Measuring Intelligence Quotients of Hyper-Intelligent Semantic Networks Alfredo Sepulveda Colorado Technical University September 9, 2013 Abstract Berners-Lee’s initial concept of the Internet was one of a complex, highly connected web of semantic knowledge built from a global collective intelligence. The Internet instead has evolved and cycled through different epochs of scale-free socio-technical-economic subnets of competing information streams, reminiscent, in part, of the growth spurts of print, television and telephony. The meaning of an intelligent semantic web transcends these stages of development – transparent and ubiquitous mobility and utility of processes leading to the construction of modular abstract web units of computational intelligence and resulting composites of computational organs and regimes. Nonetheless, in order to know what is more useful or powerful computationally begs for clarity of a spectrum and measurement of universal intelligence. This paper investigates conceptual types of abstract IQ measurement with respect to human-machine hybrid organizations. This manifestation is presented in the fold of ideas of hyper-intelligent networks that further climb the ladder of cognition in conscious-like, reflective, and thinking computational units possible in globally formed subnets of a semantic-evolving Internet. These intelligences liberate and extend the spans of the collective of human intelligence, transpersonal development possible with the integration of machines, humans, and hybrid computational species, utilizing emergent physical computational concepts, thus resulting in über-networks. The social implications for these potential super-intelligent subnets within the Internet are the vast acceleration of service lifecycles, organization transparency, and new co-opetive scenarios. 2 Table of Contents Table of Figures .............................................................................................................................. 4 Introduction ..................................................................................................................................... 5 Some Terms .................................................................................................................................... 6 Theoretical Framework ................................................................................................................... 7 Methodology ................................................................................................................................... 9 Hypothesis .................................................................................................................................... 10 Scope and Limitations .................................................................................................................. 10 Significance of the Study .............................................................................................................. 10 Background ................................................................................................................................... 11 The web as the ubiquitous social network ................................................................................ 11 Semantic networks, Petri nets, and the web .............................................................................. 12 Measurements of IQ in humans and machines.......................................................................... 13 Emergent notions of machine intelligence .................................................................................... 26 Semantic networks and web technologies .................................................................................... 36 Conclusions and Future Work ...................................................................................................... 42 References ..................................................................................................................................... 47 Appendix ....................................................................................................................................... 54 3 Table of Figures Figure 1 - Rewards-based agent-environment dynamic ............................................................... 16 Figure 2- Petri net for service agent network................................................................................ 24 Figure 3 - Basic Petri net graph components ................................................................................ 55 4 Introduction The ultimate goal of Berners-Lee’s initial concept of the Internet was to have a complex, highly connected web of semantic knowledge built from the collective intelligence of the ensuing network (Berners-Lee, Hendler, and Lassila, 2001). This grew from his concern of the façade of the early web as a social curiosity built on low level communication of knowledge slightly above that of television and telephony. However, the semantic web, as this vision has come to be known, only introduces piecewise ontological power affixed to the contagion effects of social networks (Barabási, Newman, and Watts, 2006). Two developments produce a perceived accelerated collective intelligence for networks: (1) the scale free power-law of dissipation, and (2) individual node decision logic computed on its environment. Social networks in general and social semantic networks, in particular, display notions of intelligence by introducing complexity through network multi-directional feedback and propagation in their respective structures. The true semantic web holds the promise of combining and exploiting social network and computational intelligences. The question remains, “what are these hybrid intelligences.” In this paper, the information web is modeled as a concurrent adaptive agent-based time Petri net. It is also posited that a clearer definition of a network human-machine intelligence metric can be applied to a semantic web by adding degrees of adaptability, concurrent, and non- standard logic complexity to its underlying structure modeled by properly modified time Petri nets. We then extend these definitions to quantum-gravity and general uncertainty networks as future versions of a super semantic web consisting of biologic-inorganic-cosmic agents and further generalize a notion of universal intelligence IQ for these entities. 5 Some Terms Causaloid – a tentative probabilistic causal net framework for quantum gravity that connects spacetime regions (event horizons). Gӧdel automata/machine – a computational machine or automata which is capable of self rewriting its logic in order to improve on a rewards-based optimization. GTU – general theory of uncertainty, a notion that variants of all uncertainty models can be represented by a constrained variable, a relationship functional, along with the underlying general stochastic nature of distribution of the constraint. Statistical parameters are part of this representation in which quantum logic, fuzzy logic, probability, possibility, Dempster-Shafer, and other models of uncertainty are parameterized under this meta-model for uncertainty (Zadeh, 2006). MIQ – machine intelligence quotient, adopted measurements of intelligence quotients which were later developed into standardized intelligence scores of humans applied to decision- processing of computational units and hybrid man-machines. Social networks – network structures consisting of thinking, empathic nodes such as combinations of human, animal, and machine units that help define a collective influence on each other and in outputting results to inputs that mimic decision-making. 6 Semantic web – the notion of a semantically tied Internet via the introduction of ontologies, descriptive languages, collective information, and dissemination of such knowledge and data to give an effect of learning. Petri-nets (PN) – bipartite directed graphs with two classes of nodes that help model network processes with concurrency in actions, conditions for events, states (conditions are met or not), and events. Specifically, a Petri net consists of a set of conditions and events. When certain conditions are met, certain events are triggered. An initial marking of the network is the set of initial conditions met. If a condition is met, a token (dot) is placed (inside) for that condition (usually represented as a circle). An alternative description of a Petri net is that of a mutually exclusive set of places (placeholders) P, and transitions, T, along with a set of directed arcs between the two, with an initial marking, M . A place, p, may be an input, output or both of a 0 transition t depending on either an arc maps to or from a transition to a place. A marking is the set of states of each node of the net. Formal mathematical definitions are given in the appendix. von Neumann automata/machine – a computational machine or automata that is capable of self- replication without self-referentialness. Theoretical Framework Intelligence quota (quotient) notions for computational units, including humans are based on psychometric measurements, repeatable statistical experiments in collecting the potential of general problem-solving. Intelligence quotients were initially achievement scores divided by chronological age. These were subsequently replaced by statistically standardized scores. Collective networks and computational units pose a different type of challenge to this definition of intelligence. Additionally, different, more diverse concepts of multiple human and social 7 intelligences have given rise to new controversies about the traditional view of IQ (Gardner, 1983; Goreman, 2005). Combining these multiple intelligences with an attempt to define a hybrid computational IQ taking into account network collective emergence may manifest a more apt realization of general intelligence in networks of human-machine mixtures. Additionally, concepts from recently developed universal IQs for human-machine agent systems generalize both human and machine intelligence quotas (MIQs). These metrics are more appropriate than Church-Turing-Deutsch or Searle Chinese Room intelligence tests which do not measure intelligence (Detterman, 2011; Dowe and Hernandez- Orallo, 2012). Rather, they measure similarity to human decision making or the equivalent notion that any physical or humanly logical manifestation is computable (Church, 1947; Turing, 1950; Searle, 1980; Deutsch, 1985). Information on the type, architecture, and programming characteristics of the creators of its software and hardware ethos of the machine are necessary in order to more efficiently and accurately measure a truer IQ of that machine. Machines are nonetheless extraordinarily diverse in their makeup. In a sense, the universe of possible machines crafted from anthropomorphic imaginations and intellect add, at least, another potential level of complexity to that of human circuitry. In this way, the notion of universal IQ tests for collectives and hybrids of machines, humans, and other subcombinations thereof is exceedingly difficult and imprecise. To that end, the emergent properties of networks and evolutionary processes within those collections may shed light on producing more powerful notions of universal IQ tests. Combined with novel ideas from non-Newtonian cosmology and physics and non-classical logic systems, hybrid machine/human collectives may be constructed and elevated to higher self-reflective and 8 thinking entities. These prototypical über-entities may then help define novel notions of IQ for emergence and hence for evolutionary universal IQ tests. In this study we examine some of these emergent physico-logico theories of collectives. We define a class of self-replicating and self-writing machines in the framework of emergent physical models such as quantum gravity and generalized uncertainty logics. We commence with self-replicating machines (Von Neumann machines) and universal and optimally efficient self-writing programs known as Gӧdel machines, which are combined to form evolutionary intelligence machines that can then form clusters or networks of intelligent agents in a generalized semantic network intelligence (von Neumann, 1966; Schmidhuber, 2006). The causal and physical structure of this networked machine may then be conceptualized as a quantum-gravity causal network computer utilizing a diverse representation of underlying uncertainty models and grammars (Lloyd, Mohseni, and Rebentrost, 2013; Hardy, 2007; Zadeh, 2006). Semantics of this network are manifested through the use of Petri nets to model the dynamics of concurrent machine states. Time Petri nets are used to simulate concurrency in networks where time constraints are put on the triggering of events. Here, we utilize non- standard logic versions of time Petri nets (fuzzy and quantum) to model the intelligence of web dynamics through its human-machine nodes. Universal IQs are then applied to these Petri net models to generalize semantic networks to diverse web participants and their ensuing collective intelligence. The web may also be framed as an evolutionary machine, as described above. Methodology In this conceptualization, the author utilizes the design science methodology of information systems research in which new notions or paradigms are built (artifacts) from the 9 synthesis of smaller scoped ideas applied to larger scoped ones (Hevner, March, Park, and Ram, 2004). Meta-models for constructing a computational approach to generalize IQ for hyper- intelligent semantic networks will be given based on recent generalization to IQ for human- machine agents and networks. Hypothesis Human-machine networks in general and social semantic networks, in particular, such as smarter notions of a semantic web, have measureable computational intelligence in the sense of optimal Bayesian reward seeking and parsimoniousness (Hawkins, 2004; Hernandez-Orallo and Dowe, 2010). Computational intelligence is measureable through predictability and risk assessment power. Concepts of cognitive improvement, reinforcement learning, and dynamic relationships can be conceived based upon these iterative, emergent, and evolutionary measurements. Empowered networks of self-predicting, self-writing, and self-replicating agents are then optimally intelligent and robust entities. In turn, semantic networks embodied with this agency structure are optimally efficient and robust in a Bayesian global sense. These structures can then be specialized to the web dynamic. Scope and Limitations This paper shall be a conceptual exercise in sculpturing novel ideas about defining IQs for networks of human-machine hybrid nodes and notions of general techno-socio-economic value for these hybrid networks. This study is a research-based exposition that does not collect data nor manifests traditional quantitative or qualitative experimentation. It is a concept paper on the possibilities for defining generalized intelligence for diverse networks of thinking entities. Significance of the Study 10