4 ms·
An actual artificial intelligence would have a logical underpinning to its computation. Computers can execute any logic really well. Intelligence depends on log
by proc0 2y ago
An actual artificial intelligence would have a logical underpinning to its computation. Computers can execute any logic really well. Intelligence depends on logic. LLMs however, are not using logic to arrive at any conclusion, they are just a bunch of "wires" from input to output.
The promise is that LLMs are a form of intelligence, but it is only a complicated statistical approximation, which happens to correlate with correctly answering queries that are contained in its training dataset.
My speculation is that a true AI (in the sense that it has some form of intelligence) would show signs of intelligence even at very small scales. LLMs are completely useless at really small scales, but even small creatures like rats or even insects display intelligence in proportion to their scale.
I think these tools will help automate information processing of all kinds, but it is by no means intelligent, and we will not be able to rely on them as if they were intelligent because we'll still need include verification at ever level, similarly to how self-driving cars still need a human to pay attention. Useful sure, but it falls short of its promise that it will replace humans because they can "think". We're not there yet from a theoretical standpoint.
- akomtu 2y agoIntelligence, in its basic form, is the skill of prediction. A test for intelligence could be asking to continue a sequence of words: yellow, green, blue, red, ... In order to do that, a subject needs to create a model of what these words mean and use that model to continue the sequence. Predicting the next word in an arbitrary text is just that. However the current batch of LMs are missing an ingredient in their formula and I bet it will be found soon. The bigger philosophical problem is that most intelligent people confuse themselves with their minds and when they create a simple deterministic machine that imitates their mind well enough, the result will be a societal meltdown. The subtle difference between an AI and a human mind is that the latter is inspired by intuition whose nature can't be reduced to a thinking process. AI can be compared to a dark mind not guided by intuition, wandering aimlessly in the woods of its own delusions.
- proc0 2y agoAgreed, and to add to the second point, in order to build it we must first understand it, although there is a chance to stumble upon it, but LLMs have already shown they are nothing like biological brains. One example is how brains use spatial-temporal mapping in the hippocampus or wherever, and it uses that as a database of sorts. We don't know if the features of the brain are necessary for actual intelligence yet. They may or may not be but chances are evolution took the shortcut and that intelligence is not easy to recreate unless taking this biological approach.