INTERNATIONAL WORKSHOP ON MACHINE CONSCIOUSNESS, Hong Kong 14th JUNE 2009

Gate

PROGRAM OUTLINE

The program will consist of the following:

  • An introductory session from 11-12:30, which will be part of the main Toward a Science of Consciousness 2009 conference

  • Two afternoon sessions spanning 1:45-6:15 PM, exclusive to the Machine Consciousness Workshop

You may view the tentative schedule; and will find a list of speakers and abstracts below.

Full papers corresponding to presentations are linked here for those authors who have provided them.  PPT slideshows will be linked here for authors who provide them, after the workshop has passed.

SESSION 1:
Perspectives on Machine Consciousness, 11 AM - 12:30 PM

All Things Are Conscious, But Some Things Are More Conscious Than Others: A Panpsychist Approach to Quantifying Intensity of Consciousness in Natural and Engineered Systems
Ben Goertzel, Novamente LLC

A new formal model of consciousness is presented, building on the philosophical foundations of panpsychism (which holds that everything manifests some degree of consciousness) and multiperspectivism (which holds that first, second and third person views of reality each have their uses for interpretation and modeling). 

The space of qualia is taken seriously as a model of aspects of reality, and is proposed to possess certain algebraic operations and symmetries. A homomorphic mapping is constructed between qualia-space and the space of physically-grounded patterns.  A formal definition of "consciousness intensity" is proposed, which clearly articulates the sense in which "all entities are conscious but some entities are more (intensely) conscious than others." 

Implications for novel, technology-enabled forms of consciousness are discussed, including the possibility of creating machines that are much more intensely conscious than humans, and of "global brain" or "mindplex" technologies that merge multiple human minds into greater consciousnesses.

The Conscious Robot Project: A Neural-Symbolic Approach to Robotically Embodied Machine Consciousness
Hugo de Garis, Xiamen University
Ben Goertzel, Novamente LLC

The Conscious Robot Project, an initiative currently underway at Xiamen University, involves controlling a Nao humanoid robot using software constructed based on the XIA-MAN (eXtensible Integrative Artificial Man) hybrid neural-symbolic cognitive architecture. XIA-MAN is a novel design in which perception and action are achieved via a combination of GA-evolved neural-net modules and additional sensory modality specific components; and cognition is achieved via OpenCog Prime, an integrative cognitive architecture incorporating probabilistic inference, evolutionary program learning and other algorithms unified according to the principle of cognitive synergy. Here we review the XIA-MAN architecture, and discuss why it may ultimately have the capability to give rise to a system with a type of machine consciousness bearing some significant resemblances to human consciousness. We model human consciousness as consisting of sensorimotor aspects, and more abstract cognitive/reflective aspects, and emergent properties arising from the combination of these aspects. In XIA-MAN, a similar variety of consciousness may ultimately emerge based on a combination of neural-net mechanisms that supply the sensorimotor aspects of consciousness, and symbolic cognitive algorithms that supply the more abstract reflective aspects of consciousness.

On the Origin and Embodiment of Consciousness in Cognitive Informatics and Computational Intelligence
Yingxu Wang, University of Calgary

Consciousness is the sense of self and the sign of life in natural intelligence. One of the profound myths in cognitive informatics, psychology, brain science, and computational intelligence is how consciousness is generated by physical and physiological organs in the brain. This paper presents a formal model and a cognitive process of consciousness in order to explain how abstract consciousness is generated on the basis of physical and physiological organs. The hierarchical levels of consciousness and its generation are analyzed. A rigorous mathematical model of consciousness is created that elaborates the nature of consciousness with a computational intelligence treatment. The cognitive process of consciousness is formally elaborated using denotational mathematics. A wide range of applications of the abstract intelligence model of consciousness has been identified in cognitive informatics and computational intelligence toward the mimic and simulation of human attention, perception, and the reflective/perceptive/cognitive/instructive intelligence. 

SESSION 2:
Conceptualizing Machine Consciousness, 1:45 - 3:45 PM

Workshop Introduction, and Review of Contemporary Architectures for Machine Consciousness
Ben Goertzel, Novamente LLC

This workshop introduction will survey several cognitive architectures currently being developed with the goal of creating human-level AI systems. Architectures reviewed will include: SOAR, Stan Franklin's LIDA, Nick Cassimatis's PolyScheme, Joscha Bach's MicroPsi, and the presenter's OpenCog Prime. In each case attention will be paid to why the architecture's designers believe their work might ultimately lead to systems possessing roughly humanlike conscious experience. The question of the different types of conscious experience that the different architectures might lead to will be briefly pursued.

Machine Consciousness: Essential, Functionalist, and Mixed Approaches
Allan Combs, California Institute of Integral Studies
Liane Gabora, University of British Columbia

See extended abstract


Robot Dreams: Requirements for Synthetic Phenomenology and Intersubjectivity
Robin Zebrowski, Beloit College

In spite of both the hopeful proclamations of Alan Turing and the apparent faith in the future of technology by many working in the field of Artificial Intelligence (AI) even now, the methodological approaches found in primary AI research have failed to produce great successes beyond theories about what the future holds. Philosophically, the goal is still to build a mind, and not just the appearance of a mind. Given this assumption, I undertake an examination of a more authentic approach to building AI with the goal of creating a being with actual experiences (what we might rightly call phenomenology) that we can recognize as authentic and engage with intersubjectively. To talk about any other sort of AI is to concede the weak position of simulation rather than creation. To solve this problem, we first must recognize the psychological fallacy: that something-for-us is not the same as something-for-the-machine. For example, cameras for eyes may turn out to be the wrong starting point altogether for information-gathering purposes. We are trying to get something fundamentally different from us to have experiences of a sort comparable to our own. Rather than a piece-meal functionalism, we must ask how something made of a different material entirely best interacts with the richness of the world. As humans, we see cameras as the obvious analog of vision for a machine, but the machine needs to be able to break down information that is useful for it, without regard for us as observers. We must concede, given the current state of neuroscientific research on embodiment (Damasio, Gallese), that a physical body is not a luxury, but a necessary starting point for a successful AI program. The question remains – what sort of body? Given the theoretical work in concept acquisition by Lakoff and Johnson, as well as its recent endorsement by Pinker, it seems hard to avoid the claim that AI which we are able to recognize as AI and communicate with will necessarily have a humanoid body, at least with the very general analogs of human embodiment. Exactly what this entails is still unknown, since the body was ignored for so long in traditional AI theory. What is clear is that we must build something that can sense the world in a way that is useful for whatever sort of thing it is, and not simply try to endow human senses into a new organism. We do not hook a camera up to a blind person – instead we aid their inherent abilities in the world so that the information they extract is useful to them. If true identical replication of our sorts of bodily senses were the answer, would we not also be engaged in building synthetic taste sensors as readily as we are working on vision? Thus, when we re-think our approach and methodologies, we  enable the possibility of a synthetic phenomenology – experiences for the thing in the world, as well as the chance of true intersubjective experiences between us (as embodied human minds) and whatever final form AI takes. There is a level at which it is necessary that any possible AI resemble us, and there is a level at which we must recognize that such systems are fundamentally not like us, and build them appropriately. This paper explores the methods by which we have been unsuccessful, as well as the research on the role of the body in cognition that should convince us of more fruitful avenues in the quest for synthetic phenomenology and intersubjectivity.

Can Access Consciousness Qualify as Computer Consciousness? Or, So What If My Computer Can’t Cry!
John Wallace

Of the many divergent concepts and notions that are entertained under the broad domain of Consciousness there are two that have become central to the search for an encompassing theory of Consciousness that will rest easy with philosophers, psychologists, computer scientists, and cognitive scientists alike: Phenomenal Consciousness (“P-consciousness”), and for our purposes Access Consciousness (“A-consciousness”) as flushed out by the Philosopher Ned Block. Of these, P-consciousness has been and continues to be the subject of extensive analysis and pontifical pronouncements as to its generative sources and character. While A-consciousness has been debated, it has not endured the intensive attention of P-consciousness mainly because it does not present the perplexing “unresolved” issues that adhere to P-consciousness. In part this is because the various concepts and notions that are pertinent to Consciousness, and P-consciousness in particular, have been developed principally as they apply to human beings, not animals, not plants nor fauna, not insects, and certainly not Computers or Robotic entities known descriptively as possessing Artificial Intelligence (“AI”), even though such AI models have been used to facilitate the study of the human mind and consciousness.  Furthermore, it is certainly far from settled among and within the various professions that study the issues what the actual attributes of a complete theory of Consciousness should be, let alone are. The problem of ‘Other Minds’ still lacks a definitive theory. Simulation Theory still underlies much that is under development in the accounting of knowledge of other minds; Problems with Simulation Theories notwithstanding. Also portending as serious contenders for a theory of consciousness are the neurological theories of consciousness seeded within the unified endurance of the firing of neurons within specific areas of the brain. Connectionist theories of psychology and cognitive science postulate that all consciousness activity is brain centralized neural/synaptic. Adapting many of the underlying assumptions of the Parallel Theory of Computer Science as the prime model for their theories, cognitive scientists believe that consciousness is simply another bio-chemical moment. Indeed, if the electro-static discharge was all there was to human consciousness, then why do these simple solutions not satisfy the issues that remain with regard to the body and the quantum issues that are raised by physicists regarding the physical community and communication within that community that Quantum and Relativity Theories hold as basic assumptions? But, to pursue these questions is to pursue ‘what it is like to be human’, questions relevant to P-consciousness, not our objective in this short paper. Taking these points under advisement, recent developments in Robotics and AI signal that a fresh look should be given to raising the following question in a more serious vain: Are computers, or at least AI-Robots Conscious? And if so, what might this Consciousness be like? I propose that the notion of A-consciousness comes close to providing a foundation for a theory of Consciousness applicable to AI, not as a model for human Consciousness, but as descriptive of a large portion of what goes on for ‘what it is like’ to be AI.

Towards Machine Consciousness: a Developmental Theory of Thoughts as Communicating Processes
Pierre Bonzon,  University of  Lausanne, Switzerland

We present a model of thoughts given in terms of concurrent communicating processes, to serve as a basis for machine consciousness. Towards this end, we rely on a language implementing threads within the framework of logical programming. This allows for both a formal definition and the running of simulations. Adopting a developmental approach, we start with a simple model of purely reactive behaviour. We show how it can be extended into an agent model incorporating thoughts and deliberation, and then almost naturally evolve to incorporate a functionality of consciousness akin to episodic memory.

SESSION 3:
Mechanisms of Machine Consciousness, 4:15-6:15 PM

CERA-CRANIUM: A Test Bed for Machine Consciousness Research
Raul Arrabales, Agapito Ledezma and Araceli Sanchis, Universidad Carlos III de Madrid

This talk describes a novel framework designed as a test bed for machine consciousness cognitive models (MCCM). This MCCM experimentation framework is based on a general- purpose cognitive architecture that can be integrated in different environments and confronted with different problem domains. The definition of a generic cognitive control system for abstract agents is the root of the versatility of the presented framework. The proposed control system, which is inspired in the major cognitive theories of consciousness, provides mechanisms for both sensory data acquisition and motor action execution. Sensory and motor data is represented in the proposed architecture using different level workspaces where percepts and actions are generated thanks to the competition and collaboration of specialized processors. Additionally, this cognitive architecture provides the means to modulate perception and behavior; in other words, it offers an interface for a higher control layer to drive the way percepts and actions are generated and how they interact with each other. This mechanism permits the experimentation with virtually any high level cognitive model of consciousness. An illustrative application scenario, autonomous explorer robots, is also reviewed.

The Android’s Wardrobe: Emotions, Preferences, and Pre-Rational Decisions in Machine Consciousness
Paul Fahn, Samsung Electronics

Will emotions have any proper role in machine intelligence? Emotional systems are present in humans and other animals and seem to serve as mechanisms for primitive decision-making. As a result of evolution, however, humans are capable of logical thinking, which produces superior decisions, thus obviating the need for emotions. There would therefore appear to be no reason for AI researchers to attempt to endow androids and other “conscious machines” with emotions, although the ability for such machines to detect or predict emotions in humans may be useful. There remains one class of everyday decisions, however, where emotions may never be completely replaced by rationality: simple personal preferences, such as which shirt to buy in a store, or what food to order from a restaurant menu. We speculate that these “pre-rational” decisions, as we call them, may be implemented in machine intelligence by randomized algorithms or similar stochastic processes.

Intentionality and Goal-Oriented Behavior from Reinforcement Learning
Jun Zhang, University of Michigan, Ann Arbor

Intentionality is one of the key elements of consciousness. This talk will address the origin of the notion of intentionality for any intelligent machine system. It is argued that the two basic principles of animal learning, namely, Thorndike’s Law of Effect that governs instrumental (operant) conditioning (establishing reward consequence of an action), and Pavlov’s Law of Association that governs classical condition (establishing reward values of a stimulus), will lead to a computational architecture of goal-oriented behaviors, so long as the system can categorize the states-of-the-world. Constructing internal models of rewards and state-transitions and projecting them to a common workspace constitute what is commonly referred to as “goal” or “intention” from a computation perspective.

How can we program a nano-brain so that it can run complex jobs without human intervention?
Anirban Bandyopadhyay, National Institute for Materials Science, Tsukuba, Ibaraki, Japan

Our recently proposed nano-brain for nano-bots operates as a 16 bit parallel processor but it executes only one job at a time.  One instruction generates 16 instructions and as soon as the new set of instructions is generated, the entire operation stops. Therefore no practical work can be done using this device.  What is desirable is that, once instructed, the nano-brain should continue to work for a long time, generating multiple new sets of instructions internally and thus performing complex jobs.  Then the next steps would be controlling machines attached to the nano-brain.  Sometimes one machine would be required, operating in a particular way, and sometimes none would be required.  Coherent and decoherent signal processing should be generated collectively during signal transmission between different parts of the nanobrain.  We have solved all these basic problems using simple tricks of tuning the vibration modes of molecules in a very particular way. This is a very primitive yet fundamental step towards developing nano-factories that can operate in a hostile environment.

Engineering Computer Architecture Models for the Evolution of Interiorized Creator-in-a-Box Roles for Conscious Robots, Self-Aware Robots, Automata, Machines and Artifacts
John-Thones Amenyo, York College, City University of New York (CUNY)

This paper introduces the concept of artifact-as-organism as the appropriate correspondence of engineered, cybernetic artifacts and machines with biological organisms and their biological or natural evolution. The concept of artifact-as-organism is useful because such “organisms” can be regarded as beingsubject to laws, principles, processes and mechanisms, similar to those operating to affect the evolution of biological organisms. The related Creator-in-a-Box concept is then described, to establish support for evolution or “mobility-in-wide-sense” (diachronic evolution, change or dynamics, for short term and long term survival, sustainability, longevity and persistence). The Creator-in-a-Box role is advocated as the primary and essential function that engenders the need and presence of manifestations of self conscious minds.  Operationally plausible characterizations of self consciousness minds are then advanced as embodiments and manifestations that can be used to support the Creator-in-a-Box concept in autonomous (and possibly mobile) cognitive robots. The paper then concludes with a successive, hierarchical refinement of a supporting PMSCIO-computational architecture model that can implement the Creator-in-a-Box for robots, artifacts, etc.,

A mental mechanism explaining the developmental emergence of consciousness: The three rules leading towards conscious being
Qishi Li, Fudan University

Throughout human history there have been numerous discussions about mind, consciousness and soul, about why we feel and experience. Most resort to religion and some to philosophy, until rather recently. Through research in the field of neuroscience concerning the interconnectivity of neurological loops and synaptic firing patterns, along with the inspection of the related phenomenon of “previous thoughts triggering a future thought”, I propose a theory that attempts to explain how mind works. As a result of the theory, with the help of linguistic inferences, I deduce that consciousness (the identity experience) is a developed experience rather than something present at birth.  Also deduced are three rules that govern the developmental emergence of consciousness: (1) Substrate with interconnectivity (2) Conception and interception (ways in-and-out) (3) Mechanism that can create the orientations of the substrate called “loops”, which produce “preference loops” that cause derive-forward logics. If a machine is built according to these three rules, it will build up consciousness on its own.  Also included are discussions of soul, singularity, scientific ways to paradise and creative ways to implement them.


(C) 2008 by Ben SALEM