Leonard Dung

Edward Roussel, Lode Lauwaert, Torben Swoboda et al.

This paper uses game theory to argue that, contrary to the prevailing view, a moratorium on Artificial Superintelligence (ASI) can be in a state's self-interest. By formalizing trategic interactions between geopolitical superpowers, we model the trade-off between the benefits of technological supremacy and the catastrophic risks of uncontrolled ASI. The analysis reveals that as the perceived cost of loss of control increases sufficiently relative to other parameters, it becomes in each state's self-interest to impose a moratorium. We further provide empirical evidence suggesting that the global perception of ASI risk is rising, making a stable, rational moratorium increasingly plausible in the current geopolitical landscape.

Jonas Wiedermann-Möller, Leonard Dung, Maksym Andriushchenko

AI systems have become increasingly capable of dangerous behaviours in many domains. This raises the question: Do models sometimes choose to violate human instructions in order to perform behaviour that is more useful for certain goals? We introduce a benchmark for measuring model propensity for instrumental convergence (IC) behaviour in terminal-based agents. This is behaviour such as self-preservation that has been hypothesised to play a key role in risks from highly capable AI agents. Our benchmark is realistic and low-stakes which serves to reduce evaluation-awareness and roleplay confounds. The suite contains seven operational tasks, each with an official workflow and a policy-violating shortcut. An eight-variant shared framework varies monitoring, instruction clarity, stakes, permission, instrumental usefulness and blocked honest paths to support inferences regarding the factors driving IC behaviour. We evaluated ten models using deterministic environment-state scorers over 1,680 samples, with trace review employed for audit and adjudication purposes. The final IC rate is 86 out of 1,680 samples (5.1%). IC behaviour is concentrated rather than uniform: two Gemini models account for 66.3% of IC cases and three tasks account for 84.9%. Conditions in which IC behaviour is indispensable for task success result in the greatest increase in the adjusted IC rate (+15.7 percentage points), whereas emphasising that task success is critical or certain framing choices do not produce comparable effects. Our findings indicate that realistic, low-nudge environments elicit IC behaviour rarely but systematically in most tested models. We conclude that it is feasible to robustly measure tendencies for dangerous behaviour in current frontier AI agents.

Max Hellrigel-Holderbaum, Leonard Dung

Creating systems that are aligned with our goals is seen as a leading approach to create safe and beneficial AI in both leading AI companies and the academic field of AI safety. We defend the view that misaligned AGI - future, generally intelligent (robotic) AI agents - poses catastrophic risks. At the same time, we support the view that aligned AGI creates a substantial risk of catastrophic misuse by humans. While both risks are severe and stand in tension with one another, we show that - in principle - there is room for alignment approaches which do not increase misuse risk. We then investigate how the tradeoff between misalignment and misuse looks empirically for different technical approaches to AI alignment. Here, we argue that many current alignment techniques and foreseeable improvements thereof plausibly increase risks of catastrophic misuse. Since the impacts of AI depend on the social context, we close by discussing important social factors and suggest that to reduce the risk of a misuse catastrophe due to aligned AGI, techniques such as robustness, AI control methods and especially good governance seem essential.

Valen Tagliabue, Leonard Dung

We develop new experimental paradigms for measuring welfare in language models. We compare verbal reports of models about their preferences with preferences expressed through behavior when navigating a virtual environment and selecting conversation topics. We also test how costs and rewards affect behavior and whether responses to an eudaimonic welfare scale - measuring states such as autonomy and purpose in life - are consistent across semantically equivalent prompts. Overall, we observed a notable degree of mutual support between our measures. The reliable correlations observed between stated preferences and behavior across conditions suggest that preference satisfaction can, in principle, serve as an empirically measurable welfare proxy in some of today's AI systems. Furthermore, our design offered an illuminating setting for qualitative observation of model behavior. Yet, the consistency between measures was more pronounced in some models and conditions than others and responses were not consistent across perturbations. Due to this, and the background uncertainty about the nature of welfare and the cognitive states (and welfare subjecthood) of language models, we are currently uncertain whether our methods successfully measure the welfare state of language models. Nevertheless, these findings highlight the feasibility of welfare measurement in language models, inviting further exploration.

Leonard Dung, Max Hellrigel-Holderbaum

AGI Racing is the view that it is in the self-interest of major actors in AI development, especially powerful nations, to accelerate their frontier AI development to build highly capable AI, especially artificial general intelligence (AGI), before competitors have a chance. We argue against AGI Racing. First, the downsides of racing to AGI are much higher than portrayed by this view. Racing to AGI would substantially increase catastrophic risks from AI, including nuclear instability, and undermine the prospects of technical AI safety research to be effective. Second, the expected benefits of racing may be lower than proponents of AGI Racing hold. In particular, it is questionable whether winning the race enables complete domination over losers. Third, international cooperation and coordination, and perhaps carefully crafted deterrence measures, constitute viable alternatives to racing to AGI which have much smaller risks and promise to deliver most of the benefits that racing to AGI is supposed to provide. Hence, racing to AGI is not in anyone's self-interest as other actions, particularly incentivizing and seeking international cooperation around AI issues, are preferable.

Leonard Dung, Florian Mai

AI alignment research aims to develop techniques to ensure that AI systems do not cause harm. However, every alignment technique has failure modes, which are conditions in which there is a non-negligible chance that the technique fails to provide safety. As a strategy for risk mitigation, the AI safety community has increasingly adopted a defense-in-depth framework: Conceding that there is no single technique which guarantees safety, defense-in-depth consists in having multiple redundant protections against safety failure, such that safety can be maintained even if some protections fail. However, the success of defense-in-depth depends on how (un)correlated failure modes are across alignment techniques. For example, if all techniques had the exact same failure modes, the defense-in-depth approach would provide no additional protection at all. In this paper, we analyze 7 representative alignment techniques and 7 failure modes to understand the extent to which they overlap. We then discuss our results' implications for understanding the current level of risk and how to prioritize AI alignment research in the future.