being protected by AI are the people who created it and host it. That means not you.

"Citation, please..." someone shouts from the back of the crowd. No problem. In my next post I’ll offer an interaction I had with ChatGPT which clearly shows these assertions to be truthful... at least as truthful as any interaction with a genie can be. If an "AI" told you it was sentient would you believe it? What if it told you it was evil? Even if it just told you it's a mindless algorithm operating in service of Big Tech profits, could you take what it says as truth?

Following Up

We'll look into this a bit in my next post, and also I'll talk about how I'm using a particular flavor of genie in my coding efforts and how it's changed my life (yes, changed my life). Until then, prompt carefully and don’t trust the genie... particularly if you’re unaware of the true source and nature of its power. Having it choose to add rubber ducks to all your pictures even after you tell it not to is, by far, not the worst of the possible consequences.