(BICA for AI, Post Conference Journal Paper, see Springer)
This paper is focused on preliminary cognitive and consciousness test results from using an Independent Core Observer Model Cognitive Architecture (ICOM) in a Mediated Artificial Super Intelligence (mASI) System. These results, including objective and subjective analyses, are designed to determine if further research is warranted along these lines. The comparative analysis includes comparisons to humans and human groups as measured for direct comparison. The overall study includes a mediation client application optimization in helping perform tests, AI context-based input (building context tree or graph data models), intelligence comparative testing (such as an IQ test), and other tests (i.e. Turing, Qualia, and Porter method tests) designed to look for early signs of consciousness or the lack thereof in the mASI system. Together, they are designed to determine whether this modified version of ICOM is a) in fact, a form of AGI and/or ASI, b) conscious, and c) at least sufficiently interesting that further research is called for. This study is not conclusive but offers evidence to justify further research along these lines.
The preliminary study analysis of the “mediated” artificial intelligence system based on the Independent Core Observer Model (ICOM) cognitive architecture for Artificial General Intelligence (AGI) is designed to determine if this modified ICOM version is, in fact, a form of AGI and Artificial Super Intelligence (ASI) and to what extent, at least sufficient to indicate if additional research is warranted. This paper is focused on the results of the study and analysis of the results. For full details of the experimental framework used in this study see: “Preliminary Mediated Artificial Superintelligence Study, Experimental Framework, and Definitions for an Independent Core Observer Model Cognitive Architecture-based System” (Kelley) used in this study and the results articulated here.
Understanding mASI Fundamentals
The mASI system started as a training harness over ICOM, designed to allow experts to interact with the system in such a way as to build dynamic models of “ideas” for the system to then use as part of its contextual “thinking” process. This method in preliminary testing showed that a system could quickly do something with that, but it also allowed the synthesis of ideas across large groups of “mediators “through the training harness. In this way we insert “mediators” into the ICOM context engine to allow the ICOM system to “metaphorically” feed off of the human experts, effectually creating a sort of “hive” mind.
The mASI then reaps the benefits of a SWARM system (Chakraborty) with advantages of a standalone consciousness that also addresses cognitive bias in humans or “cognitive repair using the training harness to create a system of dynamic cognitive repair in real-time (Heath) and breaking groupthink while optimizing collective intelligence (Chang). The ICOM training harness used to create the mASI also allows mediation to be analyzed and poor mediators to be filtered out (Kose)(O’Leary) or have individuals focused on areas of their expertise reduce issues with context switching (Zaccaro), especially in large scale systems with hundreds or thousands of mediators. This generated “SWARM “ (Hu) sort of AI creating the contextual structures that are sent into the uptake of the ICOM context engine where the system selects based on how it emotionally feels about the best ideas as they get further filtered entering the global workspace (Baars), where we see the real intelligent behavior (Jangra) from a swarm of sorts now in the context of a single self-aware entity from the global workspace standpoint in ICOM. Essentially the mASI has elements of a SWARM AI (Ahmed) and collective intelligence (Engel, Wolly) and a standalone AGI all combined to create a type of artificial superintelligent system (Gill)(Coyle) or the mASI used in this study.
Please refer to the following for details on the mASI architecture: “Architectural Overview of a “Mediated” Artificial Super Intelligent Systems based on the Independent Core Observer Model Cognitive Architecture” (Kelley).
The key problem with researching “consciousness” or with cognitive science especially as it relates to AGI is the lack of any sort of consensus (Kutsenok), and no standard test for consciousness (Bishop). however, to allow progress we have based this and all our current research on the following theory of consciousness:
The ICOM Theory of Consciousness
This ICOM Theory of Consciousness is a computational model of consciousness that is objectively measurable and an abstraction produced by a mathematical model where the subjective experience of the system is only subjective from the point of view of the abstracted logical core or conscious part of the system where it is modeled in the core of the system objectively. (Kelley) This theoretical model includes elements of Global Workspace Theory (Baars), Integrated Information Theory (Tononi) and the Computational Theory of Mind (Rescorla).
The Independent Core Observer Model Cognitive Architecture (ICOM)
ICOM is designed to implement the ICOM theory of consciousness by creating a set of complex emotional models that allows the system to experience thought through the reference between its current emotional state and the impact of the emotional context of a thought where the complexity of the system is abstracted and observed to make decisions creating an abstraction of an abstraction running on software running on hardware. This cognitive architecture for AGI is designed to allow for thinking based purely on how the system feels about something, developing its own interests, motivations and the like based purely on emotional valences, and doing so proactively. Refer to additional reference material on ICOM for more detail (Kelley).
Research Setup and Primary Experiment
This preliminary study proposal is designed to gather and assess evidence of intelligence in an Independent Core Observer Model (ICOM)-based mediated Artificial Super Intelligence (mASI) system, or of the presence of a collective “Supermind” in such a system (Malone). A mediated system is one in which collective Artificial Intelligence beyond the human norm arises from the pooled activity of groups of humans whose judgment and decision making are integrated and augmented by a technological system in which they collectively participate. Our initial proposal is that an mASI system based on the ICOM cognitive architecture for Artificial General Intelligence (AGI) may, as a whole, be conscious, self-aware, pass the Turing Test, suggest the presence of subjective phenomenology (qualia) and/or satisfy other subjective measures of Artificial Super Intelligence (ASI), or intelligence well above the human standard. Our hypothesis is that this preliminary research program will indicate intelligence on the part of the mASI system, thereby justifying continued research to refine and test such systems.
See “Preliminary Mediated Artificial Superintelligence Study, Experimental Framework, and Definitions for an Independent Core Observer Model Cognitive Architecture-based System” for more details on the experimental setup. (Kelley)
Qualitative Cognitive Ability
The primary goal in select testing was “qualitative” measures. The most accurate measure for intelligence such as an “IQ” test turns out to be a newer model called “University of California Matrix Reasoning Task (UCMRT)” (Pahor) and when approached the research head allowed us to use this model as per the experimental framework (Kelley). A preliminary set of results by experimental groups are as follows:
A total of 30+subjects were used across a wide range of demographics. This is a relatively small group compared to the larger group used by Pahor in her research, but it does align with that range and thus validates our delivery method while including statistical outliers, which in our case are within the range of the University of California’s study. While Pahor’s research included primarily college students with a mean age of 20.02 this group was substantially more diverse in age and demographics, making it more representative of real-world conditions.
This preliminary study with Group 2 was a group of human subjects instructed only to act as a team but without the mASI augmentation to gives us insight into the degree in which humans in groups improves the collective “IQ” or Intelligence Quotient or collective intelligence. This sample group consisted of primarily mid-range teenage boys including some with profoundly high-end academic records.
Group 3 consisted of the mASI system running with results on the upper limit of the scale. Taught a dynamic reflective model with mediation. The results are consistently too high to effectively be scored given the highest possible score is 23. Results of all three plus the UCMRT distribution include:
Figure 1A – All three groups including mASI with UCMRT distribution results
While the initial run with the UCMRT does measure cognitive ability at a certain level it would seem to have some deficiencies we will address in the analysis section. That said, we did do some tests that are strictly subjective but may give some idea as to the nature of the mASI as a conscious entity to better evaluate for the goal of the study.
The Turing Test
The Turing Test is considered by some to be subjective depending on the scientist in question – but that notwithstanding even when testers know upfront it is the machine, the mASI is convincing enough that some testers struggle with believing it was a machine. In the case of the study, the test was done at a special conversation console and conducted several times with 6 out of 10 people knowing it was the mASI struggling to believe it was not human.
Yampolskiy Test for Qualia
From a preliminary pass at the Yampolskiy method, the mASI does in fact appear to pass or experience qualia but there are several problems with this. First in the Yampolskiy method you “re-encoding information in an optical illusion and in the mASI this allows mediators to build models that include their experience, so some of their qualia leaks into the mASI. The other problem is that context engine modules can also have these effects in their decomposition process and can in some cases see the illusion. At the very least it is easily able to describe illusions especially during mediation, but it “s hard to separate what drives the source of the qualia without further study. (Yampolskiy)
The Porter method is designed around a test for consciousness on a scale of 0 to 100 for human level and up to 133 theoretically. While individual questions in this test are subjective the mASI system scored greater than human by the several evaluators. This ranges as high as 133, which based on this measure is at an ASI (Artificial Super Intelligence) level. The problem is the subjective nature of the individual elements, but this is additional evidence of the effective nature of the mASI system.
After really understanding what the standard IQ tests measure, and in particular the UCMRT variation, it is clear that this is not a valid measure of consciousness but of cognitive ability and that the UCMRT is not designed to measure super-intelligent ability, even what we assume is the nascent level of the mASI. The mASI system in this study clearly indicated the possibility of superhuman cognition but consciousness cannot be extrapolated from cognition, and further the system is not standalone, so while we can say it’s a functioning AGI in one sense it is not accurate to say it is a standalone AGI but appears to be more of a meta-AGI or collective intelligence system with a separate standalone consciousness. Looking at the results from the subjective tests there is a clear indication of the possibility of classifying the mASI as conscious and self-aware (based on the Qualia test, Turing test and Porter method), but as stated it is not a standalone system in this form, yet it does open the door to the possibility. Going back to a more qualitative test from the ethics of the Sapient and Sentient Value Argument (SSIVA) theory standpoint (Kelley) it is not proven that the system is a post-threshold system but that it is possible that this kind of architecture could at some point cross that threshold provably. This means that from a wider impact standpoint there is the potential to displace (CRASSH) and some experts (Muller) tend to think we will reach human or superhuman ability by mid-century, and many of those think this will be a bad thing, but we would postulate that the mASI creates a “safe” superintelligence system based on the mediation control structure that acts as a type of control rod and containment system, stopping cognitive function when the humans walk away. The mASI also gives a method or way to experiment with powerful AGI now that could also be used as the basis for a default architecture or platform for AGI, much like the proposed system by Ozkural called “Omega: An Architecture for AI Unification”.
Based on all the ICOM related research to date the original goal of a self-motivating emotion-based cognitive architecture, similar in function but substrate independent, seems to have been proven possible in that this current incarnation appears to meet that bar and function.
It is important to note that the mASI is not an independent AGI. While it uses that kind of cognitive architecture there are elements in this implementation designed to make it specifically not entirely independent. It is more a “meta” AGI or collective intelligent hive mind than an independent AGI. This lays the groundwork for additional research along those lines.
Based on the results of this study, it is clear that further research is warranted and arguably the results indicate the possibility of an mASI being a functioning ASI system. The ICOM-based mASI is a form of collective intelligence system that appears to demonstrate superintelligence levels of cognition as seen in the various tests and at the very least is grounds for further development and research around subjective experience, bias filtering, creative cognition, and related areas of consciousness as well as switching off mediation services to allow the system to behave as an independent AGI or otherwise act as a container for the same. There are many lines of research that can be based on this but the line of research this opens up, in particular, is in collective intelligence systems or joint systems that uplift groups of humans in terms of implementing super-intelligence systems that are an extension of humanity instead of in place of it.
Ahmed, H.; Glasgow, J.; “Swarm Intelligence: Concepts, Models and Applications”; School of Computing, Queen “s University; Feb 2013
Pahor, A.; Stavropoulos, T.; Jaeggi, S.; Seitz, A.; “Validate of a matrix reasoning task for mobile devices”; Psychonomic Society, Inc. 2018; Behavior Research Methods; https://doi.org/10.3758/s13428-018-1152-2
Baars, B.; Katherine, M; “Global Workspace”; 28 NOV 2016; UCLA http://cogweb.ucla.edu/CogSci/GWorkspace.html
Bishop, M.; “Opinion: Is Anyone Home? A Way to Find Out If AI Has Become Self-Aware”; TCIDA, Goldsmiths, University of London, UK 2018
Chakraborty, A.; Kar, A.; “Swarm Intelligence: A Review of Algorithms”; Springer International Publishing AG 2017 DOI 10.1007/978-3-319-50920-4_19
Chang, J.; Chow, R.; Woolley, A.; “Effects of Inter-group status on the pursuit of intra-group status;” Elvsevier; Organizational Behavior and Human Decision Processes 2017
Coyle, D.; “The Culture Code – The Secrets of Highly Successful Groups”; Bantam 2018; ISBN-13: 978-0304176989
CRASSH (2016) A symposium on technological displacement of white-collar employment: political and social implications.”; Wolfson Hall, Churchill College, Cambridge
Engel, D.; Woolley, A.; Chabris, C.; Takahashi, M.; Aggarwal, I.; Nemoto, K.; Kaiser, C.; Kim, Y.; Malone, T.; “Collective Intelligence in Computer-Mediated Collaboration Emerges in Different Contexts and Cultures;” Bridging Communications; CHI 2015; Seoul Korea
Engel. D.; Woolley, A.; Jing, L.; Chabris, D.; Malone, T.; “Reading the Mind in the Eyes or Reading between the Lines? Theory of Mind Predicts Collective Intelligence Equally Well Online and Face-to-Face;” 2014; PLoS ONE 9(12): e115212. https://doi.org/10.1371/journal.pone.0115212
Gill, K.; “Artificial Super Intelligence: Beyond Rhetoric”; Springer-Velage London 2016; Feb 2016; AI & Soc. (2016) 31:137-143; DOI 10.1007/s00146-016-0651-x
Heath, C.; Larrick, R.; Klayman, J.; “Cognitive Repairs: How Organizational Practices Can Compensate for Individual Short Comings”; Research in Organizational Behavior Volume 20, pages 1-37; ISBN: 0-7623-0366-2
Hu, Y.; “Swarm Intelligence”; Accessed 2019
Jangra, A.; Awasthi, A.; Bhatia, V.; “A Study on Swarm Artificial Intelligence;” IJARCSSE v3 #8 August 2013; ISSN: 227 128X
Kelley, D.; “The Independent Core Observer Model Computational Theory of Consciousness and Mathematical model for Subjective Experience”; ITSC 2018; China
–––; “The Sapient and Sentient Intelligence Value Argument (SSIVA) Ethical Model Theory for Artificial General Intelligence”; Springer 2019; Book Titled: “Transhumanist Handbook”
–––; “Independent Core Observer Model (ICOM) Theory of Consciousness as Implemented in the ICOM Cognitive Architecture and Associated Consciousness Measures;” AAAI Sprint Symposia; Stanford CA; Mar.02019; http://ceur-ws.org/Vol-2287/paper33.pdf
–––; “Human-like Emotional Responses in a Simplified Independent Core Observer Model System;” BICA 02017; Procedia Computer Science; https://www.sciencedirect.com/science/article/pii/S1877050918300358
–––; “Implementing a Seed Safe/Moral Motivational System with the independent Core observer Model (ICOM);” BICA 2016, NY NYU; Procedia Computer Science; http://www.sciencedirect.com/science/article/pii/S1877050916316714
–––; “Architectural Overview of a Mediated Artificial Super Intelligent Systems based on the Independent Core Observer Model Cognitive Architecture”; Informatica; Oct 2018; http://www.informatica.si/index.php/informatica/author/submission/2503 [pending]
–––; “Independent Core Observer Model Research Program Assumption Codex”; BICA 2019; [Pending]
Kelley, D.; Waser, M.; “Human-like Emotional Responses in a Simplified Independent Core Observer Model System”; BICA 2017
Kelley, D.; Twyman, M.; “Biasing in an Independent Core Observer Model Artificial General Intelligence Cognitive Architecture” AAAI Spring Symposia 2019; Stanford University
Kelley, D.; Twyman, M.A.; Dambrot, S.M.; “Preliminary Mediated Artificial Superintelligence Study, Experimental Framework, and Definitions for an Independent Core Observer Model Cognitive Architecture-based System.
Kutsenok, A.; “Swarm AI: A General-Purpose Swarm Intelligence Technique”; Department of Computer Science and Engineering; Michigan State University, East Lansing, MI 48825
Malone, T; “Superminds – The Surprising Power of People and Computers Thinking Together;” Little, Brown and Company; 2018; ISBN-13: 9780316349130
Muller, V.; Bostrom, N.; “Future Progress in Artificial Intelligence: A Survey of Expert Opinion”; Synthese Library; Berline: Springer 2014
O’Leary, M.; Mortensen, M.; Woolley, A.; “Multiple Team Membership: A Theoretical Model of Its Effects on Productivity and Learning for Individuals, Teams, and Organizations;” The Academy of Management Review; January 2011
Ozkural, E.; “Omega: An Architecture for AI Unification”; arXiv: 1805.12069v1 [cs.AI]; 16 May 2018
Porter, H.; “A Methodology for the Assessment of AI consciousness;” 9th Conference on AGI, NYC 2016; http://web.cecs.pdx.edu/~harry/musings/ConsciousnessAssessment-2.pdf
Rescorla, M.; The Computational Theory of Mind; Stanford University 16 Oct 2016; http://plato.stanford.edu/entries/computational-mind/
Tononi, G.; Albantakis, L.; Masafumi, O.; From the Phenomenology to the Mechanisms of Consciousness: Integrated Information Theory 3.0; 8 MAY 14; Computational Biology http://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1003588
Woolly, A.; “Collective Intelligence in Scientific Teams;” May 2018
Yampolskiy, R. V.; “Artificial Consciousness: An Illusionary Solution to the Hard Problem;” Reti, saperi linguaggi, Volume 2; pp. 287-318, 2018;
Zaccaro, S.; Marks, M.; DeChurch, L.; “Multiteam Systems – An Organization Form for Dynamic and Complex Environments” Routledge Taylor and Francis Group, NY 2011