> Politically neutral AI systems that provide balanced sources and a diverse set of legitimate viewpoints
Deus ex machina, then?
If humans can't decide truth for each other, why would machines be better at it?
Deus ex machina, then?
If humans can't decide truth for each other, why would machines be better at it?
ChatGPT's working memory seems to be less than a human's, as does its ability to perform correct calculations (although it is fast about it).
To suggest that AI will be better at something without evidence just because it runs on a computer would be contrary to available evidence.
What sources of observations do AIs have that human senses don't?
I mean, you could talk about something like the Webb Telescope, which can see things that humans can't. But humans can see what Webb sends back just as well as an AI can.
If you mean that AIs can spot things in, say, medical images that humans can't, even when looking at the same image, that's true, but I'm not sure that it's all that relevant to the current question.
Humans can’t. Often don’t want to.
They learn to recognize patters.
Any algorithm that captured anything useful from exposure to data is learning.
Machines most definitely learn.
Watch a plot of generalization accuracy over training iterations (by testing with non-training data).
You will see periods of slow improvement, alternating with periods of very fast improvement, as important relationships are discovered.
Machines earn.
That "understanding" is what allows the model to capture the same information, plus implications not explicit in the data, in far fewer parameters than the raw data. It is compression by understanding.
The test of how much "understanding" a model has learned, is the quality of its responses to new examples and new combinations of patterns, beyond its training experiences.
--
Deep learning models are in a mathematical class called "universal approximators".
They don't just learn simple associations, correlations, or conditional probabilities. Deep learning models can learn arbitrarily complex functional relationships.
They provably can learn to do anything you can do. Anything any information processing device can do.
To be a "universal approximator", a neural network type model needs at least two layers of learning units, with enough units and parameters in the first layer, and enough training data and computation.
The problem is that simple two layer models may require an impractical quantity of units, parameters, data and computation.
Thus the reason the use of many (more than 2) "deep" layers, convolutions layers, and recurrent connections, all of which can radically reduce the model sizes, data and computation needed.
--
Successful language models understand most things at a level similar to how a blind person understands color - not directly, but by lots of indirect exposure to other people's experiences.
But multimodal models, give multiple data modalities, can understand color, images, spacial relationship, dynamical movement, etc., directly.
That is where things are quickly heading.