I’m reading about Mandelbrot right now, which is making me think about a lot about high frequency feedback loops right now. They abound in nature, and no one really wanted to look at them until Mandelbrot came along. He is this kind of anti-clean, anti-ascetic temperment personality which i really enjoy, the more i get to read about his personality, the more i enjoy him.
Reading about Mandelbrot’s ideas, combined with Google’s (and others – download.com for instance) predictive search ability made me think that a feedback loop will be possible, that may eventually lead to an important form of AI. If nothing else, it will generate and interesting conversation between only computers, with no human intervention involved whatsoever. With this algo, in theory can come with ideas. All that is required is the iteration between modules. (perhaps explaining seperation of modules in the human brain).
It seems one of the many properties, humans do have which computers do not, is this sort of non-algorithmic reasoning, that comes partially from a dialectic.
For myself I believe strongly that computers will achieve some form of AI eventually, though people will want to gauge this in a typically monocultural-monospecist way.. Can i talk to this computer at the water cooler??? … no probably not, but i do believe that at least a pig-level intelligence is coming very soon. (don’t laugh).
Both pigs and computers communicate in a very different language than humans, which i think creates a level of uniformity and objectivity.. We often view people who speak a different language in this same sort of light. We wrote in our history textbooks that Columbus discovered America.
In anycase, there is some sort of reasonable comparison between pigs and computers to be made, and pigs are considered one of the most intelligent animals, so that would be quite an impressive feat already.
How I believe it will happens is through the conversation of three modules.
search, predictive text and model building (the most difficult one).
1. picker module chooses a random word from a dictionary, performs a search.
2. a version of google brings up likely results based on vector matching (what google is today already), where car and auto still mean the same thing.
3. those results are fed from the search engine module into the picker module again. The picker module follows a power law distribution in its deviation from the main search results. usually picking the top one, but not always.
4. that result and the preceding (say 5) results are logged, and piped to the modelling module, which attempts to assemble a semantic relationship between these ‘random/monte carlo-ed’ results. A semantic model is stored away and built.
Whether we can do the modelling module (create a theory of mind framework), is debatable, but certainly the first two modules can already be built as an algorithm, and will always generate interesting new and novel semantic combinations. Those semantic combinations form the raw ingredients for creating theory of mind with abstractions of the relationships between identical phenomena.
Despite the book “the emperor’s new mind’, by Roger Penrose, (whom i hold in very high respect btw) who disregards true human thinking as fundamentally unapproachable by computers, by their very nature. I think this path of randomness matched against database-like answers posed by the randomess, in a high iteration feedback loop, will create dialogue that will appear to be human.
Watch out for flying pigs. peace.