You know, if you take the time to read about what AI is and how it works exactly, it’s interesting. Cause there’s not really all that much about it that’s too difficult to understand but there’s also something quite strange about it.

Let’s take the one everyone knows: ChatGPT. The idea behind it is steeped, mostly, in probabilities. Arguably the worst of the math disciplines, what “the model” does is basically predict, based on your input, to its best ability and with high confidence, what the expected answer to your query is.

Let’s say you asked a baby of zero years what a certain color is. It will not even understand that you asked it a question but it will definitely receive the input.

At 1 year old, it might throw its food on the floor.

At some point though, the baby will realize, holy shit, this guy’s asking me a question! I can see what he’s holding! I’m gonna say dada.

At some point even further on, the baby will think, okay I saw this at that weird place with all the other kids and the baby will say, “blooo.” And the parents will rejoice and cookies will be served and the kitchen will fall apart.

What happened in this scenario? The baby has spent years looking at the parents, other people, the world, hearing noises, understanding its body. The baby realizes that when asked a question, an answer is expected. And finally, the baby realizes the value of answering correctly. And so, the baby’s understanding of how it should interact with the world is clearly defined and its ability to grow into a proper member of society is set in motion. The probability that it will conform increases.

Honestly, the human experience, when thought about on that level, is very simple. We are constantly learning and learning until we start specializing, and then our bodies deteriorate, which then breaks apart all the knowledge we’ve had into spots of wisdom and universal truths.

ChatGPT, incredibly, has had a similar trajectory. It’s important to understand that computer scientists don’t actually know what happens at the hardware level. In fact, the robustness of these predictive models is still a mystery to everyone and the ability for ChatGPT to respond in such a detailed manner to so many queries is likely something that even the developers find bizarre. And this is all for a service that only relies on two things: text and vision. You can send ChatGPT a photo or document/photo. I once sent it a medical report. It analyzed it perfectly. And I’ve started using the service much more than Google, as I find the information it spits out to mostly be just enough.

As far as I’m concerned, ChatGPT is an adequate Internet Intelligence. It’s trained on primarily internet content. In fact, it responds to you like the Internet. In bullet points and easy-to-digest snippets and blocks of code. It speaks the language that we speak into the computer. It does NOT behave like a human being (at least if you don’t include GPT 4o). It’s like a living version of every blog post ever made. Because, trained on Internet data, any human being would become… an Internetty person. That’s all they know!

And this is where the importance of the input data and the importance of the context of the model and the importance of the types of outputs are very important for us to understand how these models are going to develop and what the ceiling of what these models can do is. You can train ChatGPT all you like but at the end of the day, it is always going to be a model that is going to speak back at you the same way that any page on the Internet is going to speak back at you because fortunately, or maybe unfortunately, that is the kind of data that it is being trained on.

If you go back to the example of the baby, what sets human beings apart is that we have multiple points of input and an entire planet of data and multiple ways of output. We receive data from our eyes, ears, skin, mouth, and internal organs. And our outputs include speech, emotion, touch, thought, and more actions of the sort. What sets us apart from AI is that we simply have more to work with. We are the perfect possible intelligence for our type of body.

What is the perfect type of intelligence for the computer? Let’s take the example of the iPhone. What are the possible sensory inputs that we can give to an intelligence that runs on the iPhone? We can give it the data that is being seen on the screen. We can give it the data that is being stored in its flash memory, we can give it the data that it can see in memory. We can give it the data that it gets from the microphone. We can give it the data that it gets from being touched and we can give it the data that it gets from being moved. We can also give it the ability to automatically connect and contact the Internet. But the most important thing that we can give it is knowledge of how it works on the inside or, in other words, we can give it the data that corresponds to its internal organs. Maybe the iPhone needs to know how the alarm is set. Maybe the iPhone has to know how a calendar event is created and what the significance of it is, maybe the iPhone needs to know how a post is created on Instagram.

And more important than any of these things, the iPhone has to know what all of these actions and events mean. If I create a calendar event that is going to tell me to go and pick up my mom from the airport, it has to know who that person is and why this is important to me and what are the actions that I might want to be done to make this process easier. The iPhone has to understand how it is being used and why it is being used in the way that it is.

And the outputs? What are the outputs of a phone? The screen. And the speaker. And the vibration motor, maybe, but that’s a less significant output. So the iPhone has to be able to take the inputs that I just mentioned and then do something with them and then it’ll present the results using one of its two outputs. And if you can do this with a reasonable amount of predictability and confidence, while reducing my need to interact with the phone, and showing a reasonable understanding of my request, then we can say that the iPhone now possesses intelligence.

And therein lies the secret sauce of where this technology is going. Multiple inputs and multiple outputs along with an understanding of its internal organs. In my opinion, this is how you can give intelligence to anything. It is to allow the device, or the entity in question, to understand all of its inputs and its outputs, and how it works on the inside, and to bring all of that together in a way that is somewhat self-aware. And as the level of self-awareness improves, the higher your achieved level of intelligence is.

This is what is both exciting and scary about artificial intelligence. And this is what is both exciting and scary about what Apple just announced. Apple Intelligence, when taken to its logical extreme, could mean that you have an iPhone that knows what it is. It’s just trapped in the body of an iPhone, limited by its limited reach, and always at our command.

What is consciousness? How different are we, as organic beings, from the devices that we are slowly, through sheer force of will, beginning to understand less… as we begin to develop them further.

Is our intelligence artificial? What separates a human’s mistake from ChatGPT’s mistake?

Eid Mubarak,

⁃ Ahmed