An overview of my AI
Classic Meta-heuristics with creative hacks
If you're reading this, there's an off chance...
...that you will see some value in it.
Blog Post, Personal, AI, Learning, Self-taught, Creativity
January 25th, 2015
The necessary prologue
Idon't consider myself an expert on AI. I am as much as a student as the next fellow curious person, and will ever remain to be. However, I do consider my perspective to be different.
If you've read my résumé , perhaps you already know that I'm a college dropout, life happened and the most sensible choice was to leave my studies in order to fix more urgent matters. I could've come back, I was the recipient of three scholarships, but I didn't. Why?
I personally consider university to be inefficient in teaching me and making me love learning, than doing that on my own; and there were no hints that the higher education system was to be any different elsewhere. So I called it quits.
I believe there is a mind-set that university imposes that is restrictive to creative thought, whether the question if creativity can be taught or if university takes away the fun from learning can be debated; many that know the ropes will acquiesce that many times you will be in front of a book, without ANY drive to learn what's in it. Just to get a good grade, out of obligation.
For me, being self-taught is a part of what I am, and removes the pollution of drone-like state-paid teachers who are only there for a paycheck repeating the exact same words in the textbooks. A dreary routine that sacrifices the human parts of learning in the cold boring and lifeless altar of standardization.
There's so much joy in being self-taught, and one can learn so much faster and at its own pace; with a little self-discipline added to the passion of acquiring new knowledge, one can readily start striding not long after walking; and given some time and momentum, the most accurate feeling is that of flying.
Looking back, and below, a new perspective opens and your world changes forever. There was so much that could be seen that you were not able to perceive before, a myriad of interdisciplinary correlations opening hundreds of possibilities to exploit while developing a wide holistic overview of your knowledge.
Back to the point
Lots of the techniques from which a current AI is developed are built standing on the shoulders of giants: Artificial Neural Networks, Meta-heuristics, Optimization Algorithms and many more; are the results of the brain-work of those who came before us. If we wish to advance, we can't reinvent some of these things, we must learn and build upon them.
Nowadays the hardware is so much more potent than when these techniques were invented that we can allow ourselves to play with the technology and find fancy new applications while expanding creatively the possibilities between them.
I've played a lot with this tech and I've had ample time to test many ideas. Luckily enough I've also had the opportunity to read many papers, thesis and other research publications, saving me countless hours of personal experimentation while acquiring new perspectives.
There are many ways to do research, and even the stupid way of testing combinations can yield results. Edison knew of this when he said that genius is "1% inspiration, 99% perspiration", but he also knew how to exploit real geniuses like Tesla.
Then there's people like the tinkerer Maurice Ward, a former hairdresser turned amateur inventor whose "Starlite" heat resistant material supposedly prevented an egg from being cooked when under a torch.
Maurice died without ever doing anything meaningful with this billion-dollar technology, because nobody wanted to give him the 51% of the business with his formula.
One can question Ward's tinkering methods and greedy motives, but if his creation worked and he could produce more of it on demand (albeit reasonably priced) there is no question that his methods would matter little. Do you need the formula of Coca-Cola to drink it?
Coke has sugar, lime and caffeine. My AI has all sorts of proven meta-heuristic optimization algorithms.
Coke tastes great. My AI has no flavor, but it has results in other areas such as function prediction and creating game entities simple AIs.
Coke has a secret formula. My AI has a main heuristic encompassing a repertoire of my own experience as an optimization freak, including blatant hacks, shortcuts and personal variations of known techniques.
Can my AI withstand a flaming torch? I don't know yet, I'm actively researching and I might fail to achieve anything at all. So far my AI trading code looks promising to me but who knows, perhaps it is not trading where its faculties are best deployed.
If you're not sure that you'll succeed, then why are you doing it?
Well, I love what I do. I build stuff, even if laziness prompts me to build stuff that builds stuff. I would do it even if I had all the money in the world. Or none.