User Tools

Site Tools


public:events:challenges6-team-results

<-main workshop page
humanobs header

Challenges 6: Team Conclusions & Solutions

Lecture: FROM CONSTRUCTIONIST TO CONSTRUCTIVIST AI: ARCHITECTURE MATTERS
Author: Kristinn R. Thórisson

Challenges

1. If constructionist methodologies are sufficient, how do we create a complex architecture (by hand) that:
- can learn to act successfully in not only one but multiple - unforeseen - environments?
- can learn novel things, even invent new (useful) things? 
2. If constructivist methodologies are a necessity for achieving AGI, how do we build a complex architecture that: 
- can learn to act successfully in not only one but multiple - unforeseen - environments?
- can learn novel things, even invent new (useful) things?

Team A

Team Members:
Gudny R. Jonsdottir, IIIM, Iceland
John-Jules Meyer, U. of Utrecht, the Netherlands
Marjan Sirjani, Reykjavik U.
Ricardo Sanz, U. Madrid

Team Conclusions:

Team B

Team Members:
Pei Wang, Temple U., USA
Helgi Páll Helgason, Reykjavik U.
Mark Wernsdorfer, U. Bamberg, Germany
Yngvi Björnsson, Reykjavik U.

Team Conclusions: Architecture might be interpreted in two different ways in the constructivist context. For example, there are two different architectures in NARS: structure of system knowledge - which will eventually decide how the system works.The first is completely built by the system itself: Self organizing conceptual network. The software architecture (meta-level) completely hand coded in NARS. The constructivist classification/boundary is not so clear cut. Is modifying own source code a requirement for being constructivist? What is the limit to how far you can go in the constructivist direction while only modifying data and not code?

There is a strong relationship between constructivist and meta-control. Our environment is a hierarchy of wholes and parts. Very difficult to recover if you change your source code in the wrong way. The new methodology is that you don’t provide production rules, you want the system to generate them itself. At which levels of description should self-programming be allowed?

Constructionist:

  • Fixed way to build systems, hand-write metalevel and operation level.

Constructivist:

  • Program language capable of programming itself.
  • Architecture that allows autonomously created components.
  • Extracting operational knowledge from the environment and integrate without breaking system.

Is source code and operational knowledge the same phenomenon in the constructivist case?

Pei also stressed the importance of seperating executable objects from data.

Team C

Team Members:
Hannes Högni Vilhjalmsson, Reykjavik U.
Deon Garrett, IIIM, Iceland
Haris Dindo, U. Palermo, Italy
Jörg Siekman, DFKI, Germany

Team Conclusions: The first question is a rhetoric one! Of course you need general methods to build complex systems! But: we will shortly have complex structures resembling that of the brain in terms of number of nodes and interconnections (e.g. internet), the question is how are they organized to produce something intelligent?! What about projecting the Constructivist methodology forward: we are maybe not ready yet to build general intelligence. Take the analogy with humans creating first computers that are now creating more powerful computers out of human reach: recursive constructivist. Design intelligence vs. intelligent design!

Team D

Team Members:
Eric Baum, USA
Antonio Chella, U. Palermo, Italy
Bas Steunebrink, IDSIA, Switzerland
Hamid Pourvatan, IIIM, Iceland

Team Conclusions:

Team E

Team Members:
Hrafn Th. Thorisson, IIIM, Iceland
Anna Ingolfsdottir, Reykjavik U.
Eric Nivel, Reykjavik U.
Kristinn R. Thórisson, Reykjavik U. / IIIM, Iceland
James Bonaiuto, Cal Tech, USA

Team Conclusions:

  • Reasons why constuctionist approach is not appropriate
    • The scale is beyond our abilities
    • We don't know all possible situations in advance
      • Need code constructed on-the-fly for this
    • Robustness - constuctionist systems have limited fault tolerance
      • Some parts may be robust, but not all
  • Constructionist approaches are still of interest in bounding the system (link to Ricardo Sanz's talk)
    • The owner has to tell the system what to do
    • Apply a constructionist approach to the teacher for the system
public/events/challenges6-team-results.txt · Last modified: 2011/10/03 13:30 by thorisson