Do all the things like ++ or -- rants, post your own rants, comment on others' rants and build your customized dev avatarSign Up
wowotek6125199dHow about emergence?
Voxera7201199dThe problem here is that we have not identified exactly what is thinking.
A large neural network that partly feeds its own output to its input will change continuously even without other input, and all new input will be affected by the current output.
With enough billions of inputs and outputs and possibly some delay here and there and maybe some random switches and you might reach something that could create new ideas.
We are still far from creating such large networks and even if we do, we will most likely not let it just run for its own sake, but rather try to focus it on a problem of our choice.
That can make it even more unlikely to happen, and it might not be enough.
Kasonnara93199dI find your justification a little narrow-minded. Yes, the current AIs are as you describe them, but we're talking about the current AIs, it's like comparing the first computer, with a modern smartphone.
What happens if you increase its connections, inputs, outputs and add pre-formatted basic state like a human being? I certainly wouldn't speculate on what AIs will be like in 10, 50, 100 years.
I am not at all pessimistic to the point of saying that they will destroy the world, but to say that they will never be able to carry out the tasks you have presented seems to me very improbable.
Kasonnara93199dYes like a human, if it had bad experiences with some topics and better experiences with others and its learning algorithm take than into account different IA can have different preferences.
I remember an experience at university where an AI had to go through a complicated labyrinth with two paths of almost equal length and a little randomness in the process. This was not the objective of the experiment, but the AIs created learned one or the other in an arbitrary way.
Nikhil116199dYeah , now even carpenter talks about AI. Its that famous😂
bigus-dickus8128199dAI is all good as long as I have my sex robots.
RememberMe12255198d@ddev @Kasonnara not just that, unsupervised learning models are surprisingly flexible and work without training data - you only need to define goals (and possibly rewards) and mechanisms that the model (or AI here) can use.
Most of the media attention has been on supervised learning, the one with the huge training sets, because it has had spectacular success at solving a variety of problems.