my ai ideas
im 42 now.
when i was 11/12 the 1st home computers were making a real entry.
i wanted a vic20, as it was this or a spectrum, and thevic20 i suppose was less of
a toy in my eyes, with a proper keyboard, and it didnt
suffer from the awful character sized colour blocks that
were initially so ugly. I suppose, im justifying it after
the fact, but really, one of my friend had one, and thisis where
i saw programming could be fun, but all computers seemed almost
magical then, typing load and hearing the 1s and 0s
scream in to the computer and build a picture with so much
data from these inhumanly fast tones. And a game to play, wow.
i found i liked games, but liked writing my own more. Adventure games,
platform games(my own manic miner clone for the vic20 because i loved it and
it wasnt available for it), with moving floors, disappearing tiles, deadly platforms,
nd a pretty awful little man who jumped, changed direction
in the ait(something you will all understand if you have written this sort of thing
as needing constraining to the ground). His walking was moon-walkesque,
but hey, that was state of the art back then.
I moved on through an amstrad, atari ST etc. I had an acoustic coupler and playd
around on what came before the net, Packet Switch Stream, and PADs.
I knew most people on the ‘net’ by name, or recognised them if not. there were not a lot.
Noone other than nerds were there for a long time.
I hacked into the largest computer company at the time at 12. it wassnt at all hard. infact, it wasnt hacking then. The bigger you were,
the easier it was and the more archaic your setup.
They had the call traced and phoned up. My dad answered and told them he was a solicitor and i was 12 and they couldnt do much about it, which was probably true at the time, but orimised to stop me, and they went away.
Now they sell security and almost try to be seurity.
But, they were hard compared to say the millitary,
They didnt even know you had visited.luckily for them i was only interested in
myself and my abilities and not them.
i became interested in the brain and decided it was a collection of cells that i could simulate and so reate. I came up with the idea of a neural network. It would take an input, multiply it by a weight and add it to the others. It would need a function to keep it within range, something that curved off to infinity 1 1at each extreme.
i had barely heard of artificial intelligence back then. It didnt interest me. i didnt wasnt to create an artificial version, i wanted to actually make something like i just described. Why fake it with rules and things? this is what i imagined.
i went on coming up with lots of ideas to make it learn and then when i got to uni, i did something that i regrett. I found a book on neural networks and read it.
i couldnt remember any of my ideas, just the constant bit that was so familiar. id got myself constrained in a way.
it was a simple book, but explained the adeline,perceptron,backprop,kohenen map, hopfield, hebbian learning, biases, and that was about it.
i wrote a few demos, and avoided literature again, except to use to check out ideas id already had….i wanted to stay original and not get pulled onto a path i would rather stay off.
neural nets werent great then, they were awkward, needed pre-processing, messing with variables more etc.
i got interested in doing little experiments, playing around with the basics. data prep was undoubtedly where you eded to concentrate.
Supervised except minimal;ly was not my interest. Continuous, simple, but self scaling nets were what i thought about.
bias, normalization,reprocessing,and ouput choice and also new neurons, exotic, backwards, on its head, inside out , i tried it.
decay too. decay is exciting to me. it has sent me down a long roaad of finding that in some ways i can mess it all up and something still works and often very well.
i dont like losing information, and that is whyi resist, but decay is
necessary, and used carefully not damaging unless required , in some feedforward setups
i like ai because ideas almost verify themselves as you start to seemedthem explaining several aspects of thought.
with relative levels a prerequisite and on fromm here in different direction being ips. oops
vision-my ideas have been based on differnce/change, and zooming as i call it, which is CNN ish essentially.
i really thought i had got much further. a SIFT like feature finding system, maybe, although it varies, and i believe cant go much further atthe low level.
dealing with scale,
feedforward and everything is an input methodology.
oh, fast advanced nets, that have few constraints, built
anyway, its all about the surrounding sim type for me now. it can be simple, but must use bits of being human to really boost it#
temporal boredom nets…that perceibve time as longerr as they understand signals…short signals seem otherwise
#bootstrapping all over
reinforcement is intrinsic
concept concrete, building net around smart ideaseasy keep fast keep slim
easy sparse solutions by not doing sparse in 1st place so much as a better than sparse sparse…this sounds impossible, but is just
about pushing for data in a better way….not sparse but sparse like
human brains and lesss is more
and overkill as unnecesssary…not size braibn, just comes with body..its the setupand why we boost ours