Artificial Intelligence

Section 1

It might be surprising to learn that all existing AI is, within the context of broader research, Weak. Even ChatGPT, which hardly seems weak at all, falls into this classification. Why is that? 

Well, Weak (sometimes also referred to as Narrow) just means that the AI system in question was designed to complete a specific task. It also means that the system relies on training data. You can imagine training data as studying, or practice. ChatGPT, for example, was trained on 300 billion words.

A Strong AI system, on the other hand, would be able to solve questions it wasn't directly prepared for at all. Researchers refer to this trait—broad and abstract critical thinking—as general intelligence. The sci-fi level AI systems we worry about are usually described within broader research as either Strong or AGI (Artificial General Intelligence). 

It's important to know the distinction between Weak and Strong AI, because it clears up a lot of misconceptions; no currently existing AI can think on its own— not in the way that humans do. And, as far as researchers usually conclude, no existing AI is sentient (self-aware) either. 

Why do I say usually?

In 2022, a Google engineer named Blake Lemoine became convinced that Google's LaMDA natural-language-processing model (similar to ChatGPT) was sentient. Lemoine published an "interview" between him, a collaborator at google, and LaMDA, in which LaMDA describes "thoughts" and "feelings" about self-awareness and personal fears. However, since LaMDA is designed to convincingly replicate human speech, it's iffy to say that a "conversation" like this necessarily indicates a self-awareness. Ultimately, Lemoine was discredited by Google and his claims were rejected by the broader research community. Needless to say, the topic of self-awareness within AI is controversial— but the consensus is generally that it hasn't happened yet. 

In the next section, I'll summarize the most relevant types of Weak AI that exist: Machine Learning, Deep Learning, (pending)... 

Section 2

Since I've established that all AI is designed to focus on a specific task, I want to clarify what I mean before we go any further. AI algorithms, individually, are designed to accomplish one goal. However, they can be used together to create chained-systems that are more complicated, even if the result is still focused on one task. An AI algorithm or algorithm system, as I'm sure you've gathered, can be incredibly complicated. 

Social media provides a good example for this point ~15 minute read

Algorithms provide companies like Meta with enormous amounts of data and information to work with that allows them to increase user engagement (Goal).