Robert Long

Rapid progress in artificial intelligence (AI) capabilities has drawn fresh attention to the prospect of consciousness in AI. There is an urgent need for rigorous methods to assess AI systems for consciousness, but significant uncertainty about relevant issues in consciousness science. We present a method for assessing AI systems for consciousness that involves exploring what follows from existing or future neuroscientific theories of consciousness. Indicators derived from such theories can be used to inform credences about whether particular AI systems are conscious. This method allows us to make meaningful progress because some influential theories of consciousness, notably including computational functionalist theories, have implications for AI that can be investigated empirically.

In this report, we argue that there is a realistic possibility that some AI systems will be conscious and/or robustly agentic in the near future. That means that the prospect of AI welfare and moral patienthood — of AI systems with their own interests and moral significance — is no longer an issue only for sci-fi or the distant future. It is an issue for the near future, and AI companies and other actors have a responsibility to start taking it seriously. We also recommend three early steps that AI companies and other actors can take: They can (1) acknowledge that AI welfare is an important and difficult issue (and ensure that language model outputs do the same), (2) start assessing AI systems for evidence of consciousness and robust agency, and (3) prepare policies and procedures for treating AI systems with an appropriate level of moral concern. To be clear, our argument in this report is not that AI systems definitely are — or will be — conscious, robustly agentic, or otherwise morally significant. Instead, our argument is that there is substantial uncertainty about these possibilities, and so we need to improve our understanding of AI welfare and our ability to make wise decisions about this issue. Otherwise there is a significant risk that we will mishandle decisions about AI welfare, mistakenly harming AI systems that matter morally and/or mistakenly caring for AI systems that do not.

The field of AI safety considers whether and how AI development can be safe and beneficial for humans and other animals, and the field of AI welfare considers whether and how AI development can be safe and beneficial for AI systems. There is a prima facie tension between these projects, since some measures in AI safety, if deployed against humans and other animals, would raise questions about the ethics of constraint, deception, surveillance, alteration, suffering, death, disenfranchisement, and more. Is there in fact a tension between these projects? We argue that, considering all relevant factors, there is indeed a moderately strong tension—and it deserves more examination. In particular, we should devise interventions that can promote both safety and welfare where possible, and prepare frameworks for navigating any remaining tensions thoughtfully.

This paper considers the kind of introspection that large language models (LLMs) might be able to have. It argues that LLMs, while currently limited in their introspective capabilities, are not inherently unable to have such capabilities: they already model the world, including mental concepts, and already have some introspection-like capabilities. With deliberate training, LLMs may develop introspective capabilities. The paper proposes a method for such training for introspection, situates possible LLM introspection in the 'possible forms of introspection' framework proposed by Kammerer and Frankish, and considers the ethical ramifications of introspection and self-report in AI systems.

As machine learning informs increasingly consequential decisions, different metrics have been proposed for measuring algorithmic bias or unfairness. Two popular “fairness measures” are calibration and equality of false positive rate. Each measure seems intuitively important, but notably, it is usually impossible to satisfy both measures. For this reason, a large literature in machine learning speaks of a “fairness tradeoff” between these two measures. This framing assumes that both measures are, in fact, capturing something important. To date, philosophers have seldom examined this crucial assumption, and examined to what extent each measure actually tracks a normatively important property. This makes this inevitable statistical conflict – between calibration and false positive rate equality – an important topic for ethics. In this paper, I give an ethical framework for thinking about these measures and argue that, contrary to initial appearances, false positive rate equality is in fact morally irrelevant and does not measure fairness.

On a traditional view of perceptual justification, perceptual experiences always provide prima facie justification for beliefs based on them. Against this view, Matthew McGrath and Susanna Siegel argue that if an experience is formed in an epistemically pernicious way then it is epistemically downgraded. They argue that "wishful seeing"—when a subject sees something because he wants to see it—is psychologically and normatively analogous to wishful thinking. They conclude that perception can lose its traditional justificatory power, and that our epistemic norms should govern how experiences are formed. To make this case, the downgrader must first isolate a feature of wishful thinking that makes it epistemically defective, then show that this feature is present in wishful seeing. I present a dilemma for the downgrader. There are two features of wishful thinking that could plausibly explain why it is irrational: the fact that a desire causes you to form a belief not supported by adequate evidence, or the mere influence that desire holds over belief formation. Each option presents formidable difficulties. Although the first “bad evidence” explanation, which McGrath employs, explains the irrationality of wishful thinking, it does not transfer to wishful seeing, since experiences are not formed in response to evidence. The second “influence of desire” explanation, which Siegel employs, fails to isolate an epistemically defective feature of wishful thinking, and also does not transfer to wishful seeing. I conclude that the downgrader’s argument from wishful seeing fails.