You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
This is probably the largest deployment of AI to date and the beginning of the massive deployment of AIs across the world. The speed of change is going into hyper drive and it will be interesting to see how AIs compete with each other. I've have to watch this for I built my own theory on financial trading programs and how I would deal with them in the market in guiding my investing. Proprietary, but so far my premises have held up. I think no one knows what will happen when AIs proliferate, any thoughts or comments appreciated. Maybe they will cancel each other out to some degree?
I am not sure I would call an application specific deep learning algorithm AI per se. I think the use of the term AI is partly a marketing ploy, it sells better than calling it Deep Machine Learning Algorithm, or Bayesian Inference Neural Network or whatever describes their underlying technology. To me an intelligent agent needs to perform tasks it was not specifically trained to do, i.e. improvise to some extent.
I think that, just as fond brokers and stock market experts have trouble outsmarting each other, which becomes apparent when you consider that extremely few people can beat a given stock index, "AIs" will run into the same problematic. I think of it as a matter of "the law of big numbers." But economics is really not my field of expertise, still working on that PhD in electronics
I have found that in the market AI programs cancel each other out due to different inputs and structure, just like people don't think or act in unison. I was counting on this when I built my system which in old fashion manual and half feel. I've beat the market by a substantial margin, but am not happy with my system and it's instability.
I have found that in the market AI programs cancel each other out due to different inputs and structure, just like people don't think or act in unison. I was counting on this when I built my system which in old fashion manual and half feel. I've beat the market by a substantial margin, but am not happy with my system and it's instability.
Predicting the future of a stock is much more tricky and unreliable than analyzing the existing fact. What Google does with AI in the Google search is analyzing the known fact, not predicting the unknown future. It should work well if the implementation and environment are right. Just like many people on the Internet, I'm very often amazed by the surprisingly good quality that Google search result delivered.
It's not only the AI software and algorithm but also the huge computing power required for Google to achieve the quality and quantity they need to serve.
AI applications deployed in Google and other companies demand the sheer amount of servers and computing power. Given the fact that Google, Amazon, and Microsoft already have more than one million servers each, the server processors must meet the high performance, low heat, and low power consumption requirement. Currently Intel is the dominant processor provider for data center and cloud computing. This high performance, low heat, and low power consumption requirement will pose a big challenge to Intel and a big opportunity to many new comers.
hist 78, there are people that have done it and have amassed billions, but most of the firms using computer trading have seen their revenues dive. Very few have pulled it off over time, but those that have, have done extremely well. I feel the number doing it over the long term are less than two dozen. I feel the potential in simpler markets is greater. The old axiom rules, garbage in, garbage out.
