DECISION OPTIONS
  • Home
  • Services
  • Examples
  • People
  • Publications
  • Connect
  • Blog
  • Conferences
  • Home
  • Services
  • Examples
  • People
  • Publications
  • Connect
  • Blog
  • Conferences
Search
Picture
Scientific Sense Podcast 
Picture

Deeper  Learing

6/2/2018

0 Comments

 
A recent article (1) that demonstrates how neural networks could be used to approximate light scattering by nanoparticles is an interesting new direction. We appear to be approaching a regime in which prescriptive analytical solutions and conventional simulation become inferior to deep learning. This is exciting but it also presents a huge downside for the advancement of abstract knowledge. Models that show robust outcomes are welcome but a generation of new scientists, prone to taking to the machine to prove hypotheses, by feeding them small samples of historical data, could dampen theoretical advancement not only in Physics but also in other areas.

This struggle between empiricism and rationalism has been with humans from inception. Did they survive by predicting where the lion is likely to be by using historical data of previous (bad) outcomes or did they rationalize by abstracting the expected behavior of the animal? Did they predict when an animal is likely to attack by using historical data on the timing of previous attacks or did they understand the animal's incentives and available alternatives? Did they migrate incrementally by using predictions, originating from previous short excursions, or did they go boldly where no woman had ever gone? Were our ancestors empiricists or rationalists?

It is difficult to ascertain one way or the other. It appears that empiricism has been a hidden attribute in our psyche for long. Till the advent of computers, rationalism appears to have dominated but since then, empiricism has been on a steep rise. In Physics, they now collect and stream data to find "new particles," without even asking why such observations are important. In medicine, they "high throughput screen" looking for the needle in the haystack, without a clear understanding of the mechanism of action. In economics, they regress data to find insights without asking whether they are insights at all.

There is likely no stopping the trend. As computers get more powerful, empiricism will become ever more dominant. If this is a natural outcome of evolution, then, advanced societies elsewhere (if they do exist) would be asymptomatically approaching pure empiricism for knowledge generation. That could be there Achille's heal as it also means that their knowledge is dependent on the past. A planet full of robots, with no ability to abstract but with an infinite capacity to learn from the past, could be highly inefficient.

Would humans retain inefficient qualities of being a human? It seems unlikely.

(1) http://advances.sciencemag.org/content/4/6/eaar4206.full

Read more: http://www.scientificsense.com/2018/06/deeper-learning.html#ixzz5HJUCR99I 
Under Creative Commons License: Attribution
0 Comments



Leave a Reply.

    Author

    Gill Eapen is the author of Scientific Sense, a global blog that attracts readers from over 150 countries

    Archives

    January 2020
    October 2018
    August 2018
    July 2018
    June 2018
    May 2018

    Categories

    All

    RSS Feed

What Our Clients Are Saying

"“We are excited to work with Decision Options, who has broad experience in healthcare and AI, to effect positive changes to our patients & business in many dimensions.” said Ms. Sonja LaBarbera, CEO of Gaylord. "

“By exposing students to analytics and AI in a customized way, we hope to progress the next generation of leaders faster to the highest echelons of emerging companies. Decision Options has been involved in practical applications of AI in many industries and we are excited to collaborate with them and find ways to use their innovative tools to best prepare tomorrow’s practitioners and leaders,” said Aric Krause, Dean of Rensselaer at Hartford.



Contact Us

    Subscribe Today!

Submit
  • Home
  • Services
  • Examples
  • People
  • Publications
  • Connect
  • Blog
  • Conferences