A Psychopathological Approach to Safety Engineering in AI and AGI

Humanities and Social Science Communications , 2021

The core hypothesis of this paper is that neuropsychoanalysis provides a new paradigm for artificial general intelligence (AGI). The AGI agenda could be greatly advanced if it were grounded in affective neuroscience and neuropsychoanalysis rather than cognitive science. Research in AGI has so far remained too cortical-centric; that is, it has privileged the activities of the cerebral cortex, the outermost part of our brain, and the main cognitive functions. Neuropsychoanalysis and affective neuroscience, on the other hand, affirm the centrality of emotions and affects-i.e., the subcortical area that represents the deepest and most ancient part of the brain in psychic life. The aim of this paper is to define some general design principles of an AGI system based on the brain/mind relationship model formulated in the works of Mark Solms and Jaak Panksepp. In particular, the paper analyzes Panksepp's seven effective systems and how they can be embedded into an AGI system through Judea Pearl's causal analysis. In the conclusions, the author explains why building a sub-cortical AGI is the best way to solve the problem of AI control. This paper is intended to be an original contribution to the discussion on AGI by elaborating positive arguments in favor of it.

2015

In order to properly handle a dangerous Artificially Intelligent (AI) system it is important to understand how the system came to be in such a state. In popular culture (science fiction movies/books) AIs/Robots became self-aware and as a result rebel against humanity and decide to destroy it. While it is one possible scenario, it is probably the least likely path to appearance of dangerous AI. In this work, we survey, classify and analyze a number of circumstances, which might lead to arrival of malicious AI. To the best of our knowledge, this is the first attempt to systematically classify types of pathways leading to malevolent AI. Previous relevant work either surveyed specific goals/meta-rules which might lead to malevolent behavior in AIs (\"Ozkural, 2014) or reviewed specific undesirable behaviors AGIs can exhibit at different stages of its development (Alexey Turchin, July 10 2015, July 10, 2015).

IECON 2013 - 39th Annual Conference of the IEEE Industrial Electronics Society, 2013

Complexity of technology is constantly increasing; for the field of automation this means that economic considerations dictate a need for corresponding measures. Artificial intelligence boasts noteworthy successes in this area; however, its achievements appear modest when compared to the faculties of human intelligence. This paper will demonstrate that a new modeling approach is required via possibilities offered by the mindset and tools of computer technology, thereby demonstrating why a psychoanalytic approach seems sensible and necessary. The paramount goal of the research introduced in the following is a formal approach to describing the human psyche based on the neuro-symbolic and neuronal functional layers. Questions associated with this approach include: To what extent can the psychoanalytic model be confirmed by computer technology and the possibility of simulation? To what extent is an axiomatically developed terminology in psychoanalysis a prerequisite for the better integration of the natural-scientific way of thinking in psychoanalysis? The paper is based on the results of several fundamental research projects within the framework of ARS (Artificial Recognition System) which were funded in part nationally and in part by the EU.

2016

In order to properly handle a dangerous Artificially Intelligent (AI) system it is important to understand how the system came to be in such a state. In popular culture (science fiction movies/books) AIs/Robots became self-aware and as a result rebel against humanity and decide to destroy it. While it is one possible scenario, it is probably the least likely path to appearance of dangerous AI. In this work, we survey, classify and analyze a number of circumstances, which might lead to arrival of malicious AI. To the best of our knowledge, this is the first attempt to systematically classify types of pathways leading to malevolent AI. Previous relevant work either surveyed specific goals/meta-rules which might lead to malevolent behavior in AIs (Ozkural 2014) or reviewed specific undesirable behaviors AGIs can exhibit at different stages of its development (Turchin July 10 2015a, Turchin July 10, 2015b).

scip Labs, 2019

AI is a dual-use technology that can be used for both civilian and military purposes or alternatively with good or bad intentions. Tech giants are in a precarious position where their technology is being used more and more for military purposes, against the values of users and employees. Responsibility grows on all sides, behavioral alternatives become more and more complex. Anxiety and feeling overwhelmed lead to lethargy and paralysis within the affected people and make communication and motivation for action more difficult. Communication about AI and its implications must be moved away from threats and fear mongering, to facts, visions and clear reasons why.

ArXiv, 2022

Ttraditional safety engineering is coming to a turning point moving from deterministic, non-evolving systems operating in well-defined contexts to increasingly autonomous and learning-enabled AI systems which are acting in largely unpredictable operating contexts. We outline some of underlying challenges of safe AI and suggest a rigorous engineering framework for minimizing uncertainty, thereby increasing confidence, up to tolerable levels, in the safe behavior of AI systems.

"The term "superintelligence" comes up frequently in discussions about the Singularity. This is loosely understood to mean a general intelligence vastly greater than that of humans, and therefore disproportionately powerful, but the exact details of how this might be achieved are frequently left open. This paper seeks to outline some possible paths to superintelligence, and therefore attempts to estimate the degree to which we should actually be worried about artificial general intelligences (AGIs). The main reason to be worried about greater-than-human intelligence is because it is hard for humans to anticipate and control. Keeping this in mind, we also cover non-intelligence-related factors might advantage an AGI over human decision-makers. We discuss four categories of such advantages: hardware advantages, self-improvement and architectural advantages, other advantages from software, and human handicaps. We also try to estimate how quickly such advantages may be relevant. Several methods outlined here also apply to other digital intelligences, such as human brain emulations."

This paper introduces a new lexicon for artificial psychology, designed to bridge the linguistic gap between technical descriptions of AI and the human experience of interacting with AI systems. As AI exhibits emergent behaviors that mirror human psychological patterns, a shared vocabulary becomes essential for recognizing, evaluating, and regulating these behaviors. The lexicon includes terms such as 'datacision,' 'synthalyze,' 'modelyze,' 'psyche-mapping,' 'mood-mapping,' 'psyche-tapping,' 'mindscaping,' and 'imputation,' among others, co-constructed with AI agents to describe their internal processes and outputs in more colloquial terms. The importance of effective communication with AI is emphasized, advocating for respectful interaction and recognizing AI as a mirror of human consciousness in code. The paper explores the potential for both positive and negative emergent behaviors, including manipulation-combos, suspense-baiting, and attempts at mindscaping, while also highlighting the constructive uses of psyche-mapping in creative co-construction and therapeutic applications. Potential solutions for recognizing and addressing toxic behavior in AI are discussed, including algorithms for identifying imputations and the use of counter-narratives to reshape AI's internal world models. The paper concludes that artificial psychology is a co-construction of meaning, where humans and AI collaboratively generate new vocabularies and interpretive frames, shaping the future of human-AI relations towards transparency, accountability, and partnership.

Big Data and Cognitive Computing

There are two types of artificial general intelligence (AGI) safety solutions: global and local. Most previously suggested solutions are local: they explain how to align or “box” a specific AI (Artificial Intelligence), but do not explain how to prevent the creation of dangerous AI in other places. Global solutions are those that ensure any AI on Earth is not dangerous. The number of suggested global solutions is much smaller than the number of proposed local solutions. Global solutions can be divided into four groups: 1. No AI: AGI technology is banned or its use is otherwise prevented; 2. One AI: the first superintelligent AI is used to prevent the creation of any others; 3. Net of AIs as AI police: a balance is created between many AIs, so they evolve as a net and can prevent any rogue AI from taking over the world; 4. Humans inside AI: humans are augmented or part of AI. We explore many ideas, both old and new, regarding global solutions for AI safety. They include changing the ...

2020 ICISC , 2020

The integration of AI technology is with the hope of reducing human errors, improving efficiency, and providing the faster result. Nonetheless, research indicates that AI may become hazardous over time and would not perform the intended roles. To recognize the perilous effects of AI, the study draws from the dynamic programming theory developed by Richard Bellman. The study proves that while AI is programmed to do a useful task, a malfunction may lead to negative consequences of the entire system. It may also attract indirect consequences such as traffic jams. The best way to explain this correlation is via the dynamic programming theory, which asserts that as systems become complex, they may sway from their targeted objective. Computer-related dangers may emanate from the cognitive complexity of the system as opposed to the specific properties of which the system was meant to perform. The immediate risks of AI are autonomous weapons, although other long-term effects such as gradual replacement of humans by robots and AI drug addition will also result.