Paradigms of Personoids 
Adam Maria Gadomski
         White e-paper, since 1997, before on 
         http://wwwerg.casaccia.enea.it/ing/tispi/gadomski/per-para.html 
  ENEA >CAMO> HID Group
   Personoid does not need any human-like physical body,  it is rather an abstraction of certain functions of  human mind which can be considered as a basic entity for any, always goal-oriented, intelligent system.
The personoid cognitive skeleton  is a carrier of "reasoning" frames. Of course, various reasoning "mechanisms" is possible to insert into the personoid architecture.
The TOGA personoid architecture is not intended as an exclusive unique structural framework of intelligent agents/systems. Personoids could be seen as one of  possible abstract "species" of intelligent agents. Therefore if we use the term 'personoids' then we rather think about the TOGA intelligent cognitive agents.
The real personoid utility depends on the model of implemented "intelligence" . It could be installed and specialized on a portable computer  using a generic "intelligence kernel" available on USB key (added: Jul.2005).

The construction of personoids' mind is founded on elementary relations between following basic concepts:

information, i, inf : how situation looks (before, now, in the future)

knowledge, k :how situation may be classified and modeled, and what is possible to do

preferences, p :what is more important

goal, g : what should be achieved.

All mentioned concepts are relative and always refer to a predefined domain of activity (d-o-a) which is real or abstract. The state of d-o-a is represented by information.

Let us clear the meaning of some terms which are used in the subject matter literature.

domain-of-activity of an agent is the reference domain of its/his/her knowledge and, from the point of view of an external observer, it can be called knowledge reference domain.

information - a conceptualization either of the states of the d-o-a itself, or of the state of another world of objects which are symbolically represented in this d-o-a.

knowledge - an abstract carrier of reasoning processes, it is verified in an adequate knowledge reference domain or is acceptable after a rational meta-reasoning.

Knowledge has two components:

    • descriptive knowledge (physically passive) which describe/model possible interrelations between classes of states, and situations valid in a preselected d-o-a; any conceptualization framework which enables an efficient goal-oriented modification of the information about d-o-a (its "image") is a descriptive knowledge.
              Examples: model knowledge, rule-based knowledge, 
    • operational knowledge (physically active) which conceptualizes/describe possible actions/operations on a predefined d-o-a.
beliefs - are not validated/ falsificable components of agent's reasoning processes, and all active data of personoid, i.e.  information, preferences and knowledge . This concept belongs to the meta-knowledge ontology.

intentions - hypothetical states of domain-of-activity which either are candidates to be intervention-goals or were intervention-goals but without success.

For more see: http://erg4146.casaccia.enea.it/GKE-para.htm

All above basic concepts have object-property, i.e. can be aggregated and decomposed according to the TOGA  abstract objects framework (the TAO theory) .

One of the fundamental TOGA assumptions is that i, p, g, k are defined only all together by three generic reasoning processes executed by:

- Domain representation System, DS; it consists of a representation of d-o-a and conceptualization mechanism. ADS transforms signals from d-o-a in information. The information is memorized and is sent to the Preferences System

- Preferences System, PS; it is activated by information coming from the Domain representation System. PS consists of preferences rule bases PRB and an intervention-goal generating mechanism.

- Knowledge System, KS; it is activated by goal coming from the Preferences System. KS consists of knowledge bases and a mechanism of intervention-procedure generating.

Monad is an abstract simple agent, it is a trial system composed of the above complex objects.

Data flow among monad's components is illustrated on  Fig. 1.


Fig.1 Monad as a basic functional module of  reasoning processes (an abstract simple agent).

They can be represented as follows :

inf11 := AD inf00 ;

goal00 := PR inf11 ;

inf22 := KN [goal00] inf12 ;

where : =   denotes  ’become(s)’,

AD, PR ,KN - denote mathematical operators which, subsequently, are properties of ADS, PS, KS systems.
goal - denotes an  intervention-goal (temporal) of the agent (monad) in j-th domain, and it is a control complex parameter of KN.
infij, for j= 0,1,2,3, - denotes information about currently modified element of the domain of activity,Di,
from i -th processing level (Fig.1).
D j is a symbolic representation of the real domain of activity (d-o-a) of  physical agent, this representation is the element of every ADS. In this way, PR produces a goal and this goal activates KN which subsequently produces new information.

Monad may modify ADS but it is not able to change its own knowledge and preferences, therefore every choice of the intervention-goal depends only on the current information.

More flexible are personoids. In the above context, a personoid is an abstract entity which is able to reason about, and to modify its own knowledge and preferences (i.e. to learn and to change goals).

Personoids consist of the hierarchical pyramidal structure of monads. For the modifications of KN and PR of the basic monad, these two systems must became domains of activity for two monads located on the higher meta-level. Every next meta-level may include more monads. This structure is illustrated on  Fig. 2. up to three levels.

 In this way , the generic representation of a learning process is the following:
 inf44(KN1) := KN2 [goal12] inf34( KN1) ;

where

KNn is an operator of a n-th meta-knowledge system,

infij (ADn) is an i-th information which modifies the abstract d-o-a, ADn ,

goal12  is a goal created on the 1st meta-level of AIA, and KN1can be modified by inf44;

KN1 := KN1 Å inf44(KN1),

where X Å Ydenotes a meta-operation of a structural sum of Y and X systems.

The above presented simplified personoid model is a one-domain model.

       (for another illustrations see)

 

    Remarks:

- On the first meta-preference level, the domain-preferences are modified.
- On the first meta-knowledge level we have planning and learning.
- On the second  meta-knowledge level we have teaching.

- Using the initial personoid framework it seems to be possible to define an abstract intelligence

   scale.

 - In practice, personoids should act in two and more domains. Usually, the second domain of personoid activity is the domain of communication, it is managed, for example, by a communication monad. In this case, multi-domain personoids have three-dimensional architecture. Such architecture is now under investigation (see yet)

 

   Application Example:

 

   A.M. Gadomski, S. Bologna, G.DiCostanzo, A.Perini, M. Schaerf. Towards Intelligent Decision  Support Systems for Emergency Managers:  The IDA Approach. International Journal of Risk Assessment and Management, IJRAM, 2001, Vol 2, No 3/4.
 

On  Architecture   |  Personoids' Organization and Society  | Some Conclusions  |  TOGA| Agents and Intelligence  |       Home
Copyrights C 1997 Adam M.Gadomski, MK Eng. & Management Server, ENEA. Last graphical/links updating only: 19 Nov. 2000,
9 Aug. 2002.