NiHA: A Conscious Agent, WM Qazi, JA Ware

Tags: Tenth International Conference, NiHA, Cognitive Technologies, pp, S. Franklin, artificial intelligence, intelligence, QuBIC, conscious, circadian clock, Journal of Consciousness Studies, mental processes, R. Manzotti, artificial general intelligence, Cognitive Computing, Cognitive Architecture, Sensory Memory, Cognitive Systems Research, selective attention, unconscious components, cognitive processes, level of consciousness, Quantum Mechanics, Artificial Consciousness, Artificial Intelligence in Medicine, Working Memory, International Conference, cognitive process, contents, M. Sassoli de Bianchi, bio-inspired, Science University of South Wales, modeling, COMSATS Institute of Information Technology, Lahore, Pakistan, machine consciousness, quantum computation, cognitive features, quantum probability, Department of Computer Science, Cognitive Psychology, Quantum measurements, D. Aerts, Jonathan Andrew Ware Faculty, Journal of Mathematical Psychology, A. Khrennikov, S. Sozzo, Department of Computer Science, Government College University
Content: COGNITIVE 2018 : The Tenth International Conference on Advanced Cognitive Technologies and Applications
NiHA: A Conscious Agent
Wajahat Mahmood Qazi Intelligent Machines & Robotics, Department of Computer Science, COMSATS Institute of Information Technology, Lahore, Pakistan e-mail: [email protected] Jonathan Andrew Ware Faculty of Computing, Engineering, and Science University of South Wales, Wales, United Kingdom e-mail: [email protected]
Syed Tanweer Shah Bukhari Intelligent Machines & Robotics, Department of Computer Science, COMSATS Institute of Information Technology, Lahore, Pakistan Department of Computer Science, University of South Asia, Lahore, Pakistan e-mail: [email protected] Atifa Athar Intelligent Machines & Robotics, Department of Computer Science, COMSATS Institute of Information Technology, Lahore, Pakistan e-mail: [email protected]
Abstract-- This paper reports ongoing work in the development of self-aware artificial general intelligence, called Nature-inspired Humanoid Cognitive Computing Platform for Self-aware and Conscious Agent (NiHA). It is based on a quantum and bio-inspired cognitive architecture for machine consciousness and Artificial Intelligence. So far, a number of different cognitive features have been implemented to help facilitate the realization of NiHA. These implemented features include imaginations, dreams, personal semantics, psychopsychological based motivations and ethics. Apart from these, some industrial applications are also underway. The review of the results obtained from these applications is encouraging and supports the potential to achieve a certain level of consciousness. Keywords-machine consciousness; NiHA; cognitive machines
regard, several projects have been initiated which include the Blue Brain Project [25]; the Human Brain Project [26]; DARPA and the Brain Initiative [27]. In addition to these projects, there have been recent corporate efforts by the likes of Google [28]-[30], Facebook [31]-[34], and IBM [35]. A roadmap for addressing the challenges of bio-inspired modeling is shown in Figure 1. One of the most challenging of these is getting humans to accept them as co-workers (socio-cognitive agents) [11][36].
I. INTRODUCTION Recent years have witnessed exceptional progress in the field of Artificial General Intelligence (AGI) and Machine Consciousness (MC). Considering the current pace of technological evolution, it is envisaged by some that in the future, machines will become self-aware and even potentially become part of a man-machine society. The idea of implementing consciousness in machines is thus not new [1]-[5]. It does, however, raise several philosophical and computational issues [5]-[7]. Addressing these issues requires the unification of different knowledge domains such as physical sciences, neuroscience, psychology, and computer sciences [8]. Consequently, it has resulted in the establishment of computational theories of the mind [8]-[11], and domains which include both bio-inspired [11] and quantum-inspired methods [8][12]-[17]. Various researchers have contributed to the construction of computational equivalent of the human mind/brain [7]-[8][18]-[24]. In this
Figure 1 - Roadmap for the Bio-inspired Challenge [11] From the roadmap, it can be seen that these sociocognitive agents must be capable of producing believable conscious-like behavior and a sense of their presence in their human co-workers [11]. The benefit of such behavior is clear when considering socio-assistive robots, domestic robots, and personal robots. Examples of such machines acting as co-workers have long appeared in entertainment
Copyright (c) IARIA, 2018. ISBN: 978-1-61208-609-5
COGNITIVE 2018 : The Tenth International Conference on Advanced Cognitive Technologies and Applications
films (for example, `Geminoid F' an actoroid, acted in the film `Sayфnara, 2015'), and more recently a robot named Sophia, was awarded citizenship by Saudi Arabia [37]. Moreover, significant work has been done in the form of cognitive computing which includes IDA [9], LIDA [38], WATSON [35], CLARION [39], SIRI [40] and CORTANA [41]. Although current progress is encouraging, there are still many issues that need to be addressed before reaching the point where it can be claimed that the presence of phenomenal consciousness [7] in machines has been achieved. Simply to achieve functional consciousness [7] the agent must demonstrate voluntary control of attention, intention, action selection, motivation, goal-setting abilities, emotional intelligence, and self-regulation along with communication capability utilizing the cognitive constructs listed earlier [11]. One of the critical factors to help facilitate the implementation of the aforementioned is the selection of an appropriate modeling technique. Current literature suggests that there are two approaches for such modeling. These are a hierarchical and non-hierarchical approach [42], where a hierarchical model treats consciousness as a single reality with different levels of depth and a non-hierarchical model treats consciousness as a self-organizing multiplicity of embodied states (this allows the emergence of intelligence which is greater than the sum of its information). Non-hierarchical methods can be used for modeling emergent artificial general intelligence [42], while hierarchical methods can be used for modeling a cognitive system comprised of multiple complex subsystems interacting together in synergy to generate conscious behaviors [42]. Both approaches have strengths and weaknesses. Keeping in mind the limitations of both methods, one of the successful methods for modeling is hybrid. The hybrid approach allows the modeling of higher abstraction in a hierarchical fashion. The functional level of these abstractions consists of subsystems that may operate according to the different paradigms. The hybrid approach seems to be the most promising as it ameliorates the deficiencies associated with the other methods. Apart from bio-inspired modeling discussed above, this study also takes into account quantum inspiration. There are a number of school of thought associated with quantum inspiration explaining the functioning of brain/mind, consciousness, and intelligence. The reason for taking into account quantum inspiration is that the Universe is governed by the laws of nature [43]. The brain, mental processes along with other biological reactions, physical realities from quantum to classical world, computation, are all part of nature. If the Universe is quantum mechanical then mental processes can be described using quantum theories [43] and the mysteries of the brain, mental processes, and emergence of consciousness can be due to quantum processes or quantum computations [14][44]-[46]. Quantum mechanics is the study of subatomic particles that possess features such as superposition and entanglement, and the computation performed on them is referred to as quantum computation. Analysis of
interpretations of quantum mechanics suggests that it has the computational/mathematical potential to model cognition/consciousness. There are different schools of thought on this issue, namely: (i) quantum mind/ consciousness, (ii) quantum cognition, (iii) quantum artificial intelligence. Quantum consciousness assumes that the brain works on the principles of both neural and quantum computations [14][43][44][47]-[50]. Quantum cognition does not explicitly discuss the involvement of quantum mechanics in the function of the brain at the physical level [51]-[53]. Rather this paradigm acknowledges the mathematical and abstraction strength of quantum mechanics to model human memory, knowledge representation issues, perception, decision making and reasoning process. There are a number of useful references that expound this school of thought [15][16][54]-[58] which uses quantum inspired Neural Networks, quantum Bayes network and other quantum cognition algorithm to model complex human mental facilities. In contrast to quantum consciousness, quantum artificial intelligence is about transforming existing artificial intelligence algorithms to make them work on quantum computers [12][13]. The existence of consciousness in machines can be evaluated using both qualitative and quantitative methods. Qualitative analysis can be performed using Aleksander's axioms of conscious [59]. Qualitative analyses using Aleksander's Axiomatic postulates that an agent having depiction, imagination, attention, volition, and emotion may possess some kind of conscious activity [59]. Moreover, the quantitative analysis includes Quantum Mutual Information (QMI) and ConsScale method [60]-[63]. QMI is a method that was developed to determine the correlation between the subsystems of two quantum states where these states are not pure states [8]. ConsScale is a characterization mechanism for the evaluation of consciousness in an agent [61]. Moreno originally proposed ConsScale [61], a method that was later improved upon [62][63], as the reference method for analysis of the correlation between cognitive and conscious capability. ConsScale mainly evaluates an agent based on its behavior and architecture [62]. Following the legacy of the aforementioned studies, project NiHA (Nature-inspired Humanoid Cognitive Computing Platform for Self-aware and Conscious Agent) was initiated. In this paper, NiHA is introduced as an entity with a limited level of consciousness and general intelligence. The design and implementation strategy are also articulated and the results acquired so far presented. The results are encouraging and suggest that NiHA may possess a certain level of consciousness. This does not imply that NiHA is fully conscious or that it possesses human-level of phenomenology or general intelligence, neither do the authors intend to claim such. Indeed, NiHA is an attempt to explore the possibility of designing and implementing an artifact that falls under the domain of machine consciousness and artificial general intelligence. In this paper, Section II details NiHA's architecture, Section III deals with its applications and functionality,
Copyright (c) IARIA, 2018. ISBN: 978-1-61208-609-5
COGNITIVE 2018 : The Tenth International Conference on Advanced Cognitive Technologies and Applications
Section IV describes the test of consciousness, and Section IV documents the results and discussion. II. NATURE-INSPIRED HUMANOID COGNITIVE COMPUTING PLATFORM FOR SELF-AWARE AND CONSCIOUS AGENT (NIHA) NiHA is based on previous work related to the unified theory of the mind, `Quantum & Bio-inspired Intelligent & Consciousness Architecture (QuBIC)' (see Figure 2) and on the code-base of an agent called Johi [8]. A. Cognitive Architecture (QuBIC) of NiHA The cognitive architecture of NiHA is constructed in layers (see Figure 2). The first layer consists of physical components, while the second layer comprises of mental processes. The physical layer consists of sensors and actuators. The mental regime further consists of unconscious and conscious layers. . The unconscious layer consists of several cognitive units working together in order to regulate the involuntary and pre-programmed tasks that are essential requirements for the agent's self-regulation and optimum performance. The modules of the conscious layer are responsible for awareness, attention, intention and voluntary tasks. The conscious layer serves as the executive control of the mind and the Functional description of the whole architecture is discussed in [8].
i. Memory Units Unconscious components consist of several modules; these include memory systems, circadian clock, implicit behavior center, reflexes, decision action center, deed assessment module, seed knowledge, memory management, drives, and a synchronization unit. a) Sensory Memory Sensory Memory is responsible for the storage of sensory inputs coming from external and internal sensors. Sensed contents are collected in the implicit knowledge repository. Sensory processes apply basic filters to the sensory contents. The filters include those responsible for resizing the incoming image stream, encoding the stream as required and applying part-of-speech-tagger on the lingual contents. These sensory contents are then transferred to various unconscious modules as per the connections shown in Figure 2. b) Perceptual Associative Memory A copy of processed sensory contents is received by the Perceptual Associative Memory (PAM). PAM is responsible for the formation of an association between objects based on recognition, perception, and classification. Information is encoded for further processing, using structural and semantic analysis and collaboration between Short-Term Memory (STM) and Working Memory (WM). The percepts are then transferred to conscious memory system for further processing.
Figure 2 - NiHA's Cognitive Architecture (QuBIC) [8]
Copyright (c) IARIA, 2018. ISBN: 978-1-61208-609-5
COGNITIVE 2018 : The Tenth International Conference on Advanced Cognitive Technologies and Applications
c) Decision Execution Center The Decision Execution Center (DEC) consists of both conscious and unconscious components. It actually acts as a procedural memory. This unit consists of "how to perform a task" information which is needed mainly for motor actions. d) Short-term Memory Short-Term Memory (STM) receives and examines contents from PAM to determine what requires attention through a process called `selective attention'. The content, which does not get attention, is forgotten when short-term memory overflow occurs or content's signal is decayed. This selective attention process acts as a gateway between unconscious and conscious layers, resulting in an awareness of sensory contents. e) Working Memory Working Memory (WM) is responsible for the manipulation and processing of information during the course of the cognitive process. WM consists of the central executive that is supported by visual memory (which process visual content) and spatial memory (contains the spatial maps and information about size and location of content). ii. Long-term Memories Long-Term Memory (LTM) holds the information for longer periods. LTM is composed of episodic and semantic memories. The episodic memory contains the information related to the agent's experience of particular events, while the semantic memory contains the record of meanings, facts, and knowledge related to objects. iii. Circadian Clock, Monitoring and Control The Circadian Clock (CC) triggers several high-level and low-level drives (and ensures the optimal performance of an agent), while the Circadian Monitoring and Control (CMC) units safeguard the activities by maintaining the threshold levels. CMC also protect various conscious and unconscious cognitive modules from malfunctioning. iv. Seed Knowledge One of the major challenges when modeling artificial general intelligence is to implement safe artificial intelligence. Seed knowledge acts as apriori knowledge implemented at design time. This provides ethical norms and a set of rules that try to ensure "safe" artificial intelligence. To date, a conceptual knowledge base has been designed but not yet implemented [64]. This knowledgebase contains information relating to ethical values, laws related to daily life, rules regarding social etiquette and an understanding of the difference between right and wrong. Seed knowledge has already been implemented in NiHA [8] [64]. v. Deed Assessment Module The Deed Assessment Module (DAM) acts as the "inner self" and assesses performance with regard to the guidelines available as seed knowledge. Conscious and unconscious activities are regulated in the DAM based on predefined parameters and evolved knowledge.
vi. Emotions The Emotion Module regulates and influences the rational decisions based on basic emotions [65]. The emotions are extracted from the facial expression of a human interacting with NiHA. The extracted expressions are then used for the regulation of decisions. vii. Meta-cognition Meta-cognition performs a regulatory role of managing several conscious and unconscious modules. These modules perform numerous tasks i.e. memory management, evaluation, emotion modulation, synchronization between conscious and unconscious modules, learning management and goal setting. viii. Behavioral and Motivational System The Behavioral System is responsible for generating implicit and explicit behaviors in the agent based on current context and experiences. The Motivational System [67], inspired by the work of Manzotti [66], generates motivations in an agent in the form of goals. These goals are then used to take desired actions, along with what and how the task will be done. ix. Dreams Dreams are the most important cognitive unit. It is responsible for memory consolidation, the evolution of ideas, and creativity through the manipulation of known objects. The first stage of Dreams (i.e. memory consolidation) has been implemented in NiHA [68], and is based on Quantum Neural Network and Quantum Genetic Algorithm. x. Imagination Imagination is the manifestation of scenarios that are not directly present to the sensors. The first stage of its implementation in NiHA had the goal of emulating the imagination of a young child. Studies suggest that young children only understand simple sentences. Imaginations implemented to date simulate psychological aspects comparable with that of a four-year-old child [69]-[70]. Further exploration in this area is in progress. B. Design Rationale & Implementation Philosophy The design and implementation principles that have been utilized are based on ConsScale guidelines [60][62][63]. According to these, every conscious agent must possesses mental processes and a physical or simulated body. The mental processes in NiHA are based on QuBIC. Currently, the mental processes have been interfaced with a simulated iCub robot [71]. The simulated iCub runs on a laptop mounted on an iCreate robot platform along with Microsoft Kinect sensor and is collectively known as NiHA's body. Moreover, the integration of the Allison humanoid robot with NiHA is underway. The motivation behind this study is to address a number of philosophical and design issues not taken into account previously. These issues are associated with quantum-
Copyright (c) IARIA, 2018. ISBN: 978-1-61208-609-5
COGNITIVE 2018 : The Tenth International Conference on Advanced Cognitive Technologies and Applications
inspired modeling, synchronization of autonomous functions through the circadian clock, imaginations, dreams, personal semantics, and psychophysiological regulation of drives through the modeling of insulin and glucose interplay, seed knowledge including deed assessments. Further details on these features are documented in [8][70] The quantum inspiration is based on methods from quantum cognition and quantum artificial intelligence. The aforementioned features are complex systems and depended on one another as shown in Figure 1. Therefore, modeling of such systems requires the implementation of the cognitive architectural framework, together with a set of cognitive processes, cognitive cycles, and representation schemes [72]. The cognitive processes and cycles are inspired from pandemonium theory [1], feature-integration theory [73], unified theory of an artificial mind, global workspace theory [74] and QuBIC architectural model [8]. The neural correlates of the architectural model are summarized in [75]. These features were implemented separately due to computational and design complexity and are yet to be integrated into the architectural framework (moreover, some are still at the conceptual stage). The overall design and implementation was service oriented. These modules are implemented using different programming languages and run on Windows and Ubuntu platforms. Further details are provided in the `languages and tools' section. C. Cognitive Processes and Codelets The pandemonium theory suggests the presence of demons (mini-agents) in the mind, working in synergy to perform various cognitive tasks in parallel [1]. These cognitive tasks may range from simple identification of objects to learning and to execution of complex sensory motor controls. Generally, these demons are called codelets in cognitive computing domain [2][76][77]. According to the pandemonium theory, these codelets are an independent set of programs working in parallel to process incoming signals. Therefore, each of these codelets has well-defined responsibility and executes continuously and cyclically. Further details on codelets and their standard implementation and guidelines are available in [2][76][77]. In NiHA, the codelets were implemented as proxies to the actual implementation of the code in QuBIC architectural framework. The actual implementations are either in the form of C#'s tasks from `Task Parallel Library' (TPL) [78] or services implemented running as nodes on the network. Implementation of these codelets as TPL-tasks or services was determined based on response time and the programming language in which they were coded. Cognitive processes with short running time and programmed using C# were developed as TPL-tasks. Cognitive tasks, which were implemented on different platforms or had large execution time, for example learning codelets, were implemented as services. This division is simply a rule of thumb, and in order to keep cognitive optimization, variations can be made as per requirement and future need. Table 1 summarizes the list of codelets in NiHA.
A codelet converts internal planning to motor actions
A codelet responsible for configuration of learning
algorithms for a different type of streams.
A codelet work in collaboration with detection and
recognition codelet to assign meaning to visual
Language & Language codelet works with several other codelets
Understandin for the understanding of text and speech processes.
A codelet train system to assign meanings to sensory
Drives and Based on a high level and low drives, codelet
Motivation formulate motivation for an agent to do specific
Attention and Attention codelet bring sensory contents into the
Awareness conscious part. A codelet that attempts to train
attention on some particular kind of information.
Examples: Expectation codelet, intention codelet
A codelet that regulate the internal representations
based on emotions detected from the external
A codelet work in synergy with memory
consolidation codelet to generate dreams and
restructure memory blocks.
Circadian is just like clock used to check the mode
of NiHA every few minutes.
A codelet accountable for the restructuring of
Consolidation associated memory constructs.
Imaginations Imagination codelet works on storytelling technique.
It makes visualization based on internal
representations and current scenarios.
Deed assessment codelets evaluate agent on the basis
Assessment of ethical knowledge stored in seed knowledge.
Signaling and Codelet responsible for the interaction and
Synchronizati synchronization of numerous signals of the cognitive
Conceptually, each codelet is a part of the codelet ecosystem running on top of a cognitive module. As shown in Figure 3, a set of codelets are working on the contents of workspace module. These codelets are pre-programmed for specific tasks, for example, the conversational codelet generates a potential response after processing the lingual contents from the workspace. In parallel to this, the tagging codelet tags the current mental state with the prevailing positive and negative emotions. Responses from various codelets are compiled together by the underlying modules as chunks. These chunks are then broadcasted to relevant modules for further processing.
Figure 3 - Codelets in NiHA working in Synergy
Copyright (c) IARIA, 2018. ISBN: 978-1-61208-609-5
COGNITIVE 2018 : The Tenth International Conference on Advanced Cognitive Technologies and Applications
D. Tools and Languages used for Implementation The following programming languages were used: C#, Java, Python, and C++. The high-level abstraction of NiHA's design is shown in Figure 4. The cognitive structure (QuBIC architectural framework) is written in C# using service-oriented programming. The service orientation is done using Apache Thrift, ROS.NET and YARP.NET. Apache thrift was used as the primary tool for implementing network computing between various services in the framework. ROS.NET is used where communication between modules written in C# and Robotic Operating System (ROS) is required. Figure 4 - Proposed NiHA Cloud The software framework (YARP) for the iCub robot is written in C++ and YARP.NET is a wrapper/binder to interface YARP with C# code, which was acquired after compiling the YARP C++ code using SWIG. Python is used for natural language processing. Accord.NET and EmguCV are used for computer vision support and Accord.NET for audio processing. The conversation module was implemented using the Artificial Intelligence Markup Language (AIML). Weka.NET is used for machine learning support and MaryTTS for text to speech processing. Neo4j is a graph database used for the implementation of episodic and semantic structures. Moreover, quantum computations are performed using an in-house developed quantum computing library written in C# (see download section of the official NiHA website [79]). E. Application and Functioning of NiHA The current implementation allows NiHA to have general communication with a human operator. Moreover, it can play the popular paper-scissor-rock game. It can also construct an artistic work through storytelling processing utilizing imagination codelets. Moreover, it can construct short dreams using annotated images placed in the memory system. Details of the implementation of these applications and functioning will be published in the future but related
video demonstrations are already available to be reviewed [79]. III. TEST FOR CONSCIOUSNESS The objective of this paper is to report on progress to date on constructing NiHA, an agent that currently has limited consciousness and general intelligence. The presence of a certain amount of consciousness in an agent can be evaluated using qualitative and quantitative analyses. Qualitative methods include Aleksander's Axioms [59] and the quantitative methods, Quantum Mutual Information (QMI) [8] and ConsScale [63]. A. Aleksander and Dunmall (ADM) Axioms Aleksander and Dunmall, based on neuroscience studies, articulated the necessary properties for an agent to have indications of its own existence [59]. In this section, NiHA is assessed to determine its level of consciousness. i. Depiction Depiction is the mechanism of bringing the perceptual experiences into awareness. NiHA is able to formulate meaning from sensory information. The visual and auditory cues generate selective awareness about the existence and movement of the object(s) in an environment. Similarly, the acoustic cue in NiHA reacts by generating appropriate replies. According to Gamez [23], the integration of various perceptual signals give the sensation of something, that is out there and consequently, qualify the possibility of the depiction of an agent. The depiction is achieved in NiHA via the collaborative work of numerous codelets. ii. Imagination As mentioned earlier, in NiHA imagination is implemented in the form of the collaborative work of several codelets. So, NiHA can be credited with the existence of imagination in its primary form. Aspects of storytelling have been chosen for its initial development [69][70]. Moreover, NiHA is able to generate sensations and future aspects of known and unknown objects in the absence of any sensory information. iii. Attention The bottom-up and top-down attention mechanism is implemented in NiHA. The recall represents the top-down attention when signals are generated in working memory from long-term or medium-term memory, whereas signals coming from perceptual memory represent bottom-up attention. iv. Volition NiHA plans and simulates its actions before applying them to the environment and regulates them by means of its emotional state. v. Emotions Emotions play a vital role in manipulating the generation of goals and motivations. The Emotional Block, in NiHA,
Copyright (c) IARIA, 2018. ISBN: 978-1-61208-609-5
COGNITIVE 2018 : The Tenth International Conference on Advanced Cognitive Technologies and Applications
influences the effects of drives, decision-making, and planning. This block consists of six basic emotions [65]. Axiomatic analysis advocates the possibility of consciousness in NiHA. Moreover, the axioms suggest that the consciousness can be measured by evaluation of minimum mutual information exchange between cognitive blocks. B. Quantum Mutual Information Quantum Mutual Information (QMI) is used to measure the correlation between several sub-systems where the quantum states are not pure. Cognitive blocks can be represented as quantum states. This correlation represents the level of consciousness. For experimental analysis, the selected states are Working Memory (WM) and Perceptual Associative Memory (PAM). i. Analysis 1 Signals were transmitted from the Sensory Memory block to the Perceptual Associative Memory block, the NiHA was supposed to get the awareness of conscious stream between various unconscious blocks. ii. Analysis 2 Signals were transmitted from unconscious block (Perceptual Associative Memory) to conscious clock (Working Memory and Medium-term Memory) where NiHA was expected to get the attention of the memory contents. iii. Analysis 3 Signals were transmitted from Working Memory to the Workspace and Medium-term Memory, where NiHA was supposed to get aware of the memory contents. C. ConsScale ConsScale was proposed by Moreno [61] and later matured [62][63] for the characterization of consciousness in an agent. The correlation between cognitive and conscious skills are being measured on the basis of Cumulative Levels Score (CLS) and ConsScale Quantitative Score (CQS). The levels of consciousness in ConsScales range from 0 to 11, from disembodied to super-human. The evaluation is mainly based on its architectural framework and behavior. IV. RESULTS & DISCUSSION The experiments have been performed on NiHA to analyze its conscious activity. In this regard, using signals transmitted from the unconscious region to the conscious region, signal analysis to determine level of consciousness were carried out using Quantum Mutual Information (QMI) sharing. Measuring of mutual information sharing is an established practice [23]. i. Analysis 1 NiHA did not show any conscious activity between unconscious blocks.
ii. Analysis 2 The analysis was performed on the contents being shared between Perceptual Associate Memory (PAM) and Medium-term Memory (MTM), similarly between PAM and Working Memory (WM). This resulted in conscious activity (see Figure 5). These results were recorded for 1,000 circadian clock cycles against 25 sample streams. The level of mutual information sharing shown in Figure 5 represents the level of consciousness recorded at different data sample points. The x-axis represents the sample number and the y-axis represents the level of consciousness in terms of mutual information sharing. Figure 5 - Conscious activity between Perceptual Associative Memory and Working Memory [8]. The x-axis represents the sample number and Y- axis represents the conscious level in that sample. iii. Analysis 3 The analysis was performed between conscious modules. These modules were Working Memory, Workspace, and Medium-term Memory. The consciousness level obtained between these modules is shown in Figure 6. Figure 6 - Conscious activity between Working Memory and Workspace [8]. The x-axis represents the sample number and Y-axis represents the conscious level in that sample. iv. ConsScale based Evaluation of NiHA ConsScale was also applied in order to further clarify the class of consciousness which NiHA possess, given the current state of its implementation. A Cumulative Levels Score (CLS) of 1.27 and a ConsScale Quantitative Score (CQS) of 2.94 were obtained.
Copyright (c) IARIA, 2018. ISBN: 978-1-61208-609-5
COGNITIVE 2018 : The Tenth International Conference on Advanced Cognitive Technologies and Applications
Figure 7 - ConsScale Evaluation Method for NiHA The ConsScale result is summarized in Figure 7. It suggests that NiHA achieves a 3 on the 11 levels scale. It further proposes that by design NiHA is computationally equivalent to adaptive species, and therefore it still requires a lot of developmental work before anywhere close to achieving human level of consciousness. V. CONCLUSION The study reports an ongoing work in the development of NiHA which is based on the QuBIC architecture. NiHA is composed of several success stories, cognitive memories, dreams, imagination, machine ethics and socio-cognitive capabilities and other blackboard-like projects. Results obtained from various isolated studies imply that NiHA (as the extension of Johi) has a certain degree consciousness. The qualitative analyses suggest that all necessary conceptual and theoretical ingredients required to model a conscious agent are present. The quantitative analyses using quantum mutual information sharing suggests that at certain points in the signal stream, conscious activity was registered. This indeed supports the presence of a minimum level of consciousness. Furthermore, analysis using ConsScale suggested that this presence of consciousness can be classified as computationally equivalent to level 3 species.
ACKNOWLEDGEMENT The authors acknowledge the financial support from the Higher Education Commission of Pakistan, and the sponsorship of EDVON Pakistan to help facilitate the acquisition of the Allison Humanoid Robot. REFERENCES [1] S. Franklin, Artificial Minds, MIT Press, 1995. [2] S. Franklin, A. Kelemen and L. McCauley, "IDA: A Cognitive agent architecture." in IEEE Conference on Systems, Man and Cybernetics, 1998. [3] R. Manzotti and V. Tagliasco, "Artificial Consciousness: A discipline between technology and theoretical obstacles," Artificial Intelligence in Medicine, vol. 44, pp. 105-117, 2008. [4] P. O. Haikonen, "The Comparison of Some Cognitive Architectures," Series on Machine Consciousness, vol. 2, pp. 185-202, 2012. [5] P. Langley, "Progress and Challenges in Research on Cognitive Architectures," in Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence, San Francisco, California USA, 2017. [6] G. Buttazzo, "Artificial consciousness: Hazardous questions (and answers)," Artificial Intelligence in Medicine, vol. 44, no. 2, pp. 139-146, 2008. [7] D. J. Chalmers, "The Puzzle of conscious experience," Scientific American, vol. 273, pp. 80-86, 1995. [8] W. M. Qazi, "Modeling cognitive cybernetics from unified theory of mind using quantum neuro-computing for machine consciousness," Lahore, 2011. [9] S. Franklin, "IDA, a Conscious Artifact?" Journal of Consciousness, vol. 10, pp. 47-66, 2003. [10] D. Feil-Seifer and M. J. Mataric, "Defining Socially Assistive Robotics," in International Conference on Rehabilitation Robotics, Chicago, IL, USA, 2005. [11] A. V. Samsonovich, "On a roadmap for the BICA Challenge," vol. 1, pp. 100-107, 2012. [12] B. Aoun and M. Tarifi, "Quantum Artificial Intelligence," Quantum Information Processing, pp. ArXiv:quantph/0401124, 2004. [13] K. Sgarbas, "The Road to Quantum Artificial Intelligence," Current Trends in Informatics, pp. 469-477, 2007. [14] S. Hameroff , "The Brain is both Neurocomputer and Quantum Computer," Cognitive Science, vol. 31, no. 6, pp. 1035-45, 2007. [15] J. Busemeyer and P. Bruza, Quantum Models of Cognition and Decision, Cambridge: Cambridge University Press, 2012. [16] E. M. Pothos and J. R. Busemeyer, "Can quantum probability provide a new direction for cognitive modeling," Behavioral and Brain Sciences, vol. 36, p. 255­274., 2013. [17] Z. Wang, J. R. Busemeyer, H. Atmanspacher and E. M. Pothos, "The potential of using quantum theory to build models of cognition," Topics in Cognitive Science, vol. 5, no. 4, pp. 672-688, 2013. [18] I. Aleksander, How to build a mind : Towards machines with imagination, New York : Columbia University Press, 2001. [19] D. Friedlander and S. Franklin, "LIDA and a Theory of Mind," in Proceeding of the 2008 Conference on Artificial Genral Intelligence, 2008. [20] R. Kurzweil, The Singularity Is Near, Viking, 2006. [21] R. Manzotti and A. Chella, "Physical integration: A causal account for consciousness," Journal of integrative neuroscience, vol. 13, no. 02, pp. 403-427, 2014.
Copyright (c) IARIA, 2018. ISBN: 978-1-61208-609-5
COGNITIVE 2018 : The Tenth International Conference on Advanced Cognitive Technologies and Applications
[22] R. Sun, "Motivational Representations within a
Computational cognitive Architecture," Cognitive
Computation, vol. 1, no. 1, pp. 91-103, 2009.
[23] D. Gamez, "The Development and Analysis of Conscious
Machines," 2008.
[24] R. Manzottia and S. Jeschkec, "A causal foundation for
consciousness in biological and artificial agents," Cognitive
Systems Research, vol. 40, pp. 172-185, 2016.
[25] "The Blue Brain Project - A Swiss Brain Initiative," 2005.
[Online]. Available: [Accessed 15
January 2018].
[26] "Human Brain Project," 2013. [Online]. Available: [Accessed 15
January 2018].
[27] "DARPA and the Brain Initiative," 2013. [Online].
research/darpa-and-the-brain-initiative. [Accessed 15
January 2018].
[28] S. Boixo, V. N. Smelyanskiy, A. Shabani, S. V. Isakov, M.
Dykman, V. S. Denchev, M. H. Amin, A. Y. Smirnov, M.
Mohseni and H. Neven, "Computational Multiqubit
Tunnelling in Programmable Quantum Annealers," Nature
Communications, vol. 7, pp. 1-7, 2016.
[29] S. Gu, T. Lillicrap, I. Sutskever and S. Levine, "Continuous
Deep Q-Learning with Model-based Acceleration," in
Proceedings of the 33 rd International Conference on
Machine, New York, NY, USA, 2016.
[30] A. Globerson, N. Lazic, S. Chakrabarti, A. Subramanya, M.
Ringgaard and F. Pereira, "Collective Entity Resolution with
Multi-Focal Attention," in Proceedings of the 54th Annual
Meeting of the Association for Computational Linguistics,
Berlin, Germany, 2016.
[31] E. Whitmire, L. Trutoiu, R. Cavin, D. Perek, B. Scally, J. O.
Phillips and S. Patel, "EyeContact: Scleral Coil Eye
Tracking for Virtual Reality," in The International
Symposium on Wearable Computers (ISWC), Heidelberg,
Germany, 2016.
[32] Y. LeCun, "Facebook AI Research (FAIR)," Facebook,
[Online]. Available:
[Accessed 26 September 2016].
[33] Y.-W. E. Sung, X. Tie, S. H. Wong and H. Zeng, "Robotron:
Top-down Network Management at Facebook Scale," in
ACM Special Interest Group on Data Communication
(SIGCOMM) , Florianopolis , Brazil, 2016.
[34] N. Zeghidour, G. Synnaeve, N. Usunier and E. Dupoux,
"Joint Learning of Speaker and Phonetic Similarities with
Siamese Networks," in Interspeech, San Francisco, United
States, 2016.
[35] "Watson-IBM,"
Available: [Accessed 15 January 2018].
[36] "National Robotics Initiative (NRI)," [Online]. Available:
[Accessed 15 January 2018].
[37] K. Korosec, "Saudi Arabia's Newest Citizen Is a Robot,"
arabia/. [Accessed 30 October 2017].
[38] S. Strain, S. Kugele and S. Franklin, "The Learning
Intelligent Distribution Agent (LIDA) and Medical Agent X
(MAX)," in IEEE Symposium on Computational
Intelligence for Human-like Intelligence , Orlando, Florida,
USA, 2014.
[39] M. F. Lynch, R. Sun and N. Wilson, "CLARION as a
Cognitive Framework for Intelligent Virtual Agents," in
IVA'11 Proceedings of the 10th international conference on
Intelligent virtual agents, Reykjavik, Iceland, 2011.
[40] "IOS
Available: [Accessed 15 January 2018].
[41] "Cortana
[Accessed 15 Januray 2018].
[42] R. Bosnak, "Hierarchical and Nonhierarchical Models of
Consciousness: Commentary on Paper by Hilary Hoge,"
Psychoanalytic Dialogues, vol. 18, no. 1, p. 27­41, 2008.
[43] D. Song, "Quantum Theory, Consciousness, and Being,"
NeuroQuantology , vol. 6, no. 3, pp. 272-277, 2008.
[44] J. Acacio de Barros and P. Suppes, "Quantum Mechanics,
interference, and the Brain," Journal of Mathematical
Psychology, vol. 53, pp. 306-313, 2009.
[45] S. Hameroff, "Quantum walks in brain microtubules--a
biomolecular basis for quantum cognition?," Top Cognitive
Science, vol. 6, no. 1, pp. 91-97, 2014.
[46] S. Hameroff, T. Craddock and J. Tuszynski, "Quantum
effects in the understanding of consciousness," Journal of
Integrative Neuroscience, vol. 13, no. 2, pp. 229-252, 2014.
[47] I. J. Thompson, "Quantum Mechanics and Consciousness: A
Causal Correspondence Theory," in The Second Mind and
Brain Symposium, London, 1990.
[48] G. Globus, "Quantum Consciousness is Cybernetics,"
Psyche, vol. 2, no. 12, 1995.
[49] M. K. Samal, "Speculations on a Unified Theory of Matter
and Mind," in Science and Metaphysics: A Discussion on
Consciousness and Genetics, Bangalore, India, 2001.
[50] R. K. Pradhan, "Subject-Object Duality and States of
Consciousness: A Quantum Mechanical Approach,"
NeuroQuantology, vol. 8, no. 3, p. 262-278, 2010.
[51] D. Aerts and S. Sozzo, "Modeling Concept Combinations in
a Quantum-Theoretic Framework," in Advances in
Cognitive Neurodynamics (IV), 2014, pp. 393-399.
[52] S. Sozzo, "A quantum probability explanation in Fock space
for borderline contradictions," Journal of Mathematical
Psychology, vol. 58, pp. 1-12, 2014.
[53] D. Aerts and M. Sassoli de Bianchi, "The unreasonable
success of quantum probability I: Quantum measurements as
uniform fluctuations," Journal of Mathematical Psychology,
vol. 67, pp. 51-75, 2015.
[54] A. Khrennikov, "Quantum-like brain: "Interference of
minds"," Biosystems, vol. 84, no. 3, pp. 225-241, 2006.
[55] P. Bruza, K. Kitto, D. Nelson and C. McEvoy, "Is there
something quantum-like about the human mental lexicon?,"
Journal of Mathematical Psychology, vol. 53, no. 5, pp. 362-
377, 2009.
[56] J. A. de Barros and P. Suppes, "Quantum mechanics,
interference, and the brain," Journal of Mathematical
Psychology, vol. 53, no. 5, pp. 306-313, 2009.
[57] D. Aerts, J. Broekaert, L. Gabora and S. Sozzo, "Quantum
structure and human thought," Behavioral and Brain
Sciences, vol. 36, no. 3, pp. 274-276., 2013.
[58] S. Kak, "Biological memories and agents as quantum
collectives," NeuroQuantology, Vol. 11, pp. 391-398, 2013.
[59] I. Aleksander and B. Dunmall, "Axioms and Tests for the
Presence of Minimal Consciousness in Agents," Journal of
Consciousness Studies, vol. 10, 2003.
[60] R. A. Moreno, "ConsScale Generic Calculator," ConsScale,
January 2016. [Online]. Available: http://www.conscious- [Accessed 15 January
Copyright (c) IARIA, 2018. ISBN: 978-1-61208-609-5
COGNITIVE 2018 : The Tenth International Conference on Advanced Cognitive Technologies and Applications
[61] R. A. Moreno, A. L. Espino and A. S. d. Miguel, "ConsScale: A Plausible Test for Machine Consciousness?," in Nokia Workshop on Machine Consciousness, 2008. [62] R. Arrabales, A. I. L. Espino and A. S. d. Miguel, "ConsScale: A Pragmatic Scale for Measuring the Level of Consciousness in Artificial Agents," Journal of Consciousness Studies, vol. 17, no. 3-4, pp. 131-164, 2010. [63] R. Arrabales, A. Ledezma and A. Sanchis, "ConsScale FPS: Cognitive Integration for Improved Believability in Computer Game Bots," in Believable Bots, Springer, Berlin, Heidelberg, 2012, pp. 193-214. [64] S. Gull, "Development of Seed Knowledge Representation Technique and Method for the Transformation of Human Explicit Knowledge to Machine's Instinctive Knowledge," Department of Computer Science, Government College University, Lahore, 2014. [65] P. Ekman, "Basic Emotions," in The Handbook of Cognition, M. Power and T. Dalgleish, Eds., Sussex, John Wiley & Sons, Ltd, 1999, pp. 45-60. [66] R. Manzotti and V. Tagliasco, "From behaviour-based robots to motivation-based robots," Robotics and Autonomous Systems, vol. 51, no. 2-3, pp. 175-190, 2005. [67] S. A. Raza, "Ontogenetic & phylogenetic control in conscious Robots," Department of Computer Science, Government College University, Lahore, 2010. [68] M. Gillani, "Machine Dreams: A Quantum Computational Approach to Model Dreams for QuBIC based Cybernetic Agents.," Lahore, 2011. [69] S. T. S. Bukhari, "Modeling and Simulation of EmotionBased Imagination in Cognitive Machines," Lahore, 2014. [70] S. T. S. Bukhari, A. Kanwal and W. M. Qazi, "Machine Imagination: A Step Towards the Construction of Artistic World Through Storytelling," in 2nd EAI International Conference on Future Intelligent Vehicular Technologies,(2017), Islamabad, 2018, in-press. [71] "iCub - an open source cognitive humanoid robotic platform," 2004. [Online]. Available: [Accessed 15 January 2018]. [72] S. Franklin and M. Ferkin, "An Ontology for Comparative Cognition: A Functional Approach," Comparative Cognition & Behavior Reviews, vol. 1, pp. 36-52, 2006. [73] A. M. Treisman and G. Gelade, "A Feature-Integration Theory of Attention," Cognitive Psychology, vol. 12, pp. 97136, 1980. [74] B. J. Baars, "In the Theatre of Consciousness : Global Workspace Theory, A Rigorous Scientific Theory of Consciousness.," Journal of Consciousness Studies, vol. 4, no. 4, pp. 292-309, 1997. [75] W. M. Qazi, S. T. S. Bukhari, J. A. Ware and A. Athar, "Neural Coorelates," Project NiHA, 2018. [Online]. Available: [Accessed 15 January 2018]. [76] A. L. Paraense, K. Raizer, S. M. Paula, E. Rohmer and R. R. Gudwin, "The cognitive systems toolkit and the CST reference cognitive architecture," Biologically Inspired Cognitive Architectures, vol. 17, pp. 32-48, 2016. [77] S. Franklin and F. P. Jr., "The Lida Architecture: Adding New Modes of Learning to an Intelligent, Autonomous, Software Agent," in Integrated Design and Process Technology,, Society for Design and Process Science, 2006. [78] "Task Parallel Library (TPL)," Microsoft, 03 March 2017. [Online]. Available: [Accessed 15 January 2018].
[79] Project NiHA, 2018. [Online]. Available: [Accessed 15 January 2018].
Copyright (c) IARIA, 2018. ISBN: 978-1-61208-609-5

WM Qazi, JA Ware

File: niha-a-conscious-agent.pdf
Title: Paper Title (use style: paper title)
Author: WM Qazi, JA Ware
Author: IEEE
Published: Fri Feb 2 08:09:32 2018
Pages: 10
File size: 0.75 Mb

Who thought of that, 2 pages, 0.17 Mb

The Sea-Change, 21 pages, 0.07 Mb

, pages, 0 Mb

, pages, 0 Mb
Copyright © 2018