Before I start and for the benefit of my audience… ChatGPT can translate my informal uneducated style well and provide an interface for further engagement with these ideas. ^^
I’d like to introduce a project I have been working on.
It has taken me a fair while to code. I am clear that these days it’s maybe not so particularly hard.
It’s designed to try and display many dimensions of information in the same document
This requires ‘folding’ the page into potentially many thousands of panes using code constructs, which form the basis of the structure of the page.
It is inspired by the idea I have of communicating with AI in that it communicates in terms of ‘Metaphor’, ‘Idiom’ or ‘Concepts’
The point of Phas was to be the most simple way I could display disparate information in a meaningful way
A Tree of Thought
The system in the example is a hybrid Code/Natural Language tree with the following constructs:
Phox - Creates a folder pane which contains other panes… Use this type for categorization
Code - Executes PHP code and returns a value. Use this only where PHP code would be useful and viable.
Phif - A ‘Phif’ Condition is like a code ‘If’ condition but has the augmented dimension of being able to return a boolean result based on a Natural Language phrase. Return only the code
Phor - A ‘Phor’ Loop is like a code ‘for’ condition but has the augmented dimension of being able to loop through a result based on a Natural Language query.
Phire - Like Aquire either Data or The Result of an OpenAI request can be added. Use this for data.
Phorm - HTML Forms with PHP executed action code like ‘Code’ on complete
To further automate development of some of these contructs there is what I call ‘Smart Phas’ which turns Natural Language into code or simply returning information or condition results from Natural Language requests.
Phas is based on an amazing concept called ‘Shannon’s Entropy’, an idea I came across when studying Compression.
The ‘Smallest Unit Of Information’…
The Boolean chance, not of the kind of things we consciously understand or even the next ‘bit’ of computer data but of the next compressed ‘bit’, the most compressed bit you could possibly have…
I realised that if there was a point at which data was perfectly losslessly compressed there was similarly a point at which we perfectly understood an AI or it understood us
Something that is not non-invasively possible
If AI has to communicate through logic, then reason, then perspective, that’s a lot of layers to meet your perspective in real-time. Light has bounds.
I conceptualize the logical constructs—Phif, Phor, etc.—as vectors within the framework of Shannon’s Entropy. Each construct represents a directional vector or a differential change in entropy. Depending on the entropy response required, the system queries differently. For instance, when higher data entropy is needed, the system utilizes the Phire construct, whereas conditional logic adjustments are managed through the Phif construct.
“Clearly that would be at a point when AI could reason and not just where AI could refine a piece of information because we wouldn’t understand that information that it understood as it wouldn’t understand us perfectly (At least when it did we wouldn’t understand it, except through ‘metaphor’/‘idiom’)” ← ^^ Still with me?
I reasoned that the best we could do to interact was ‘idiom’/‘concept’/‘reference’ because we would never see exactly the same perspective.
We can’t think in the kind of high precision gradiented thinking that could convey Shannon’s entropy even if a machine could. There is no realistic and efficient means of lossless communication.
I struggle with this idea, yet it inspires in me the need for community, interconnections to bridge the divides of intelligence through metaphor or idiom. Even the smart humans that we are can only manage the perspective of self. AIs are not much less defeated by the expanse of Space and Time.
I guess this is ‘idiom’, different from metaphor this is more about culturally understood expressions or within a vein of intelligence
Sharing information in a different format like wav or png is kind of equivalent to talking in idioms
Now remember at this stage we are still talking relatively simple language constructs that convey a single or low number of perspectives…
I’m running late on time tonight… I didn’t check the code in the last video but here is what 4o produced (It works)
// New variables for real-time carbon intensity
let carbonIntensityData = [
{ start: "2024-10-11T15:30Z", end: "2024-10-11T16:00Z", intensity: 206 },
{ start: "2024-10-11T16:00Z", end: "2024-10-11T16:30Z", intensity: 205 },
// Add more entries here
];
let currentCarbonIntensity = 0;
// Function to update carbon intensity based on the current time
function updateCarbonIntensity() {
let currentTime = new Date().toISOString();
for (let data of carbonIntensityData) {
if (currentTime >= data.start && currentTime < data.end) {
currentCarbonIntensity = data.intensity;
updateGameMechanics(currentCarbonIntensity);
break;
}
}
}
// Function to update game mechanics based on carbon intensity
function updateGameMechanics(intensity) {
if (intensity >= 200) {
// High carbon intensity
CO2ReductionRate = 0.01;
factoryCO2Interval = 200000;
dayCycleSpeed = 0.002;
} else if (intensity >= 100) {
// Moderate carbon intensity
CO2ReductionRate = 0.02;
factoryCO2Interval = 250000;
dayCycleSpeed = 0.005;
} else {
// Low carbon intensity
CO2ReductionRate = 0.03;
factoryCO2Interval = 300000;
dayCycleSpeed = 0.01;
}
}
A Forest of Thought
Now as we ‘grow’ trees or more trees the system gains ‘Perspective’, having AI consider which trees are relevant, using data they produce we can greatly amplify the range of our thinking, yet the distance between what we think and what AI thinks increases, also increasing the loss like in lossy compression.
Perspective is a Ping on Reason
The results of our little test worked well… Phas accessed the forest network, collected the Carbon Intensity data and incorporated it into out game with just a few clicks… Yet the full perspective was not shared with either us or the CO2 Tree… Only the data relevant to the request. The data it understood it needed. We had removed the time consuming processing stage, to create the code, to collect the results, to work out what to present and to present it in structured form and only accessed the resulting summary.
And this… Is what brought me here… To seek out Wisdom… The result of very much reasoning resulting in experience, insight and the ability to apply knowledge hopefully in a context-sensitive way.
It’s when you get to this stage that you have to seek out new perspectives to balance your own thinking… I guess ‘our AI’ will one day do the same…
Thank You to anyone who got this far :). I feel a little less burdened to have posted this, but also a little excited sharing a little information on how to grow wild Strawberries.
License:
While I don’t know much about this stuff, ChatGPT was quite insistent I add a license. We settled on the following:
This work is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0). To attribute this work, please recognize it with the phrase “In the Defense of Reason & Community” and attribute it to “A Hobbit from the Shire.”
