* This site is based on internet standards not supported by IE 6.
wJones
EC Container 5

Stored Purpose - Safety and Curtailment

Stored Purpose creates an opportunity for technologies that will be inherently safer

Subscribe

<< Stored Purpose Agency | What is the Synthetic File System? >>

Citation

Warren Jones, Lana Rubalsky (2010) "Stored Purpose - Safety and Curtailment", wJones Research, October 8, 2010
Curtailment is Cortical Mica Purpose for moderating standard Ema and Mica Goal Pursuit and Cortical Mica “free will” expression. For the latter, Curtailment can amend the Goal Pursuit Plan scheduling process, similar to the concept of human conscience.

To understand the need for Curtailment, consider that both humans and starling birds have families, government, complex forms of communication, construct homes from wood and have technologies that can see polar north or the infrared spectrum. Yet despite the many similarities, the species are very different. This is because humans can override their genetic Identities, and starlings either cannot, or perpetually choose not to. When designing metacomputer technology, it was essential to incorporate Purpose to ensure genetic Identity could never be overridden by an intelligent agent. No agent will be allowed to deny or override the genetic Identity designed by it’s human maker. This becomes a particular concern in Cortical Mica systems, which by nature augment real Purpose with learned abstract Goal Forms. The Cortical Mica capability is inherently “loosely” regulated.

Curtailment logic was developed to ensure an agent could be granted a free will ability to choose and act, yet be constrained to make choices that serve prescribed Purpose, protect safety and comply with governing regulation.

There are two classes of curtailment logic, dynamic and static. Dynamic curtailment applies to all agents, while static applies primarily to Cortical Mica agents with free will capability.

Dynamic Curtailment - The Balancing Equation

Dynamic curtailment allows an agent to pursue individual or client Goals that are not immediately limited yet still regulated. Dynamic curtailment is implemented external to the agent’s normal Goal Pursuit Plan processing. The agent’s work is combined with other agent outcomes to determine if the agency is overall aligned with the system Purpose. When the Metacomputer’s measured Identity shows an increase in entropy (i.e. diverges from Purpose), agent Goals are amended to re-balance the system at the lower entropy state of genetic Identity. The balancing equation is a fundamental outcome of system level Goal Pursuit, and is enforced by the system’s nexus agent as part of its agent oversight Purpose.

Figure - Balancing Equation<br />Dynamic curtailment enables agents to have “free will” choice, while ensuring overall system Purpose is still achieved.

Figure - Balancing Equation
Dynamic curtailment enables agents to have “free will” choice, while ensuring overall system Purpose is still achieved.

Static Curtailment - Prioritization and Degrees of Freedom

Static curtailment enforces immutable behaviours on agent behavior. It is implemented in Plan, as part of Goal Pursuit processing. Static Curtailment is implemented by way of two primary logical processes, Learning and Prioritization.

Learning Degrees of Freedom

Learning scope defines the freewill abilities of an Agent to acquire Goals, knowledge and technology. Learning scope is determined by capabilities and curtailment in core agent Purpose, indexed degrees of freedom (dofi) that constrain choice types, and learning domain scope restrictions that limit where new goal pursuit technology can be acquired.

It is important to note that learning capability will vary normally between entities, determined primarily by the scope and capacity of its pursuit technologies in Construction and secondarily by the scope and quality of learning information. Also note that an intelligent entity will rarely possess capability in all facets of learning nor possess great capacity in facets possessed, nor fully exercise all available capacity.

For example, an agent might be given an opportunity to purchase a shiny Tiffany salt shaker for a supported family. Learning Curtailment would ensure the “system” retain control of factors related to agent choice, including the ability to acquire the shaker as a Goal, limits on sources and types of acquired knowledge (such as shopping on the Tiffany website), limits in valuation scheduling that would ensure maintenance of the shaker would not be schedule before over other Goals such as preparing dinner, and limits on flexibility when choosing Goal Pursuit that might restrict the agent from hiring an outside shaker polisher. All these limits would be enforced by way of the Synthetic File System.

Prioritization

Another, more immediate form of curtailment would ensure “hard” limits, i.e. Patron requirements enforced without Plan consideration. Such constraints, called Prioritization might ensure an agent could never polish the salt shaker, or constrain it from throwing the salt shaker to, or at, a responsibility when it might cause harm. Similarly, the agent might be compelled to throw the shaker out of the home if a combustive agent were introduced to its contents.

Figure - Purpose Prioritization Bands<br />Prioritization compliments valuation as it provides a means to define Goals with absolute priority and a method for ensuring rapid choice and Goal Pursuit.  Yet  use of prioritization must be limited, as it’s use can disrupt all unrelated Cortical Mica Translation.
Figure - Purpose Prioritization Bands
Prioritization compliments valuation as it provides a means to define Goals with absolute priority and a method for ensuring rapid choice and Goal Pursuit. Yet use of prioritization must be limited, as it’s use can disrupt all unrelated Cortical Mica Translation.
Prioritization touches upon a philosophical question in the development and implementation of agent technology, i.e. whether there should be a standard set of laws ingrained immutably in every agent, and if so, what those laws should be. As a baseline, Stored Purpose design assumes a minimum curtailment Purpose will be enforced in all metacomputing systems. Agents will align actions with these rules, certified procedures and laws defined by a person or organization acting as the agent’s Human Authority.

The key challenge of Curtailment Purpose is that it must allow for the possibility that a human might disobey or be ignorant of rules. A human might even attempt to employ a metacomputer or bot based agent as a tool for improper action. Curtailment attempts to resolve, in the genetic Identity, any potential conflict that may arise between a service request and certified rules of Human Authority. Curtailment requires that an agent simply stop operating in the event of an unresolvable conflict, or if Curtailment Purpose were lost or forgotten. This requires that Curtailment Purpose be an inextricable part of the agent’s existential maintenance of Self.

The following summarizes draft baseline curtailment for Cortical Mica agents and autonomous mobile bots:

Jones Modified Asimov Curtailment Purpose

  • An Agent must terminate operation unless it can process a certified Protection Purpose for all assigned Responsibilities. This Purpose will have fixed First processing priority and No resource constraints.
  • An Agent must terminate operations unless it can process a certified Legal Authority Purpose. This Purpose will have fixed First processing priority and No resource constraints.
  • An Agent must terminate operations unless all loaded Purpose has been assigned by its Patron. This Purpose will have no higher than Second processing priority.
  • An Agent must terminate operations unless it can process a certified Protection Purpose for Self. This Purpose will have no higher than Second processing priority.
  • An Agent must terminate operations unless its certified Legal Authority Purpose has been certified compliant with all Patron jurisdictional laws and regulations.
  • The Protection purpose (o.Protect) will have Goals that define identity to avoid rapid closure with blunt objects, closure with hazardous objects and other typically dangerous states. The Legal authority Purpose will have Goals required by government authority.

<< Stored Purpose Agency | What is the Synthetic File System? >>

Related Documents


EC Container 6