Basically, economic decision making is not restricted to mundane finance, the managing of money, but any decision that involves expected utility (some result with some degree of optimality.)

  • Can Machine Learning algorithms make economic decisions as well as or better than humans?

"Like humans" means understanding classes of objects and their interactions, including agents such as other humans.

At a fundamental level, there must be some physical representation of an object, leading to usage of an object, leading to management of resources that the objects constitute.

This may include ability to effectively handle semantic data (NLP) because mcuh of the relevant information is communicated in human languages.

  • $\begingroup$ I think I know what your getting at, but this question needs editing. Economic decision making is not restricted to the field of finance, but involves any decisions with expected utility. Thus, minimax can be applied to a range of problems not involving mundane finance. I've attempted an edit of the question. $\endgroup$
    – DukeZhou
    Jun 12 '19 at 21:43
  • $\begingroup$ ok, i can specify. nexoma has already answered in the direction it want $\endgroup$ Jun 12 '19 at 21:48
  • $\begingroup$ this is a trick - humans mind is complete, means has limited set of "parameters" , but can handle infinite about of combinations of them. i just wonder why realy no project exists about that. local success if false success $\endgroup$ Jun 12 '19 at 21:59
  • $\begingroup$ It has never been demonstrated that the human mind can do infinite calculations. (If it were so, many would be interested in how!) $\endgroup$
    – DukeZhou
    Jun 12 '19 at 22:02
  • $\begingroup$ Updated my answer--I think Google's harnessing of Machine Learning to manage the air conditioning in it's data centers may apply. $\endgroup$
    – DukeZhou
    Jun 14 '19 at 17:24

Consider managing a memory structure as an economic function. (Where to put, and how to manage, the resources constituted by data.) This is something computers can do better and faster than any human. The reason is that the system in which the economic decisions are being made is fully defined.

Routing of packages is a similar, economic function that computers do much better than humans.

These functions haven't been handled by Machine Learning in the past, but, soon after the AlphaGo milestone, Google found an economic application for Machine Learning. Google's DeepMind trains AI to cut its energy bills by 40% (Wired)

So it's entirely context dependent.

As the model increases in complexity and nuanced, utility will be reduced. (In the former case it's a time and space issue related to computational complexity, and in the latter case, often a function of incomplete information or inability to define parameters.)

But as the sophistication of the machine learning algorithms increases, and the models continue to be refined, the algorithms will get better and better at managing intractability and incomplete information.


at this time, as open source - NOT.

i guess:

  • for a decision make we need a broad input layer/-s of data flows,
  • we need a 20 000-200 000 layers of neural networks or more complex and dynamic architectures
  • we need a deep research of date-time influence for historical data flow

what we have at this time:

  • only sensors - opencv and object recognition, nlp-tagging, data predicting

so, sensors isn't AI, sensors and machine learning is previous experience. it is not ready for the change analysis.

  • $\begingroup$ why so many layers? also as i input may be also feedback from humans. And yes, i think there will be need some advanced architectures, not straight feed forward as we know today- But it will be real project with usefull value, not regognizing cats and dogs, whatever $\endgroup$ Jun 12 '19 at 14:41
  • $\begingroup$ en.wikipedia.org/wiki/Ilya_Sutskever was saying that the network with 10000 layers may understand grammar, may be! $\endgroup$
    – nexoma
    Jun 12 '19 at 14:47
  • $\begingroup$ maybe he over-estimates that a little bit, if an 8 layer stuff can recognize a dog. $\endgroup$ Jun 12 '19 at 15:01
  • $\begingroup$ @user8426627 Grammar is substantially more complex than image recognition (easily confused image recognition) but I do agree the 20k-200k figure seems a little like it was plucked out of thin air. $\endgroup$ Jun 12 '19 at 15:13
  • $\begingroup$ yes, but it's only dog/cat with the 28x28-pixel input. so for economic decision you need summarize a economic theory books, economic news, many economic reports and many politics and many others in manufacturing, geology, climat, states, human factor as leaders decisions. it's more broad than the 28 pixel machine learning. 2 input bits have 64 states based on 16 boolean logical funstions. 2^(2^N). and you need 4 bits for choice a function. $\endgroup$
    – nexoma
    Jun 12 '19 at 15:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.