The problem is the machines don’t understand the context of their work which can cause problems. All the work of AI is a result of trying to make a machine that can.
I am deeply confused by this statement.
A robot that assembles cars does not need to “understand” anything about what it’s doing. It just needs to make the same motions with its welding torch over and over again for eternity. And it does that job pretty well.
Further, neural networks as they stand cannot truly understand anything. All classification networks know how to do is point at stuff and say “That’s a car/traffic light/cancer cell”, and all generation networks know how to do is parrot. Any halfway decent teacher will tell you that memorizing and understanding are completely different things.
No but a robot that does the dishes needs to know how to know what a dish is and how to clean all different types and what’s not a dish. The complexity of behavior needed to automate human tasks that cannot be done by a assembly line robot is immense. Most manual labor jobs are still manual labor because they are too full of unknowns and nuances for a simple logic diagram to be of any use. So yes some robots need to understand what’s going on
And as for parroting vs remembering current LLMs are very limited in the capacity of creating new things but they can create novel things bash smashing together their training data. Think about it, that’s all humans are too. A result of our training data. If I took away every single one of your sense since the day you where born and removed your ability to remember anything you wouldn’t be very intelligent either. With no inputs youcould produce no outputs other than gibberish which an AI can do to. ( And I mean ALL senses you have no form of connection with the outside world )
The problem is they didn’t focus research this tech, or try to make image generators specifically, it was an scientific discovery coming from emulating how brains work and then it worked wonders in these fields
Which is why STEM is so cool. Because one is dedicated to an interaction with physical reality, which exists outside the mind, novelty can arise unexpectedly from a simple and honest conversation with deep structures nobody knows about.
STEM is cool because it involves discovery. The fact that amazing things can exist without anyone being (yet) aware of them makes it an open and unpredictable undertaking.
I am deeply confused by this statement.
A robot that assembles cars does not need to “understand” anything about what it’s doing. It just needs to make the same motions with its welding torch over and over again for eternity. And it does that job pretty well.
Further, neural networks as they stand cannot truly understand anything. All classification networks know how to do is point at stuff and say “That’s a car/traffic light/cancer cell”, and all generation networks know how to do is parrot. Any halfway decent teacher will tell you that memorizing and understanding are completely different things.
No but a robot that does the dishes needs to know how to know what a dish is and how to clean all different types and what’s not a dish. The complexity of behavior needed to automate human tasks that cannot be done by a assembly line robot is immense. Most manual labor jobs are still manual labor because they are too full of unknowns and nuances for a simple logic diagram to be of any use. So yes some robots need to understand what’s going on
And as for parroting vs remembering current LLMs are very limited in the capacity of creating new things but they can create novel things bash smashing together their training data. Think about it, that’s all humans are too. A result of our training data. If I took away every single one of your sense since the day you where born and removed your ability to remember anything you wouldn’t be very intelligent either. With no inputs youcould produce no outputs other than gibberish which an AI can do to. ( And I mean ALL senses you have no form of connection with the outside world )
My dish washing robot doesn’t need to know anything. It does depend on me loading it, and putting the more heat affected stuff on the top shelf
The problem is they didn’t focus research this tech, or try to make image generators specifically, it was an scientific discovery coming from emulating how brains work and then it worked wonders in these fields
Which is why STEM is so cool. Because one is dedicated to an interaction with physical reality, which exists outside the mind, novelty can arise unexpectedly from a simple and honest conversation with deep structures nobody knows about.
STEM is cool because it involves discovery. The fact that amazing things can exist without anyone being (yet) aware of them makes it an open and unpredictable undertaking.
Right. That’s why making cars is already automated. But a robot that digs ditches needs to understand context because no two ditches are the same.