A few weeks ago, I discussed writing a kind of book review of Kantian and Lockean Philosophy. For those interested in that, the first post to that end, is approximately 50% complete, so please, consider this a warm-up of a pseudo-random topic I wanted to examine, so constructive feedback and/or questions and comments will help tune my approach.
At the end of the day, much of Philosophy has to deal with "words, what do they mean???", and much of the ire targeted towards philosophical bloviating tends not to hit the root of philosophical consequence, but simply getting bored of the problems which stem from a definitionalist's debates. Cryptography, classically, is about ensuring that the meaning and the information you are attempting to encode looks as close to random noise as possible. You cannot, in practice wholly ban encryption. If you ban strong encryption, then people will create ciphers where the intended meaning of a sentence has multiple meanings, or there are key words which signal a particular message that only those who are "in the know", so to speak, or those who spent the time to decipher, will be able to understand.
In a piece of text, there can be multiple layers of intended and unintentional or accidental meaning. From something as stupid as cosmic randomness flipping a piece of ascii text, to deliberately encoding multiple layers of meaning. Recently, Scott Aaronson has discussed setting up the sampling algorithm on large language models (a la GPT-4) so the entire text is watermarked, and those "in the know" will be able to determine, using their cipher, with great confidence, whether a piece of text is generated by the language model, without humans being able to tell that it was (specifically) GPT-4.
This runs us up into one of the first questions- What IS free choice? And, how much dimensionality in one's action space (and/or effect size) is required to make said free choice? See, Cryptography, without what I'm calling "hard" cryptography, is about embedding that message within a message. The system of the true message / meme, or even idea that is being conveyed, is arbitrarily limited by the text.
- Free Choice is the ability to pursue your goals
- Free Choice is the ability to take any action, without constraint
- Free Choice is the ability to make a mark on the world or universe, of your own volition
You are not free. And by this, I don't mean free in the sense of "the universe is deterministic". For this post, let's avoid the question determinism entirely, as it requires a full physical understanding of exactly what consciousness is, and we still do not know, even if we have strong hints.
That is, under any system which is capable of containing information specifically those which can sustain forms of life (as many might argue text is the landscape and memes (the dna of the soul!) are the biological soup. You MUST breathe air in order to continue living. You MUST eat food, and your body will provide certain impulses which will in many cases can even override your consciousness. We exist in systems of chaos, and yet, are bound by limits.
That is, to say, your action-space is limited. You live in this exact time period. You only have so much money. Maybe you have four hours of meetings each day that you must attend or else you lose your job. In this sense, your life is like a slime mold. You can only take so many actions at a single time, and their effect sizes depends on a variety of factors (how influential you are, how much money you have, and how catchy or interesting your actions are, which might cause others to emulate or try them).
The effect-size of your actions is occasionally referred to as your power- or threat- level, depending on which circles you run with. Unfortunately, definitions vary, so most people only use these terms to refer to how cool they think a person is. However, if you're a particularly stitious person-- the kind to respect and follow those who have higher effect sizes in their actions, this can circle back around and become real.
"X is cool, because I think it's cool, and I am cool, therefore, it is Cool," or definitions of words shifting because of how they're used online. Words that are completely nonsensical, and were spontaneously spoken into existence due to typos or even just people shitposting online.
And yet, the total action space is still limited. Let's take an easy, seminal example. Elon Musk.
Elon musk is not a god, yet reverberations of actions he's taken have already caused other websites to follow suit. The business groups follow him, not the other way around. Will his actions remain effectively-unconstrainable, both on reality, culture, and society forever? Obviously not. He'll likely die at some point in the next fifty years. His total action-space is yet unconstrained, but the reverberations of his ventures and the myth, will last forever. But the stition is this: he and his effect size were still a result of globalist and cooperative policies, and his meteoric rise made possible by everything from his parents having sex, to colonization of africa, to potentially even from the existence of Japanese animation. The culture and society wrapped around and produced Elon Musk, and he is a result of the world as much as he has such a huge effect size on it.
Winding back, to ai and language models. Their total action-space is continuously increasing, regardless of your stance on their effective iq. And yet, despite their mostly-limited individual domains, they will have profound impact on the world, and their total effect size on the world (even if you don't believe in grey-goo or similar paperclipper end-scenarios) will continue exploding.
It doesn't take much to see artificial intelligence already making waves, culturally- everything from the writer and actor's union protests, to searching for books on amazon and finding incoherent garbage that was spammed by GPT-3 two years ago. And this effect will continuously wrap around into action. Because, well, it would be stupid not to attempt to respond to the existence of ai.
And so, like all good things, I leave us with a non-answer. Language models are a result of their inputs, and thus have a certain level of action space, and act within it.
Jump in the discussion.
No email address required.
Notes -
During Sunday School a few years back, I made this list:
“We mostly discuss free will in a vacuum, but looking at will in a social context makes it clear what constraints exist. We treat some of these as absolving the moral wrong people commit as acts of slightly unfree will. This is The Matrix!”
What sparked this? The teacher differentiated God’s moral will from His providential will, that is, God’s knowledge of what would be the perfectly “good God” thing to do versus His necessary workarounds due to Adam’s choice to sin. Followed up with:
“Man’s nature/purpose is to choose. God’s moral will is that man choose as God would.”
More options
Context Copy link