Paper Discussion 9

[Hannes Vilhjalmsson]

On Monday we will be joined by Kristinn R. Thórisson, who is one of the co-founders of CADIA at Reykjavík university and currently the director of the Icelandic Institute for Intelligent Machines (IIIM). The reading that he has assigned is the following:

Thórisson, K. R. (2009). From Constructionist to Constructivist A.I. Keynote, AAAI Fall Symposium Series: Biologically Inspired Cognitive Architectures, Washington D.C., Nov. 5-7, 175-183. AAAI Tech Report FS-09-01, AAAI press, Menlo Park, CA.

Post your questions here before Monday and Kris will get a chance to sort through them. Comments and other discussion points are also welcome.

[grimurtomasson]

How does moving to a massively parallel meta-programming system affect model checking - verifying that the software system does what it is intended to.

Have there been any studies on the ability of humans to think in terms of a very high level of concurrent processes, small as they may be? In other words, how viable is it to have humans create the bootstrapping part of the general AI?

[carmine]

- Is it possible to convert an already developed constructionist system in a constructivist one or you should start again from 0, throwing away all the code?
- Do you think that a general AI could obtain the same result (in terms of efficiency and optimality) than an AI specialized on a specific problem? An example using humans: could a scientist (that have a huge generic knowledge) be better than workers specialized in a particular domain (drivers, mathematicians, artists..)?

[Helgi Siemsen Sigurðarson]

1) Could you explain Cognitive Map ?

2) are there any systems that use Constructivist to date (in the commercial market or other wise) ?

[lorenzo]

- Can you give us more infromation about the science of "self-organization "?

- Can you give us other example on how we can build an operational knowledge of time into AI architecture?

[kristofer kristofersson]

1. what is HeLD
2. is I disagreeumption right that from the viewpoint of the writer that the A.I programs that we have now are not true A.I's since they are made to do only specific things (today's A.I's are built specifically for a specific problem instead of trying to make an A.I that can learn to do anything) and only a true general A.I that could learn to do anything should be called a true A.I

[Ásgeir Jónasson]

1.

Do you think there will be a greater need for systems with general A.I. in the future ? Most A.I. systems today have a specific job to do and only need to be able to do a subset of the things humans can do. So, do you think industry will start to demand systems with more general intelligence ?

2.

The article states that an intelligent system must have a sense of external and mental time and understand how these relate. Has this generally been a big concern in the past when constructing A.I. systems and how greatly does, for example, ASIMO factor in time when making decisions ?

[thorsteinnth]

Mar 12, 2011 13:02:44 GMT -5 Helgi Siemsen Sigurðarson said:

1) Could you explain Cognitive Map ?

This is something I would like to know also.


2. Quote from the article;

...the annual Robot
Challenge competition at AAAI calls for competing teams
to create a social robot that, starting at the conference building’s entrance, can locate the registration counter and register for the conference, find its way to the proper lecture
hall and give a two-minute lecture about itself, as well as
take questions from its (all-human, as of yet) audience

How can problems like these be abstracted?

[finnur]

in the AAAI, how well have the robots been performing, especially the part about taking questions from the audience?

[Eiríkur Fannar Torfason]

1. Have you built a general, modular AI framework and if so, are you actively pushing it towards a constructivist approach?

2. Have any AI systems been built that exhibit self-organizing behavior?

[Elín Carstens]

1) Would it be possible to verify theories, in psychology and other social sciences for example, using Constructivist methods, where the theories would “emerge” from the system in some way? If so, would the results be considered valid?

2) What kind of new general-intelligence architectures/systems do you expect to emerge from / because of Constructivist methods? Would they be human-/ animal-like or something completely different?

[Jón Þór Kristinsson]

Are there any programing languages available today that fit the need of the Constructivist approach?

When using a system composed of hundreds of thousands of modules, does each module have a limited function and are there then modules that use these to carry out more complex tasks?

Has it been determined whether an extremely small size of peewee granularity is required for self-construction?

[Hrafn J. Geirsson]

1)
In what ways exactly should we stray from traditional software design methodologies ?

2)
Can you name an example of a successful Constructivist A.I. System ? How can we know that construtivism truly beats constructionism ?

[una]

1)
In the paper it's written
"some developed by the same team during the same
periods (e.g. the visual perceptors modules in the Cognitive Map) while others are developed in separate parts of
the world by separate companies (e.g. speech recognition)."
In your opinion and/or experience, do you think that development works better when it's done by the same team in the same period or when it's developed by seperate companies in seperate places?
2)
In these AAAI competitions, are there any AI machines that have performed exceptionally well that you know of? Does the audience have a certain range of subjects that it can question the machine, and if so, what are these subjects?

[niccolo]

Can you explain meta-programming and why do you think new languages are needed? What characteristic are you looking for in the design of these future languages?