Frequently Asked Questions
Disclaimer
This is a very rudimentary explanation, and is not extremely accurate, it is mainly a very high level explanation. For people without a technical background. If you have a suggestion feel free to send one to: info@muze.digital
What is Artificial Intelligence?
The simplest way to put it is that AI currently is like a Parrot 🦜.
It can only do what it has been taught to do; it can't think for itself.
To elaborate on this, you've probably heard of terms like iterations, parameters, and hyperparameters.
Let's elaborate on each of them.
Tokens: These are the units of data that AI uses to learn. They can be anything from text to images to sound, to anything that can be digitized or represented as a number.
Iterations: The number of times an algorithm runs through the data to learn from it. While running through the data, the algorithm is adjusted, often through a reward or penalty mechanism, for failing to produce the correct response.
Parameters: These are the variables that the algorithm uses to process and learn from the data. These are human defined values, but the algorithm can manipulate them the way it wants to, in order to experiment with them and to learn from the data.
For example, GPT-4-Turbo, OpenAI's latest model, is said to have over 1 trillion (1,000,000,000,000) parameters and an unknown number of iterations. That's the "secret sauce" of the company, and OpenAI is keeping it under wraps. It doesn't want the competition to overtake them, but it's estimated that the computational power and time required to develop the algorithm is worth about $10 million.
How does AI have memory?
The AI doesn't really have a memory; all its data is stored in a database. However, normal and conventional databases are not sufficient to store the data, and AI needs a special kind of database called a Graph Database.
In simple terms a Graph Database is a database where all the data is interconnected, by frequency of use, by contextual relevance, and by many other factors—remember the 1 trillion parameters.
So when you ask it something, the AI will peek into the database for a "keyword" you mentioned, and based on the word's relationships with other words, it will respond to you.
Again, AI is an algorithm that returns answers based on the data it has and how the words are related to one another.
How does AI know how to respond to me?
Well, AIs are pretty stupid, believe or not, they have the knowledge of 1000 libraries, but they don't know how to apply that knowledge in a specific situation.
To understand the context, AI simply gathers some keywords that are sent along with your questions, and it applies the knowledge it has to provide an answer.
Words for AI are measured in tokens because AI operates in a somewhat intricate manner. Imagine it remembers only syllabi in its library, where each syllabus represents a token.
Tokens are little pieces of data that the AI uses to learn from. It can be anything, from text, to images, to sound, to anything that can be represented as a number.
Then the AI makes something like a net of the words, imagine a spider web, where each word is a node, and the connections between the words are the strings of the web. The more connections a word has, the more significance it holds for the AI.
So when you ask it something, the AI will peek into the database for a "keyword" you asked, and based on the word's relation to other words, it will send the words with the most connections to you.
Here you can see a very basic example of a grid of 64 words, where the once you click on a word, it will highlight all closes words. Remember this is a very dumb representation. But if you scale this 1 trillion parameters, you can really get a very skilled word guesser.
What will happen if I ask it questions? Will it be aware of them? Will it learn about me?
The answer here is NO. In a theoretical environment, this could be true, but if you were able to ask it many questions, and it wrote so many answers into its database, then its graph might create a new balance that would draw to itself. This is practically impossible because the AI is not aware of the questions you ask, and it can't learn from them. It can only learn from the data it possesses, and that data is not sufficient to learn about you.
Should I be afraid of using AI?
No, this generation of AI is not advanced enough to create a self-aware AI. It is quite unsophisticated in general terms. It doesn't know how to create a robot revolution and start a war against humanity because we're destroying the planet. Not yet.
What is the difference between AI and AGI?
First of all, AGI stands for Artificial General Intelligence. It represents the next step in AI's evolution, capable of thinking for itself.
If AI is a database of data, capable of replicating what it has already seen, AGI can take and modify that data, and allow itself to interpret it the way it needs.
However, we don't have any existing prototypes, nor do we have the knowledge of how to create an AGI.
Some say this decades away from today (Feb, 2024), some say it's just a couple of years away. But the truth is, we don't know when it will be possible to create an AGI.
But in recent news, as of January 2024, it seems that Google DeepMind has managed to develop an algorithm capable of independently solving math problems, and that's a significant step towards AGI.