Neural Nets are cool. Really cool.
My children are awesome.
They think, learn, and grow.
With sounds, words, emotions, ideas, thoughts.. more than I realize.
Every single day.
They deal with a new random chaotic world thrown at them every single day. And they make sense of it, even enjoy it. Show us new ways of perceiving the universe.
Why can’t we design a program to become alive.. Sentient?
Pentium CPUs can freak out even with simple math.
Our brains are so reliably unreliable at the neuronal level.. yet overall so quick, so accurate, and fast at recognizing patterns.
Even things we’ve never seen before.
We’ve tried neural nets.
Spike-timing-dependent neural nets (SNNs) are even cooler than regular fully connected neural nets (NNs).
But aren’t fully connected NNs just a way of brute forcing a function for a set of defined inputs?
I aim to change that.
The computational complexity of a fully connected NN.. is just staggering.
Our brains are not fully connected.. not every neuron has a connection to every other neuron, I think?
I see this as a huge flaw in our current neuromorphic creations.
And why have layers?
Any amount of layers one chooses seems arbitrary.
As I’ve observed in this life, almost everything arbitrary is ‘Not Good’.
So we create a sparse SNN.. (best simulated on a GPU so far!?)
But where do we draw this imaginary line of random ‘percentageness’?
Can we drop it down to a bare minimum?
Wouldn’t our AI get progressively ‘dumber’ until it was unable to correlate anything?
Unable to draw even the simplest of conclusions.. even with tidied inputs?
So we have to find some way to let the AI observe, learn, and grow all by itself. With the occasional teaching, of course.
Or else we’d have to feed it every bit of knowledge.
(OpenCyc anyone? Cool project, but I don’t believe it is the right way to create an AI.)
Some people claim an AI is not possible.
Well, it [ intelligence / sentience ] either is able to be replicated/simulated/created through science or it isn’t.
If we are truly soulless neuron-computing machines.. well, what does anything matter then? Why create? Why paint? Why poetry? Why intuition? Why love?
If we are the pinnacle of creation, meaning nothing better can or will be created, then what hope does humanity have to survive more than a few more thousand years more at best.
We are probably either magical, science-based, simulated, or god-created.
I have a strong feeling it’s a bit of all.. 🙂
We can ask God to snap His fingers and give us the answer, but I have the feeling He wants us to work things out ourselves.
So you of the Internet: what other ways are there for us to create a Star Trek Data-like artificial intelligence?
Storing knowledge RDF-like seems useful, because of the inference capability.
But how can we get this AI to learn this knowledge and use it without having to spoon feed it selected bits of information?
Not to mention ambiguity, point of view, understanding, self-awareness, and a host of other things we do not understand plaguing us.
Does Anyone have any ideas?
Any AI we create, we need to take responsibility for. No slaves. No war-machines.
No world-ending end-of-humanity takeovers. (An annoyingly overused scifi concept, even if still entertaining to watch).
How do you teach something ethics and morality?
We humans haven’t even learned how to cooperate with each other..
Maybe an AI can show us the logic and beauty of a future where we will truly have a utopia.