Following link shows the definition of Narrow / Weak AI.
http://whatis.techtarget.com/definition/narrow-AI-weak-AI?src=5562889&asrc=EM_ERU_65692481&uid=6904983&utm_content=recipe3&utm_medium=EM&utm_source=ERU&utm_campaign=20161006_ERU+Transmission+for+10%2F06%2F2016+%28UserUniverse%3A+2198073%29_myka-reports%40techtarget.com
Questions:
Obviouly that this question is difficult. If we just consider the Definition of Strong AI as Ray Kurzweil, ie. The amount of intelligence a machine can generally exhibit regardless its internal workings, then two Major Problems are posed:
The first Problem is that most of the Weak AIs are situated. That is, they are built for and integrated in particular or specific Goals and Environments. In this case, even with a federation, the whole System will still be considerably limited.
The second Problem is a little bit more complex and quite not easy to formulate, but it is all about the existence of such a System. The Thing is, if GENERAL=No Particular GOAL. Question: When you start your System, What does it do? Are you going to situate the System all the times?
Now assume that GENERAL=GENERIC GOAL. What is a GENERIC GOAL? To answer this question, we may ask ourselves: What is a HUMAN BEING'S GOAL, when the Person is Born, since we aim to develop complete human-like intelligence. A nice answer would be THE BEST AND LONGEST POSSIBLE LIFE. To reach this Goal, Sub-Goals have be to reached, repeatedly followed by Learning and Self-development. By repeatedly and intensively solving some Tasks while trying to achieve some Goals, humans become experts in such Domains and consequently intelligent at doing them. This illustrates the fact that Intelligence is typically domain-related: Can we find a Human that is intelligent at all Tasks? From what we just said above, the answer would be probably no. The Thing is, we become more and more intelligent(learning from experience) while trying to achieve our Goals. And achieving our Goals constraints each Human to only Focus on a set of Problems, namely what the Person usually Encounters. So Humans that are intelligent at all Tasks can be difficult to find; hence the question, Why do we want to build complete human-like intelligence which is more powerful than Human intelligence? This is contradictory.It is like trying to do more than we can actually do. So this causes us to reformulate the Problem as follow:
First reformulation: Do we want to develop artificial Systems which completely act like Humans? but how do Humans act?A completee artificial Human should be created?
Second reformulation: Do we want to build a stack of AI-Systems for solving Problems in differents Domains? In this case the whole System will have to be situated each time and there will be no global autonomy and efficiency.
Third reformulation: Do we want to build a single AI-System that combines and emulates several human cognitive faculties to reach a set of well-defined Goals as intelligently as Humans do? Yes, we can and intelligent robots are typical examples.
According to me, Goal is fundamental to System design. However, there are conceptually infinitely many Goals. Without Goal, there is nothing. Strong AI is a serious philosophical issue in AI.
What you need is an ensemble classifiers which is also called a meta-classifier where you can combine different weak or standalone strong classifiers and build a stronger and more accurate one.
Have a look at AdaboostM1 or Majority Voting based ensembled classifiers.
- Badges
- Science topic