Blog‎ > ‎

AI Boom Bust, Social Networking, My passion, Baltic security review

posted Apr 23, 2017, 10:04 AM by Sami Lehtinen   [ updated Apr 23, 2017, 10:05 AM ]
  • This AI Boom Will Also Bust - This article actually pretty much is on same lines as I am. Many organizations got lot of data which isn't being utilized in anyway. Even lot more data could be very easily gathered, if required. Using the state of art machine learning on that data would be just computationally too intensive and wouldn't produce worthy results. Best bang for buck would be using some very simple charts, KPI and linear regression on that data. Aka, basic traditional BI stuff or Data Science if done it bit more advanced form. That would revel where there are inefficiencies and which units / items / personnel etc is doing better than others and help the organization to seek knowledge why so. Thinking that you'll just forward data from your "data lake" to "AI" which would then run your business more efficiently is pretty fantasy, at least so far. How about replacing the investors on shark tank / dragon's den with AIs wouldn't that sound great plan? At least the AI could properly analyze the business background and wouldn't be so "limited" on available resources than the busy investors currently are. Wouldn't the advanced AI automatically clean and merge the data, as well as figure out what really matters in that data. That's the dream of AI. But as it comes clear in this article, it's not the truth yet. So called AI is often just very complex (at times) statistical model, nothing more, currently. That's also why some of the simpler models could produce similar or better results with much less computational resources. This is also one of the primary reasons why I haven't yet applied real "machine learning" on anything. Most commonly I use simple statistical models like Bayesian Filtering or Collaborative Filtering. Even those simple methods can require lot of computation resources on larger data sets. Requiring multiple optimizations to make the actual processing simpler and reducing amount of data being processed using several shortcuts. Of course there are cases where accurate prediction does make all the difference, but as usual. I like balancing resource consumption with required output. Shouldn't smart AI (or Strong AGI) do that completely automatically? These are the things that matter and matter this much, this is how we efficiently allocate resources to produce best results form this data. This data is ignored, this data is slightly analyzed to see if it causes any alarms, and this data is the key data which will be very carefully studied and analyzed using the most advanced methods. You don't need great AGI to make business predictions for the board, it should be able to replace the board. ;) - But as said this is something which will be probably over estimated in short term and under estimated in long term. There are real benefits coming and AI research is progressing. But you can't yet replace full business department with a rack of servers, at least in many business sectors. Something like AlphaGo and Watson are going to change things in future, but those are still limited and aren't running in your pocket. kw: deep learning, Monte Carlo method, inefficiency, inefficient
  • Social Networking: Show Passion, Show Expertise, Signal Flexibility, Ask for Advice -> Good!
  • My passion is to make reliable, efficient software without wasting resources and execute projects which truly benefit the end customer. Without hype or false exceptions. Sometimes taking perspective can be seen as negative thing, I'm taking lot of perspective in this blog. But I think some of the perspectives also just reflect my experience and views, boosting my expertise. And sometimes I'm just plain wrong, but that's just very human. I might have some bad background data or experience, but if something is wrong. I'm glad to review my opinions when new data or perspectives are brought to light / my knowledge. So as said, opinion might be strong A, but in light of certain circumstances it's totally natural to choose option B. Everything in this blog is mostly generalized, so this doesn't reflect "absolute opinion" in no means. - Thanks for feedback.
  • Read dozens of pages of Baltic Sea security review, Finland (Åland), Sweden (Gotland), Denmark (Bornholm), Estonia, Latvia, Lithuania, Russia (Kalningrad, St. Petersbug), etc.