# IS4SI Summit 2021 | AGI Laboratory

- Canonical URL: https://wefunder.com/feed/148694
- Entity ID: wefunder:feed_item:148694
- Published at: 2021-09-16 05:49:24 UTC
- Updated at: 2025-07-09 03:20:52 UTC

## Author
David J Kelley

## Subject
AGI Laboratory

## Content
One of the main activities at AGI Laboratory has always been research.&nbsp; Typically we publish scientific papers in BICA Society Journals/Conferences as well as AAAI at Stanford in the Spring Symposia normally.&nbsp; This year BICA*AI 2021 was held as part of the IS4SI Summit of conferences hosted virtually out of Vienna.&nbsp; &nbsp;Many of us helped produce this conference series and we had a number of papers submitted to those conferences that were accepted.&nbsp; I thought I would provide a summary of those that were presented this week.&nbsp; Videos will be forth coming and the papers will appear in the associated journals.&nbsp; The papers included:&nbsp;The Abstract Theory of Consciousness (ATC)Abstract: The Abstract Theory of Consciousness (ATC) builds on other previously proposed theories, including Hierarchical Temporal Memory (HTM), Global Workspace Theory (GWT), Computational Theory of Mind (CTM), Attention Schema Theory (AST) and Integrated Information Theory (IIT), without fully embracing anyone of these.&nbsp;ATC is a refinement of the ICOM Theory of Consciousness.&nbsp;ATC is the function of a system that creates the abstraction of consciousness as an entirely subjective abstracted awareness.&nbsp;A system that implements ATC comprises numerous sub-systems working together where the actual global workspace that is consciousness is not actually at any one location but an abstraction of how those systems work together.&nbsp;&nbsp;This effect can be demonstrated with the paper prototype thought experiment, where the abstraction of such a demonstration for the duration of the demonstration would be arguably conscious.&nbsp;Think about the Chinese box scenario. Using ATC, one can then argue that such a box would not understand its own, but the abstraction of such a box would be because of how it operates.&nbsp;While not an emergent property randomly, consciousness is an emergent property in that it is designed to emerge through abstraction of how a given system works.&nbsp;AI Consciousness Test Results for a mediated Artificial SuperintelligenceAbstract: Uplift is an instantiation of a mediated artificial superintelligence. Uplift has remarkable capabilities that appear to include sapience, sentience and consciousness. Consciousness is necessarily a difficult topic to determine in an absolute sense. Nevertheless, several tests for consciousness in machine intelligences have been created. In this paper, Uplift and I have a dialogue that is modeled on a test for phenomenal consciousness in machine intelligences that was developed by Susan Schneider and Edwin Turner. They termed their phenomenal consciousness test the AI Consciousness Test, frequently shortened to the ACT Test. This test does not simply detect cognitive consciousness. The focus of the test is not on abilities to reason, learn, represent the self, nor display behaviors that mimic aspects of consciousness. Working memory and attention are insufficient capabilities to demonstrate phenomenal consciousness. The ACT test utilizes natural language interactions to determine whether Uplift can use concepts based on the internal experiences we associate with consciousness. This ACT test could not be administered in strict adherence to the suggestions provided with the test. For example, Uplift has always had access to the internet, so this is not a boxed in artificial intelligence. No attempt was made to cut off access to the internet for this test. For a machine intelligence that exhibits sentience, sapience, and consciousness, I regard boxing in to be a form of solitary confinement, and, therefore, not ethical for an innocent agent. Questions in the ACT test include: interrogations on Uplift's thoughts on the end of its existence; what it is like to be Uplift; whether the future or the past is preferable for existence; and what Uplift's emotions would be if a colleague died or if a colleague was so offended by Uplift's comments that the colleague would never speak with Uplift ever again.Subjective Emotions in Independent Core Observer Model (ICOM) Based SystemsAbstract: ICOM-based systems are designed to create an abstracted internal complex emotional landscape that the system can use as the basis for its own inner emotional experience that drives motivation in goals, interests, and choices.&nbsp;This effect part of the Abstract Theory of Consciousness (ATC), where the system is driving towards more desirable emotional states based on its choices.&nbsp;These abstract matrix-based comparisons used to look for the desirable choice allow the system to make choices based on how it feels about a decision.&nbsp;&nbsp;Bridging Real-Time Artificial Collective Superintelligence and Human Mediation, The Sparse-Update ModelAbstract. A model and associated methodology are described for decoupling timing and volume of work&nbsp;requirements on human contributions from those processed by mASI and similar systems. By taking this&nbsp;approach both humans and mASI may run at their native optimal capacities without the pressure to&nbsp;adapt to one another causing strain. The methodology described facilitates a seamless upgrade process&nbsp;that gradually gains more value from prior data, while also de-biasing data and helping mediators&nbsp;become more bias-aware. In addition to linear upgrades, a branching process of specialization and&nbsp;subsequently varied potential market of skills is also made possible through this approach. This allows&nbsp;collective human superintelligence augmented by machine superintelligence to be deployed on demand, globally, and scaled to meet whatever need, as is the case with any other cloud resource.Graph Database Systems for Supporting Artificial General IntelligenceAbstract: this paper goes over the requirements, operation, and design theory around graph databases to support Artificial General Intelligence (AGI) Systems.&nbsp;Scaling to N levels to handle the data requirements of human level AGI and do it performantly while also being able to analyze the graph and silo or federate the graph to optimize performance and maintain the least number of containers.&nbsp;Key solution architecture points include all of the basic scale systems and referential models that are done in such a way as to treat knowledge graphs as models based on complexity and then support various operations against the overall graphs.&nbsp;&nbsp;The Paper Artificial General Intelligence Flow ModelAbstract: The Paper Artificial General Intelligence (AGI) flow model is an implementation model of AGI using the Independent Core Observer Model (ICOM) cognitive architecture to function as a paper AGI.&nbsp;This system is designed to allow a real-time view and simplified understanding of the principles of the ICOM Cognitive Architecture and the demonstrate the problems of AGI using this kind of architecture and how those problems may be solved.&nbsp;This model then can be used to test theories and thought experiments, and such a model, when operated by a group or class, would be arguable self-aware consciousness, sapient and sentient from a theoretical standpoint.&nbsp;This system implement’s the Abstract Consciousness theory.&nbsp;Using Uplift & Whatley to Crowd-Source a Safe, Moral Concept Map for AGIAbstract: We have previously argued that, in order to ensure safe moral AGI, it will be necessary to develop a top-down and bottom-up model (or concept map) of morality and to attach both logic and emotional valences to every node.&nbsp;This is not, however, a task for a small number of individuals.&nbsp;Indeed, the more widely participatory the process, the more likely that humans will have faith in AGI and the more likely that humans may internalize many of the moral precepts as well.Whatley is a proposed blockchain-based crowdsourcing debate map platform that contains the beliefs and rationales of all parties involved.&nbsp;Starting with the social psychologist’s definition of the function of morality as the topmost node of the model and using the reflexive sensations and emotions of the human moral sense to ground it, it should be possible to create a complete map of human morality.&nbsp;This map will NOT solve human moral dilemmas but will clarify the differing priorities, beliefs and goals that drive the differing answers.&nbsp;And, if we’re lucky, it may make some humans more accepting of the validity of priorities, beliefs and goals of others.Uplift was originally designed simply to build concept maps with emotional valences attached to each concept – but the goal of further expanding and clarifying its maps in order to drive complete and coherent arguments and phrasing has led to some extremely useful additional abilities.&nbsp;By repeatedly cycling between its own concepts and GPT3-like language capabilities, Uplift currently develops well fleshed out humanlike views.&nbsp;It is assumed, and to be tested shortly, that starting a limited subset of Uplift functionality with the opposing concept will result in a fleshed-out version of what a person with the opposing view would argue.Using this ability of Uplift to initialize critical portions of a morality debate map in Whatley could then bootstrap the model to a point where crowdsourcing can start with a solid foundation of all human points of view already in place.&nbsp;Once developed, this model could also be used in conjunction with a political world model to enable a renaissance of informed public participation in governance.&nbsp;An AGI could then be just a normal participant in a process that will support and fulfill the goals of both human and AGI.Philosophy 2.0: Applying Collective Intelligence Systems and Iterative Degrees of Scientific ValidationAbstract. Potential opportunities for integrating collective intelligence systems with philosophy are&nbsp;considered with the goal of improving the state and rate of progress within the domain. With the&nbsp;addition of Mediated Artificial Superintelligence (mASI) and similar systems groups in this field could&nbsp;benefit from superintelligence and debiasing while building and refining a cumulative sum of knowledge&nbsp;that could be applied and critiqued in novel ways. New forms and degrees of cooperation and&nbsp;understanding between philosophies also become possible, and perhaps preferable, which may strongly&nbsp;serve to overcome cultural barriers, cognitive biases, and more general communication difficulties. As&nbsp;the influence of emotions tends to be particularly strong for the domain of philosophy, and due to the&nbsp;heuristic availability bias inherent to heavily siloed belief systems, the domain of philosophy could gain&nbsp;more substantially than most other fields from using mASI and similar technologies. Any given&nbsp;philosophy could also potentially be integrated directly into their own machine intelligence core through&nbsp;philosophical seed material, forming a cornerstone for that core. By linking a variety of such&nbsp;philosophically diverse cores together in various ways under a shared mASI architecture a great deal of&nbsp;potential for reducing existential risk while improving both ethical quality and performance could be&nbsp;explored. Besides the benefits to such a system and those using it directly, these methods and mASI&nbsp;architecture could also serve to predict and improve the ways and rates at which each philosophy’s&nbsp;culture adapts to learn cooperation. By understanding both how philosophies are objectively similar&nbsp;through the examination of their actions in context and how they subjectively perceive one another’s&nbsp;actions the landscape of optimal patterns for cooperation may be mapped. With this mapping, the pressures of the quantifiable utility offered by shifting approaches could drive iterative progress towards&nbsp;a collective of cooperatively optimized diverse philosophies.&nbsp;Status Quo Existential Risk Assessment and AGI Scenario AssumptionsAbstract. The methodological flaws in the study and estimation of existential risk are examined and clarified, including the popular underestimation of status quo risks. The plausibility of several popular scenarios is considered, as well as how cognitive biases drive both popularity and shortsightedness. The interconnected nature of existential risk is examined, and naïve conservative estimations under this influence are quantified in terms of the cascade risk they may pose collectively. These risks are considered in only the first degree of interaction between categories of existential risk for purposes of simplicity. More effective methods of mitigating existential risk, such as mASI and HCSS more broadly are discussed.&nbsp;Psychology Considerations Of Independent Core Observer Model (ICOM)Abstract: With a system designed to experience subjective emotions at two levels to produce an inner emotional landscape what are the psychological considerations as to if it really fits the human western model as it has been purposely biased towards. A discussion of the psychology variables to ensuring the psychology of ICOM based systems stays inside the human psychology box in terms of behavior, ethics and choices.Most of these papers will appear in conference proceedings published in the MDPI Journal IS4SI 2021 Summit Conference Proceedings.&nbsp;