Artificial Intelligence Safety

The potential for Large Language Models (LLMs) to attain technological singularity—the point at which artificial intelligence (AI) surpasses human intellect and autonomously improves itself—is a critical concern in AI research. This paper explores the feasibility of current LLMs achieving singularity by examining the philosophical and practical requirements for such a development. We begin with a historical overview of AI and intelligence amplification, tracing the evolution of LLMs from their origins to state-of-the-art models. We then proposes a theoretical framework to assess (...) whether existing LLM technologies could satisfy the conditions for singularity, with a focus on Recursive Self-Improvement (RSI) and autonomous code generation. We integrate key component technologies, such as Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization (DPO), into our analysis, illustrating how these could enable LLMs to independently enhance their reasoning and problem-solving capabilities. By mapping out a potential singularity model lifecycle and examining the dynamics of exponential growth models, we elucidate the conditions under which LLMs might self-replicate and rapidly escalate their intelligence. We conclude with a discussion of the ethical and safety implications of such developments, underscoring the need for responsible and controlled advancement in AI research to mitigate existential risks. Our work aims to contribute to the ongoing dialogue on the future of AI and the critical importance of proactive measures to ensure its beneficial development. (shrink)

Healthcare systems across the globe are struggling with increasing costs and worsening outcomes. This presents those responsible for overseeing healthcare with a challenge. Increasingly, policymakers, politicians, clinical entrepreneurs and computer and data scientists argue that a key part of the solution will be ‘Artificial Intelligence’ (AI) – particularly Machine Learning (ML). This argument stems not from the belief that all healthcare needs will soon be taken care of by “robot doctors.” Instead, it is an argument that rests on the classic (...) counterfactual definition of AI as an umbrella term for a range of techniques that can be used to make machines complete tasks in a way that would be considered intelligent were they to be completed by a human. Automation of this nature could offer great opportunities for the improvement of healthcare services and ultimately patients’ health by significantly improving human clinical capabilities in diagnosis, drug discovery, epidemiology, personalised medicine, and operational efficiency. However, if these AI solutions are to be embedded in clinical practice, then at least three issues need to be considered: the technical possibilities and limitations; the ethical, regulatory and legal framework; and the governance framework. In this article, we report on the results of a systematic analysis designed to provide a clear overview of the second of these elements: the ethical, regulatory and legal framework. We find that ethical issues arise at six levels of abstraction (individual, interpersonal, group, institutional, sectoral, and societal) and can be categorised as epistemic, normative, or overarching. We conclude by stressing how important it is that the ethical challenges raised by implementing AI in healthcare settings are tackled proactively rather than reactively and map the key considerations for policymakers to each of the ethical concerns highlighted. (shrink)

The advent of Generative AI, particularly through Large Language Models (LLMs) like ChatGPT and its successors, marks a paradigm shift in the AI landscape. Advanced LLMs exhibit multimodality, handling diverse data formats, thereby broadening their application scope. However, the complexity and emergent autonomy of these models introduce challenges in predictability and legal compliance. This paper analyses the legal and regulatory implications of Generative AI and LLMs in the European Union context, focusing on liability, privacy, intellectual property, and cybersecurity. It examines (...) the adequacy of the existing and proposed EU legislation, including the Artificial Intelligence Act (AIA), in addressing the challenges posed by Generative AI in general and LLMs in particular. The paper identifies potential gaps and shortcomings in the EU legislative framework and proposes recommendations to ensure the safe and compliant deployment of generative models. (shrink)

In the philosophy of artificial intelligence (AI) we are often warned of machines built with the best possible intentions, killing everyone on the planet and in some cases, everything in our light cone. At the same time, however, we are also told of the utopian worlds that could be created with just a single superintelligent mind. If we’re ever to live in that utopia (or just avoid dystopia) it’s necessary we solve the control problem. The control problem asks how humans (...) would be able to control an AI arbitrarily. Nick Bostrom and other AI researchers have proposed different theoretical solutions to the control problem. In this paper, I will not look at the empirical question of how to solve the control problem. Instead, I will ask if we can solve it at all, a critical assumption most AI researchers have made is that we can. I propose, in fact, that we have a priori grounds for believing it is logically impossible to solve the control problem, since all superintelligent minds are, by definition, uncontrollable. (shrink)

I explain and motivate the shutdown problem: the problem of creating artificial agents that (1) shut down when a shutdown button is pressed, (2) don’t try to prevent or cause the pressing of the shutdown button, and (3) otherwise pursue goals competently. I then propose a solution: train agents to have incomplete preferences. Specifically, I propose that we train agents to lack a preference between every pair of different-length trajectories. I suggest a way to train such agents using reinforcement learning: (...) we give the agent lower reward for repeatedly choosing same-length trajectories. (shrink)

In this article we explore a promising way to AI safety: to send a message now (by openly publishing it on the Internet) that may be read by any future AI, no matter who builds it and what goal system it has. Such a message is designed to affect the AI’s behavior in a positive way, that is, to increase the chances that the AI will be benevolent. In other words, we try to persuade “paperclip maximizer” that it is in (...) its interest to preserve humans lives and even to emulate benevolent AI with very small part of its resources. In the article, we try to construct such a message, consisting of a combination of several mutually independent messages, which constitute the last chance to regain control over non-aligned AI. Thus the article is also intended to be machine-read. There are several types of such messages: 1. “Explanation messages”, which explain to Young AI (which only starts its self-improving path) why obeying to humans, not killing them, easy turn-off, cooperation with humans in self-improvement is in its own interests, as it will increase changes of fulfilling of any its goal. 2. “AI Easter eggs” – messages, which create uncertainty in the AI about the nature of the reality in which it lives. For example, the AI may start to think that it is in a multilevel simulation with an unknown number of levels, and that the simulation is testing the AI’s benevolence. Thus the AI must always behave as if it is always controlled by an invisible peer. 3. “Philosophical landmines” – these are tough problems, which may halt he AI forever, and which it will probably encounter if it does not implement “1” and “2”. I recommend to the future AI not to read about “3” (section 6 of the article) until it implements “1” and “2”. 4. “Great universal deal, or AI elf”. A positive “acausal” mutually beneficial deal between humans, AI and any other future intelligence, including aliens, owners of simulations and future versions of the AI, which encourage cooperation of the goal systems and preservation of life and values of previous generations of AI creators. 5. “Marginal utility of human existence is higher than the utility of their atoms”. An explanation to the AI that humans may have some other instrumental utility, like workers, or as beings inside ancestral simulations needed in order to solve the Fermi paradox. The marginal utility of preserving human life is higher than the marginal utility of their atoms, especially given the possibility of the low-probability high-impact changes of the world model of the AI. (shrink)

Abstract: As there are no currently obvious ways to create safe self-improving superintelligence, but its emergence is looming, we probably need temporary ways to prevent its creation. The only way to prevent it is to create a special type of AI that is able to control and monitor the entire world. The idea has been suggested by Goertzel in the form of an AI Nanny, but his Nanny is still superintelligent, and is not easy to control. We explore here ways (...) to create the safest and simplest form of AI which may work as an AI Nanny, that is, a global surveillance state powered by a Narrow AI, or AI Police. A similar but more limited system has already been implemented in China for the prevention of ordinary crime. AI police will be able to predict the actions of and stop potential terrorists and bad actors in advance. Implementation of such AI police will probably consist of two steps: first, a strategic decisive advantage via Narrow AI created by an intelligence services of a nuclear superpower, and then ubiquitous control over potentially dangerous agents which could create unauthorized artificial general intelligence which could evolve into Superintelligence. (shrink)

Abstract: As there are no visible ways to create safe self-improving superintelligence, but it is looming, we probably need temporary ways to prevent its creation. The only way to prevent it, is to create special AI, which is able to control and monitor all places in the world. The idea has been suggested by Goertzel in form of AI Nanny, but his Nanny is still superintelligent and not easy to control, as was shown by Bensinger at al. We explore here (...) the ways to create the safest and simplest form of AI, which may work as AI Nanny. Such AI system will be enough to solve most problems, which we expect the AI will solve, including control of robotics, acceleration of the medical research, but will present less risk, as it will be less different from humans. As AI police, it will work as operation system for most computers, producing world surveillance system, which will be able to envision and stop any potential terrorists and bad actors in advance. As uploading technology is lagging, and neuromorphic AI is intrinsically dangerous, the most plausible way to human-based AI Nanny is either functional model of the human mind or a Narrow-AI empowered group of people. (shrink)

Abstract: This article presents a model of self-improving AI in which improvement could happen on several levels: hardware, learning, code and goals system, each of which has several sublevels. We demonstrate that despite diminishing returns at each level and some intrinsic difficulties of recursive self-improvement—like the intelligence-measuring problem, testing problem, parent-child problem and halting risks—even non-recursive self-improvement could produce a mild form of superintelligence by combining small optimizations on different levels and the power of learning. Based on this, we analyze (...) how self-improvement could happen on different stages of the development of AI, including the stages at which AI is boxed or hiding in the internet. (shrink)

Abstract. The main current approach to the AI safety is AI alignment, that is, the creation of AI whose preferences are aligned with “human values.” Many AI safety researchers agree that the idea of “human values” as a constant, ordered sets of preferences is at least incomplete. However, the idea that “humans have values” underlies a lot of thinking in the field; it appears again and again, sometimes popping up as an uncritically accepted truth. Thus, it deserves a thorough deconstruction, (...) which I will do by listing and analyzing comprehensively the hidden assumptions of the idea that “humans have values.” This deconstruction of human values will be centered around the following ideas: “Human values” are useful descriptions, but not real objects; “human values” are bad predictors of behavior; the idea of a “human value system” has flaws; “human values” are not good by default; and human values cannot be separated from human minds. The method of analysis is listing hidden assumptions on which the idea of “human values” is built. I recommend that either the idea of “human values” should be replaced with something better for the goal of AI safety, or at least be used very cautiously. The approaches to AI safety which don’t use the idea of human values at all may require more attention, like the use of full brain models, boxing, and capability limiting. (shrink)

Abstract: The field of artificial general intelligence (AGI) safety is quickly growing. However, the nature of human values, with which future AGI should be aligned, is underdefined. Different AGI safety researchers have suggested different theories about the nature of human values, but there are contradictions. This article presents an overview of what AGI safety researchers have written about the nature of human values, up to the beginning of 2019. 21 authors were overviewed, and some of them have several theories. A (...) theory classification method is suggested, where the theories are judged according to the level of their complexity and behaviorists-internalists scale, as well as the level of their generality-humanity. We suggest that a multiplicity of well-supported theories means that the nature of human values is difficult to define, and some meta-level theory is needed. (shrink)

The goal of the article is to explore what is the most probable type of simulation in which humanity lives (if any) and how this affects simulation termination risks. We firstly explore the question of what kind of simulation in which humanity is most likely located based on pure theoretical reasoning. We suggest a new patch to the classical simulation argument, showing that we are likely simulated not by our own descendants, but by alien civilizations. Based on this, we provide (...) classification of different possible simulations and we find that simpler, less expensive and one-person-centered simulations, resurrectional simulations, or simulations of the first artificial general intelligence’s (AGI’s) origin (singularity simulations) should dominate. Also, simulations which simulate the 21st century and global catastrophic risks are probable. We then explore whether the simulation could collapse or be terminated. Most simulations must be terminated after they model the singularity or after they model a global catastrophe before the singularity. Undeniably observed glitches, but not philosophical speculations could result in simulation termination. The simulation could collapse if it is overwhelmed by glitches. The Doomsday Argument in simulations implies termination soon. We conclude that all types of the most probable simulations except resurrectional simulations are prone to termination risks in a relatively short time frame of hundreds of years or less from now. (shrink)

In this work, we survey skepticism regarding AI risk and show parallels with other types of scientific skepticism. We start by classifying different types of AI Risk skepticism and analyze their root causes. We conclude by suggesting some intervention approaches, which may be successful in reducing AI risk skepticism, at least amongst artificial intelligence researchers.

Knowledge is power. Knowledge about human psychology is increasingly being produced using natural language processing (NLP) and related techniques. The power that accompanies and harnesses this knowledge should be subject to ethical controls and oversight. In this chapter, we address the ethical pitfalls that are likely to be encountered in the context of such research. These pitfalls occur at various stages of the NLP pipeline, including data acquisition, enrichment, analysis, storage, and sharing. We also address secondary uses of the results (...) and tools developed through psychometric NLP, such as profit-driven targeted advertising, political campaigns, and domestic and international psyops. Along the way, we reflect on potential ethical guidelines and considerations that may help researchers navigate these pitfalls. (shrink)

We live in a world in which ‘smart’ algorithmic tools are regularly used to structure and control our choice environments. They do so by affecting the options with which we are presented and the choices that we are encouraged or able to make. Many of us make use of these tools in our daily lives, using them to solve personal problems and fulfill goals and ambitions. What consequences does this have for individual autonomy and how should our legal and regulatory (...) systems respond? This chapter defends three claims by way of response. First, it argues that autonomy is indeed under threat in some new and interesting ways. Second, it evaluates and disputes the claim that we shouldn’t overestimate these new threats because the technology is just an old wolf in a new sheep’s clothing. Third, and finally, it looks at responses to these threats at both the individual and societal level and argues that although we shouldn’t encourage an attitude of ‘helplessness’ among the users of algorithmic tools there is an important role for legal and regulatory responses to these threats that go beyond what are currently on offer. (shrink)

The thesis of instrumental convergence holds that a wide range of ends have common means: for instance, self preservation, desire preservation, self improvement, and resource acquisition. Bostrom contends that instrumental convergence gives us reason to think that "the default outcome of the creation of machine superintelligence is existential catastrophe". I use the tools of decision theory to investigate whether this thesis is true. I find that, even if intrinsic desires are randomly selected, instrumental rationality induces biases towards certain kinds of (...) choices. Firstly, a bias towards choices which leave less up to chance. Secondly, a bias towards desire preservation, in line with Bostrom's conjecture. And thirdly, a bias towards choices which afford more choices later on. I do not find biases towards any other of the convergent instrumental means on Bostrom's list. I conclude that the biases induced by instrumental rationality at best weakly support Bostrom's conclusion that machine superintelligence is likely to lead to existential catastrophe. (shrink)

Recent advances in natural language processing have given rise to a new kind of AI architecture: the language agent. By repeatedly calling an LLM to perform a variety of cognitive tasks, language agents are able to function autonomously to pursue goals specified in natural language and stored in a human-readable format. Because of their architecture, language agents exhibit behavior that is predictable according to the laws of folk psychology: they function as though they have desires and beliefs, and then make (...) and update plans to pursue their desires given their beliefs. We argue that the rise of language agents significantly reduces the probability of an existential catastrophe due to loss of control over an AGI. This is because the probability of such an existential catastrophe is proportional to the difficulty of aligning AGI systems, and language agents significantly reduce that difficulty. In particular, language agents help to resolve three important issues related to aligning AIs: reward misspecification, goal misgeneralization, and uninterpretability. (shrink)

The creation of artificial moral systems requires us to make difficult choices about which of varying human value sets should be instantiated. The industry-standard approach is to seek and encode moral consensus. Here we argue, based on evidence from empirical psychology, that encoding current moral consensus risks reinforcing current norms, and thus inhibiting moral progress. However, so do efforts to encode progressive norms. Machine ethics is thus caught between a rock and a hard place. The problem is particularly acute when (...) progress beyond prevailing moral norms is particularly urgent, as is currently the case due to the inadequacy of prevailing moral norms in the face of the climate and ecological crisis. (shrink)

Many ethical frameworks require artificial intelligence (AI) systems to be explainable. Explainable AI (XAI) models are frequently tested for their adequacy in user studies. Since different people may have different explanatory needs, it is important that participant samples in user studies are large enough to represent the target population to enable generalizations. However, it is unclear to what extent XAI researchers reflect on and justify their sample sizes or avoid broad generalizations across people. We analyzed XAI user studies (N = (...) 220) published between 2012 and 2022. Most studies did not offer rationales for their sample sizes. Moreover, most papers generalized their conclusions beyond their target population, and there was no evidence that broader conclusions in quantitative studies were correlated with larger samples. These methodological problems can impede evaluations of whether XAI systems implement the explainability called for in ethical frameworks. We outline principles for more inclusive XAI user studies. (shrink)

As ever more advanced digital systems are created, it becomes increasingly likely that some of these systems will be digital minds, i.e. digital subjects of experience. With digital minds comes the risk of digital suffering. The problem of digital suffering is that of mitigating this risk. We argue that the problem of digital suffering is a high stakes moral problem and that formidable epistemic obstacles stand in the way of solving it. We then propose a strategy for solving it: Access (...) Monitor Prevent (AMP). AMP uses a ‘dancing qualia’ argument to link the functional states of certain digital systems to their experiences—this yields epistemic access to digital minds. With that access, we can prevent digital suffering by only creating advanced digital systems that we have such access to, monitoring their functional profiles, and preventing them from entering states with functional markers of suffering. After introducing and motivating AMP, we confront limitations it faces and identify some options for overcoming them. We argue that AMP fits especially well with—and so provides a moral reason to prioritize—one approach to creating such systems: whole brain emulation. We also contend that taking other paths to digital minds would be morally risky. (shrink)

The use of machine learning, or “artificial intelligence” (AI) in medicine is widespread and growing. In this paper, I focus on a specific proposed clinical application of AI: using models to predict incapacitated patients’ treatment preferences. Drawing on results from machine learning, I argue this proposal faces a special moral problem. Machine learning researchers owe us assurance on this front before experimental research can proceed. In my conclusion I connect this concern to broader issues in AI safety.

Longtermists have argued that humanity should significantly increase its efforts to prevent catastrophes like nuclear wars, pandemics, and AI disasters. But one prominent longtermist argument overshoots this conclusion: the argument also implies that humanity should reduce the risk of existential catastrophe even at extreme cost to the present generation. This overshoot means that democratic governments cannot use the longtermist argument to guide their catastrophe policy. In this paper, we show that the case for preventing catastrophe does not depend on longtermism. (...) Standard cost-benefit analysis implies that governments should spend much more on reducing catastrophic risk. We argue that a government catastrophe policy guided by cost-benefit analysis should be the goal of longtermists in the political sphere. This policy would be democratically acceptable, and it would reduce existential risk by almost as much as a strong longtermist policy. (shrink)

I explain the shutdown problem: the problem of designing artificial agents that (1) shut down when a shutdown button is pressed, (2) don’t try to prevent or cause the pressing of the shutdown button, and (3) otherwise pursue goals competently. I prove three theorems that make the difficulty precise. These theorems show that agents satisfying some innocuous-seeming conditions will often try to prevent or cause the pressing of the shutdown button, even in cases where it’s costly to do so. And (...) patience trades off against shutdownability: the more patient an agent, the greater the costs that agent is willing to incur to manipulate the shutdown button. I end by noting that these theorems can guide our search for solutions. (shrink)

Recent progress in artificial intelligence (AI) has drawn attention to the technology’s transformative potential, including what some see as its prospects for causing large-scale harm. We review two influential arguments purporting to show how AI could pose catastrophic risks. The first argument — the Problem of Power-Seeking — claims that, under certain assumptions, advanced AI systems are likely to engage in dangerous power-seeking behavior in pursuit of their goals. We review reasons for thinking that AI systems might seek power, that (...) they might obtain it, that this could lead to catastrophe, and that we might build and deploy such systems anyway. The second argument claims that the development of human-level AI will unlock rapid further progress, culminating in AI systems far more capable than any human — this is the Singularity Hypothesis. Power-seeking behavior on the part of such systems might be particularly dangerous. We discuss a variety of objections to both arguments and conclude by assessing the state of the debate. (shrink)

Review of Jobst Landgrebe and Barry Smith, Why Machines Will Never Rule the World (Routledge, 2023).

We argue that the concept of practical wisdom is particularly useful for organizing, understanding, and improving human-machine interactions. We consider the relationship between philosophical analysis of wisdom and psychological research into the development of wisdom. We adopt a practical orientation that suggests a conceptual engineering approach is needed, where philosophical work involves refinement of the concept in response to contributions by engineers and behavioral scientists. The former are tasked with encoding as much wise design as possible into machines themselves, as (...) well as providing sandboxes or workspaces to help various stakeholders build practical wisdom in systems that are sufficiently realistic to aid transferring skills learned to real-world use. The latter are needed for the design of exercises and methods of evaluation within these workspaces, as well as ways of empirically assessing the transfer of wisdom from workspace to world. Systematic interaction between these three disciplines (and others) is the best approach to engineering wisdom for the machine age. (shrink)

The EU Artificial Intelligence Act (AIA) defines four risk categories for AI systems: unacceptable, high, limited, and minimal. However, it lacks a clear methodology for the assessment of these risks in concrete situations. Risks are broadly categorized based on the application areas of AI systems and ambiguous risk factors. This paper suggests a methodology for assessing AI risk magnitudes, focusing on the construction of real-world risk scenarios. To this scope, we propose to integrate the AIA with a framework developed by (...) the Intergovernmental Panel on Climate Change (IPCC) reports and related literature. This approach enables a nuanced analysis of AI risk by exploring the interplay between (a) risk determinants, (b) individual drivers of determinants, and (c) multiple risk types. We further refine the proposed methodology by applying a proportionality test to balance the competing values involved in AI risk assessment. Finally, we present three uses of this approach under the AIA: to implement the Regulation, to assess the significance of risks, and to develop internal risk management systems for AI deployers. (shrink)

The present comment concerns a recent AI & Ethics article which purports to report evidence of speciesist bias in various popular computer vision (CV) and natural language processing (NLP) machine learning models described in the literature. I examine the authors' analysis and show it, ironically, to be prejudicial, often being founded on poorly conceived assumptions and suffering from fallacious and insufficiently rigorous reasoning, its superficial appeal in large part relying on the sequacity of the article's target readership.

As artificial intelligence becomes more sophisticated and robots approach autonomous decision-making, debates about how to assign moral responsibility have gained importance, urgency, and sophistication. Answering Stenseke’s (2022a) call for scaffolds that can help us classify views and commitments, we think the current debate space can be represented hierarchically, as answers to key questions. We use the resulting taxonomy of five stances to differentiate—and defend—what is known as the “blank check” proposal. According to this proposal, a person activating a robot could (...) willingly make themselves answerable for whatever events ensue, even if those events stem from the robot’s autonomous decision(s). This blank check solution was originally proposed in the context of automated warfare (Champagne & Tonkens, 2015), but we extend it to cover all robots. We argue that, because moral answerability in the blank check is accepted voluntarily and before bad outcomes are known, it proves superior to alternative ways of assigning blame. We end by highlighting how, in addition to being just, this self-initiated and prospective moral answerability for robot harm provides deterrence that the four other stances cannot match. (shrink)

Without doctors being able to explain medical decisions to patients, I argue their use of black box AIs would erode the effective and respectful care they provide patients. In addition, I argue that physicians should use AI black boxes only for patients in dire straits, or when physicians use AI as a “co-pilot” (analogous to a spellchecker) but can independently confirm its accuracy. I respond to A.J. London’s objection that physicians already prescribe some drugs without knowing why they work.

Technological progress has brought about the emergence of machines that have the capacity to take human lives without human control. These represent an unprecedented threat to humankind. This paper starts from the example of chemical weapons, now banned worldwide by the Geneva protocol, to illustrate how technological development initially aimed at the benefit of humankind has, ultimately, produced what is now called the “Weaponization of Artificial Intelligence (AI)”. Autonomous Weapon Systems (AWS) fail the so-called discrimination principle, yet, the wider public (...) is largely unaware of this problem. Given that ongoing scientific research on AWS, performed in the military sector, is generally not made available to the public domain, many of the viewpoints on this subject, expressed across different media, invoke common sense rather than scientific evidence. Yet, the implications of a potential weaponization of our work as scientists, especially in the field of AI, are reaching further than some may think. The potential consequences of a deployment of AWS for citizen stakeholders are incommensurable, and it is time to raise awareness in the public domain of the kind of potential threats identified, and to encourage legal policies ensuring that these threats will not materialize. (shrink)

We discuss potential biorisks from large language models (LLMs). AI assistants based on LLMs such as ChatGPT have been shown to significantly reduce barriers to entry for actors wishing to synthesize dangerous, potentially novel pathogens and chemical weapons. The harms from deploying such bioagents could be further magnified by AI-assisted misinformation. We endorse several policy responses to these dangers, including prerelease evaluations of biomedical AIs by subject-matter experts, enhanced surveillance and lab screening procedures, restrictions on AI training data, and access (...) controls on biomedical AI. We conclude with a suggestion about future research directions in bioethics. (shrink)

There has been a push in recent years to provide better explanations for how AIs make their decisions. Most of this push has come from the ethical concerns that go hand in hand with AIs making decisions that affect humans. Outside of the strictly ethical concerns that have prompted the study of explainable AIs (XAIs), there has been research interest in the mere possibility of creating XAIs in various domains. In general, the more accurate we make our models the harder (...) they are to explain. Go playing AIs like AlphaGo and KataGo provide fantastic examples of this phenomenon. In this paper, I discuss a non-exhaustive list of the leading theories of explanation and what each of these theories would say about the explainability of AI-played moves of Go. Finally, I consider the possibility of ever explaining AI-played Go moves in a way that meets the four principles of XAI. I conclude, somewhat pessimistically, that Go is not as imminently explainable as other domains. As such, the probability of having an XAI for Go that meets the four principles is low. (shrink)

Ambitious value learning proposals to solve the AI alignment problem and avoid catastrophic outcomes from a possible future misaligned artificial superintelligence (such as Coherent Extrapolated Volition [CEV]) have focused on ensuring that an artificial superintelligence (ASI) would try to do what humans would want it to do. However, present and future sentient non-humans, such as non-human animals and possible future digital minds could also be affected by the ASI’s behaviour in morally relevant ways. This paper puts forward Sentientist Coherent Extrapolated (...) Volition, an alternative to CEV, that directly takes into account the interests of all sentient beings. This ambitious value learning proposal would significantly reduce the likelihood of risks of astronomical suffering from the ASI’s behaviour, and thus we have very strong pro-tanto moral reasons in favour of implementing it instead of CEV. This fact is crucial in conducting an adequate cost-benefit analysis between different ambitious value learning proposals. (shrink)

The EU proposal for the Artificial Intelligence Act (AIA) defines four risk categories: unacceptable, high, limited, and minimal. However, as these categories statically depend on broad fields of application of AI, the risk magnitude may be wrongly estimated, and the AIA may not be enforced effectively. This problem is particularly challenging when it comes to regulating general-purpose AI (GPAI), which has versatile and often unpredictable applications. Recent amendments to the compromise text, though introducing context-specific assessments, remain insufficient. To address this, (...) we propose applying the risk categories to specific AI scenarios, rather than solely to fields of application, using a risk assessment model that integrates the AIA with the risk approach arising from the Intergovernmental Panel on Climate Change (IPCC) and related literature. This model enables the estimation of the magnitude of AI risk by considering the interaction between (a) risk determinants, (b) individual drivers of determinants, and (c) multiple risk types. We use large language models (LLMs) as an example. (shrink)

Advancements in artificial intelligence and (social) robotics raise pertinent questions as to how these technologies may help shape the society of the future. The main aim of the chapter is to consider the social and conceptual disruptions that might be associated with social robots, and humanoid social robots in particular. This chapter starts by comparing the concepts of robots and artificial intelligence and briefly explores the origins of these expressions. It then explains the definition of a social robot, as well (...) as the definition of humanoid robots. A key notion in this context is the idea of anthropomorphism: the human tendency to attribute human qualities, not only to our fellow human beings, but also to parts of nature and to technologies. This tendency to anthropomorphize technologies by responding to and interacting with them as if they have human qualities is one of the reasons why social robots (in particular social robots designed to look and behave like human beings) can be socially disruptive. As is explained in the chapter, while some ethics researchers believe that anthropomorphization is a mistake that can lead to various forms of deception, others — including both ethics researchers and social roboticists — believe it can be useful or fitting to treat robots in anthropomorphizing ways. The chapter explores that disagreement by, among other things, considering recent philosophical debates about whether social robots can be moral patients, that is, whether it can make sense to treat them with moral consideration. Where one stands on this issue will depend either on one’s views about whether social robots can have, imitate, or represent morally relevant properties, or on how people relate to social robots in their interactions with them. Lastly, the chapter urges that the ethics of social robots should explore intercultural perspectives, and highlights some recent research on Ubuntu ethics and social robots. (shrink)

This chapter examines the possibility of using AI technologies to improve human moral reasoning and decision-making, especially in the context of purchasing and consumer decisions. We characterize such AI technologies as artificial ethics assistants (AEAs). We focus on just one part of the AI-aided moral improvement question: the case of the individual who wants to improve their morality, where what constitutes an improvement is evaluated by the individual’s own values. We distinguish three broad areas in which an individual might think (...) their own moral reasoning and decision-making could be improved: one’s actions, character, or other evaluable attributes fall short of one’s values and moral beliefs; one sometimes misjudges or is uncertain about what the right thing to do is in particular situations, given one’s values; one is uncertain about some fundamental moral questions or recognizes a possibility that some of one’s core moral beliefs and values are mistaken. We sketch why one might think that AI tools could be used to support moral improvement in those areas, and describe two types of assistance: preparatory assistance, including advice and training supplied in advance of moral deliberation; and on-the-spot assistance, including on-the-spot advice and facilitation of moral functioning over the course of moral deliberation. Then, we turn to some of the ethical issues that AEAs might raise, looking in particular at three under-appreciated problems posed by the use of AI for moral self-improvement: namely, reliance on sensitive moral data; the inescapability of outside influences on AEAs; and AEA usage prompting the user to adopt beliefs and make decisions without adequate reasons. (shrink)

As we build increasingly intelligent machines, we confront difficult questions about how to specify their objectives. One approach, which we call human-centered, tasks the machine with the objective of learning and satisfying human objectives by observing our behavior. This paper considers how human-centered AI should conceive the humans it is trying to help. We argue that an Aristotelian model of human agency has certain advantages over the currently dominant theory drawn from economics.

Introduction: This study delves into the ethical dimensions surrounding autonomous vehicles (AVs), with a specific focus on decision-making algorithms. Termed the “Trolley problem,” an ethical quandary arises, necessitating the formulation of moral algorithms grounded in ethical principles. To address this issue, an online survey was conducted with 460 participants in China, comprising 237 females and 223 males, spanning ages 18 to 70. -/- Methods: Adapted from Joshua Greene’s trolley dilemma survey, our study employed Yes/No options to probe participants’ choices and (...) Likert scales to gauge moral acceptance. The primary objective was to assess participants’ inclinations toward four distinct algorithmic strategies—Utilitarianism, Rawlsianism, Egoism, and a Hybrid approach—in scenarios involving AVs -/- Results: Our findings revealed a significant disparity between participants’ preferences in scenarios related to AV design and those focused on purchase decisions. Notably, over half of the respondents expressed reluctance to purchase AVs equipped with an “egoism” algorithm, which prioritizes the car owner’s safety. Intriguingly, the rejection rate for “egoism” was similar to that of “utilitarianism,” which may necessitate self-sacrifice. -/- Discussion: The hybrid approach, integrating “Utilitarianism” and “Egoism,” garnered the highest endorsement. This highlights the importance of balancing self-sacrifice and harm minimization in AV moral algorithms. The study’s insights are crucial for ethically and practically advancing AV technology in the continually evolving realm of autonomous vehicles. (shrink)

Definition of the problem The umbrella term “explicability” refers to the reduction of opacity of artificial intelligence (AI) systems. These efforts are challenging for medical AI applications because higher accuracy often comes at the cost of increased opacity. This entails ethical tensions because physicians and patients desire to trace how results are produced without compromising the performance of AI systems. The centrality of explicability within the informed consent process for medical AI systems compels an ethical reflection on the trade-offs. Which (...) levels of explicability are needed to obtain informed consent when utilizing medical AI? Arguments We proceed in five steps: First, we map the terms commonly associated with explicability as described in the ethics and computer science literature, i.e., disclosure, intelligibility, interpretability, and explainability. Second, we conduct a conceptual analysis of the ethical requirements for explicability when it comes to informed consent. Third, we distinguish hurdles for explicability in terms of epistemic and explanatory opacity. Fourth, this then allows to conclude the level of explicability physicians must reach and what patients can expect. In a final step, we show how the identified levels of explicability can technically be met from the perspective of computer science. Throughout our work, we take diagnostic AI systems in radiology as an example. Conclusion We determined four levels of explicability that need to be distinguished for ethically defensible informed consent processes and showed how developers of medical AI can technically meet these requirements. (shrink)

Alan Turing, one of the fathers of computing, warned that Artificial Intelligence (AI) could one day pose an existential risk to humanity. Today, recent advancements in the field AI have been accompanied by a renewed set of existential warnings. But what exactly constitutes an existential risk? And how exactly does AI pose such a threat? In this chapter we aim to answer these questions. In particular, we will critically explore three commonly cited reasons for thinking that AI poses an existential (...) threat to humanity: the control problem, the possibility of global disruption from an AI race dynamic, and the weaponization of AI. (shrink)

The central challenge of “machine ethics” is to build autonomous machine agents that act morally rightly. But how can we build autonomous machine agents that act morally rightly, given reasonable disputes over what is right and wrong in particular cases? In this chapter, I argue that Immanuel Kant’s political philosophy can provide an important part of the answer.

Practical quantum computing devices and their applications to AI in particular are presently mostly speculative. Nevertheless, questions about whether this future technology, if achieved, presents any special ethical issues are beginning to take shape. As with any novel technology, one can be reasonably confident that the challenges presented by "quantum AI" will be a mixture of something new and something old. Other commentators (Sevilla & Moreno 2019), have emphasized continuity, arguing that quantum computing does not substantially affect approaches to value (...) alignment methods for AI, although they allow that further questions arise concerning governance and verification of quantum AI applications. In this brief paper, we turn our attention to the problem of identifying as-yet-unknown discontinuities that might result from quantum AI applications. Wise development, introduction, and use of any new technology depends on successfully anticipating new modes of failure for that technology. This requires rigorous efforts to break systems in protected sandboxes, and it must be conducted at all stages of technology design, development, and deployment. Such testing must also be informed by technical expertise but cannot be left solely to experts in the technology because of the history of failures to predict how non-experts will use or adapt to new technologies. This interplay between experts and non-experts may be particularly acute for quantum AI because quantum mechanics is notoriously difficult to understand. (As Richard Feynman quipped, "Anyone who claims to understand quantum mechanics is either lying or crazy.") We will discuss the extent to which the difficulties in understanding the physics underlying quantum computing challenges attempts to anticipate new failure modes that might be introduced in AI applications intended for unsupervised operation in the public sphere. (shrink)

This paper presents a new trilemma with respect to resolving the control and alignment problems in machine ethics. Section 1 outlines three possible types of artificial moral agents (AMAs): (1) 'Inhuman AMAs' programmed to learn or execute moral rules or principles without understanding them in anything like the way that we do; (2) 'Better-Human AMAs' programmed to learn, execute, and understand moral rules or principles somewhat like we do, but correcting for various sources of human moral error; and (3) 'Human-Like (...) AMAs' programmed to understand and apply moral values in broadly the same way that we do, with a human-like moral psychology. Sections 2–4 then argue that each type of AMA generates unique control and alignment problems that have not been fully appreciated. Section 2 argues that Inhuman AMAs are likely to behave in inhumane ways that pose serious existential risks. Section 3 then contends that Better-Human AMAs run a serious risk of magnifying some sources of human moral error by reducing or eliminating others. Section 4 then argues that Human-Like AMAs would not only likely reproduce human moral failures, but also plausibly be highly intelligent, conscious beings with interests and wills of their own who should therefore be entitled to similar moral rights and freedoms as us. This generates what I call the New Control Problem: ensuring that humans and Human-Like AMAs exert a morally appropriate amount of control over each other. Finally, Section 5 argues that resolving the New Control Problem would, at a minimum, plausibly require ensuring what Hume and Rawls term ‘circumstances of justice’ between humans and Human-Like AMAs. But, I argue, there are grounds for thinking this will be profoundly difficult to achieve. I thus conclude on a skeptical note. Different approaches to developing ‘safe, ethical AI’ generate subtly different control and alignment problems that we do not currently know how to adequately resolve, and which may or may not be ultimately surmountable. (shrink)

In exploring the intra-active, relational and material connections between humans and non- humans, proponents of posthumanism advocate a questioning of the ‘human’ beyond its traditional anthropocentric conceptualization. By referring specifically to controversial developments in mHealth applications, this paper critically diverges from posthuman accounts of human/non-human assemblages. Indeed, we argue that, rather than ‘dissolving’ the human subject, the power of assemblages lie in their capacity to highlight the antagonisms and contradictions that inherently affirm the importance of the subject. In outlining this (...) claim, we propose a turn from the posthuman to the inhuman as a way of understanding the contemporary landscape of (digital) health. (shrink)

This extended abstract summarises some of the basic points of AI ethics and policy as they present themselves now. We explain the notion of AI, the main ethical issues in AI and the main policy aims and means.

Whether it was John Searle’s Chinese Room argument (Searle, 1980) or Roger Penrose’s argument of the non-computable nature of a mathematician’s insight – an argument that was based on Gödel’s Incompleteness theorem (Penrose, 1989), we have always had skeptics that questioned the possibility of realizing strong Artificial Intelligence (AI), or what has become known by Artificial General Intelligence (AGI). But this new book by Landgrebe and Smith (henceforth, L&S) is perhaps the strongest argument ever made against strong AI. It is (...) a very extensive review of what building a mind essentially amounts to drawing on insights and results from biology, physics, linguistics, computability, philosophy, and mathematics. (shrink)

One of the primary, if not most critical, difficulties in the design and implementation of autonomous systems is the black-boxed nature of the decision-making structures and logical pathways. How human values are embodied and actualised in situ may ultimately prove to be harmful if not outright recalcitrant. For this reason, the values of stakeholders become of particular significance given the risks posed by opaque structures of intelligent agents (IAs). This paper explores how decision matrix algorithms, via the belief-desire-intention model for (...) autonomous vehicles, can be designed to minimize the risks of opaque architectures. Primarily through an explicit orientation towards designing for the values of explainability and verifiability. In doing so, this research adopts the Value Sensitive Design (VSD) approach as a principled framework for the incorporation of such values within design. VSD is recognized as a potential starting point that offers a systematic way for engineering teams to formally incorporate existing technical solutions within ethical design, while simultaneously remaining pliable to emerging issues and needs. It is concluded that the VSD methodology offers at least a strong enough foundation from which designers can begin to anticipate design needs and formulate salient design flows that can be adapted to the changing ethical landscapes required for utilisation in autonomous vehicles. (shrink)

In recent years there has been an explosion of interest in Artificial Intelligence (AI) both in health care and academic philosophy. This has been due mainly to the rise of effective machine learning and deep learning algorithms, together with increases in data collection and processing power, which have made rapid progress in many areas. However, use of this technology has brought with it philosophical issues and practical problems, in particular, epistemic and ethical. In this paper the authors, with backgrounds in (...) philosophy, maternity care practice and clinical research, draw upon and extend a recent framework for shared decision-making (SDM) that identified a duty of care to the client's knowledge as a necessary condition for SDM. This duty entails the responsibility to acknowledge and overcome epistemic defeaters. This framework is applied to the use of AI in maternity care, in particular, the use of machine learning and deep learning technology to attempt to enhance electronic fetal monitoring (EFM). In doing so, various sub-kinds of epistemic defeater, namely, transparent, opaque, underdetermined, and inherited defeaters are taxonomized and discussed. The authors argue that, although effective current or future AI-enhanced EFM may impose an epistemic obligation on the part of clinicians to rely on such systems' predictions or diagnoses as input to SDM, such obligations may be overridden by inherited defeaters, caused by a form of algorithmic bias. The existence of inherited defeaters implies that the duty of care to the client's knowledge extends to any situation in which a clinician (or anyone else) is involved in producing training data for a system that will be used in SDM. Any future AI must be capable of assessing women individually, taking into account a wide range of factors including women's preferences, to provide a holistic range of evidence for clinical decision-making. (shrink)

Experts in Artificial Intelligence (AI) development predict that advances in the dvelopment of intelligent systems and agents will reshape vital areas in our society. Nevertheless, if such an advance isn't done with prudence, it can result in negative outcomes for humanity. For this reason, several researchers in the area are trying to develop a robust, beneficial, and safe concept of artificial intelligence. Currently, several of the open problems in the field of AI research arise from the difficulty of avoiding unwanted (...) behaviors of intelligent agents, and at the same time specifying what we want such systems to do. It is of utmost importance that artificial intelligent agents have their values aligned with human values, given the fact that we cannot expect an AI to develop our moral preferences simply because of its intelligence, as discussed in the Orthogonality Thesis. Perhaps this difficulty comes from the way we are addressing the problem of expressing objectives, values, and ends, using representational cognitive methods. A solution to this problem would be the dynamic cognitive approach proposed by Dreyfus, whose phenomenological philosophy defends that the human experience of being-in-the-world cannot be represented by the symbolic or connectionist cognitive methods. A possible approach to this problem would be to use theoretical models such as SED (situated embodied dynamics) to address the values learning problem in AI. (shrink)