Training vs learning
Machines never learn
Training and learning are quite similar but I think they are the outcome of different processes.
Both cause changes in our structure of understanding that can change our behavior, Training is a process of mastering a skill while learning is more about exploring our world.
In many ways learning depends of training because we need to be trained before we can explore very much.
For instance, I was once trained to drive a car legally and safely. But I learned the backroads of New Brunswick like the back of my hand.
Generative AI systems get trained. I'm not really clear on the process but it involves looking for patterns in vast amounts of data. A prediction is made about what happens next when a pattern is found. The system checks whether the prediction is accurate and adjusts itself so that the predictions get to be more and more accurate.
It's amazing that it works but I've tried it and it does work. These are systems of hardware and software that simulate the neuronal connections in the neocortex of our brains. You start off with a system that doesn't make any sort of prediction and end up with a system that does make accurate predictions.
Systems that can translate text into speech were early systems that worked that way.
The point is that once the system is trained then you can easily replicate it without having to train each new one. The system is just hardware with a neural network that has a set of weightings that were created by the training. But those weights are just numbers within computer code and that code can be run on any compatible machine.
This has a very ominous aspect. The "intelligence" of such a system might grow without limit leaving human intelligence far behind because each human has to be trained from scratch. It takes decades.
Well - babies don't start as blank slates. They respond to the emotions of the people around them. They have mirror neurons so they can imitate. A smile generates a smile. Babies have a propensity to master a language but start off not understanding language at all. But if in their burbling the baby says mama and Mama comes and smiles - that sets a weight in the baby's neuronal network. Baby cries and Mama brings food - that sets another weight. Remarkably, a process like that can make a baby fluent in a language in just a few years.
For each baby !
Humans have capabilities that machines don't. Babies laugh and enjoy playing. Babies, and people, do things for fun. Babies live in a social world of other babies and adults. All of those people are responding to motivations like "having fun" or "making baby smile" or "I'm hungry".
The life of a person is vastly complex. I'm a socially isolated old guy these days and yet I interact pleasantly with dozens of people a day. I read the news (an addiction I admit). I get smiles a lot as I'm out doing errands and I know not to make eye contact with the crazy people on the street. How many terabytes of data does my brain process each day as I walk doing my errands? Others are way more active that way than I am :-)
So, we are told that AI can process a few terabytes of stolen data in a week or so - pikers ?
I learned about the backroads in NB for fun. I bet that by now google maps has explored them too - but not for fun. I learned how to skid around a gravelly corner going yahoo. That was a skill that actually practical years later when I had to climb up a steep road with a few switchbacks during a heavy snow storm.
The great limitation of AI is that it doesn't have fun and so can't be creative or exploratory. AI might be used against humanity by some evil dictator - but AI doesn't want to - it doesn't want anything
What do you think
I present regular philosophy discussions in a virtual reality called Second Life.
I set a topic and people come as avatars and sit around a virtual table to discuss it.
Each week I write a short essay to set the topic.
I show a selection of them here.