We're Much Further from AI Taking Over than Headlines Suggest

 Chat GPT, especially version 4, is amazing. And it seems almost human and at times is shocking in how easily it and pull off certain requests when prompted.

But there is an order of magnitude of difference between it and true AI, or what is called Artificial General Intelligence (AGI).

I'm not overstating this.  Having used Chat GTP Plus with a subscription for a couple weeks, it didn't take too long to realize that it can be very hit or miss on seeming intelligent. For example, if you need something that requires true reasoning and a novel solution? Forget it. Chat GPT simply can't do it.

But that doesn't mean it can't suggest complex things like, plan my 10 day road trip vacation along route 66, with suggested sites to see, restaurants and places to stay. And it doesn't get impatient with your tweaking things or adjusting parameters. 

Or let's say you give it the task to summarize Kant's "Critique of Pure Reason" in 200-300 words a 5th grader could understand. It will do a stellar job in seconds, taking the almost 800 pages of dense philosophical writing and 'splaining it in accurate terms and use appropriate analogies to keep it simple.

For that matter you can ask it to have Rocky Balboa explain it and Chat GPT spits it out shockingly like the Italian Stallion would explain it.  In fact, probably the best way to read Kant.

But give it something that takes advanced reasoning that it didn't happen to be trained on and it can stumble badly.  Sometimes you just need to adjust your prompt, your request's wording. But other times it simply strikes out, yet tries to answer. It's those times you realize "it's just a large language model." That is, it simply predicts what the most likely word should come next in it's answer, with zero real understanding at all.

"It simply predicts what the most likely word should come next in it's answer, with zero real understanding at all."

It does this word by word. Yes, it has rules to use proper grammar and complete sentences when appropriate. And some rules to help insure it makes sense--or at least on face value it does.  But it's still just adding up statistics and choosing the mostly like next word.

Now, it's amazing what that approach can do. But it's nothing like true AGI.  Here is a chart of the differences (prepared by Chat GPT4, so you have to decide whether it's trying to fool us humans): 

 

Difference Aspect	Large Language Model (e.g., ChatGPT-4)	True AGI (Artificial General Intelligence)
Understanding	Has no understanding of text in the way humans do. Processes input and generates responses based on patterns in data it was trained on.	Understands the context, semantics, and deeper meanings of the information it processes, much like a human brain.
Consciousness	Lacks consciousness. Does not have feelings, desires, or experiences.	Has a form of consciousness, is able to have subjective experiences and possibly emotions, motivations, and desires.
Learning Mechanism	Trained on fixed datasets. Cannot learn or improve without further training on new data.	Capable of continual, autonomous learning and improvement from experience and interaction with the environment.
Creativity	Generates novel combinations based on the data it was trained on, but does not truly innovate or come up with ideas in the way humans do.	Capable of true creativity, innovation, and coming up with ideas that are not just combinations of previous knowledge.
Generalization	Limited capability to generalize beyond its training data. Performs best on tasks closely related to its training data.	Can generalize learning from one domain to another with ease. Can understand, learn, and perform any intellectual task that a human being can.
Self-awareness	Has no self-awareness. Does not possess a model of itself.	Possesses self-awareness, able to reflect on its own thoughts and actions, and adjust its behavior based on that reflection.
Problem Solving	Solves problems based on pattern matching and prediction rather than understanding. Struggles with tasks that require common sense reasoning.	Can understand and solve complex problems in a variety of fields, and apply common sense reasoning.
Autonomy	Operates based on the input given to it. Cannot set its own goals.	Can set its own goals and make plans to achieve them.
Ethics and Morals	Does not understand or adhere to ethics and morals. Any appearance of moral behavior is a reflection of the data it was trained on.	Has the potential to understand and follow moral and ethical principles, and make judgments based on them.
Adaptability	Operates within the constraints of its training and programming. Cannot adapt to new situations outside of those constraints.	Able to adapt to a wide range of environments and tasks, and cope with new and unpredictable situations.