The Ethics of Advanced AI Assistants

Gabriel, I., Manzini, A., et al. (2024).

Google Deep Mind

This paper focuses on the opportunities and the ethical and societal risks posed by advanced AI assistants. We define advanced AI assistants as artificial agents with natural language interfaces, whose function is to plan and execute sequences of actions on behalf of a user – across one or more domains – in line with the user’s expectations. The paper starts by considering the technology itself, providing an overview of AI assistants, their technical foundations and potential range of applications. It then explores questions around AI value alignment, well-being, safety and malicious uses. Extending the circle of inquiry further, we next consider the relationship between advanced AI assistants and individual users in more detail, exploring topics such as manipulation and persuasion, anthropomorphism, appropriate relationships, trust and privacy. With this analysis in place, we consider the deployment of advanced assistants at a societal scale, focusing on cooperation, equity and access, misinformation, economic impact, the environment and how best to evaluate advanced AI assistants. Finally, we conclude by providing a range of recommendations for researchers, developers, policymakers and public stakeholders.

Our analysis suggests that advanced AI assistants are likely to have a profound impact on our individual and collective lives. To be beneficial and value-aligned, we argue that assistants must be appropriately responsive to the competing claims and needs of users, developers and society. Features such as increased agency, the capacity to interact in natural language and high degrees of personalisation could make AI assistants especially helpful to users. However, these features also make people vulnerable to inappropriate influence by the technology, so robust safeguards are needed. Moreover, when AI assistants are deployed at scale, knock-on effects that arise from interaction between them and questions about their overall impact on wider institutions and social processes rise to the fore. These dynamics likely require technical and policy interventions in order to foster beneficial cooperation and to achieve broad, inclusive and equitable outcomes. Finally, given that the current landscape of AI evaluation focuses primarily on the technical components of AI systems, it is important to invest in the holistic sociotechnical evaluations of AI assistants, including human–AI interaction, multi-agent and societal level research, to support responsible decision-making and deployment in this domain.

The paper is here.

Here are some summary thoughts:

The development of increasingly advanced AI assistants represents a significant technological shift, moving beyond narrow AI for specific tasks to general-purpose foundation models that enable greater autonomy and scope.

These advanced AI assistants can provide novel services (like summarization, ideation, planning, and tool use), with the potential to become deeply integrated into our economic, social, and personal lives.

Ethical and Societal Implications

Profound Impact Potential: AI assistants could radically alter work, education, creativity, communication, and how we make decisions about our lives and goals.

Safety, Alignment, and Misuse: The autonomy of AI assistants presents challenges around safety, ensuring alignment with user intentions, and potential for misuse.

Human-Assistant Interactions: Issues around trust, privacy, anthropomorphism, and the moral limits of personalization need to be considered.

Social Impacts: AI assistants could affect the distribution of benefits and burdens in society, as well as how humans cooperate and coordinate.

Evaluation Challenges: New methodologies are needed to evaluate AI assistants as part of a broader sociotechnical system, beyond just model performance.

Responsible Development: Ongoing research, policy work, and public discussion are required to address the novel normative and technical challenges posed by advanced AI assistants.

Concluding Thoughts

The development of advanced AI assistants represents a transformative technological shift, and the choices we make now will shape their future path. Coordinated efforts across researchers, developers, policymakers, and the public are needed to ensure these assistants are developed responsibly and in the public interest.

Evangelical literary tradition and moral foundations theory

Christopher Douglas

The Journal of American Culture

Originally published 26 Feb 24

Here is an excerpt:

What can MFT tell us about the topography of evangelical ethics as displayed in its bestselling fiction of the last 20 years? In many ways, there is nothing surprising in these findings. As Haidt himself suggests, the five primary foundations discernably track onto political orientations, with conservatives balancing all five criteria but liberals prioritizing care and fairness (as equality): “it's not just members of traditional societies who draw on all five foundations; even within Western societies, we consistently find an ideological effect in which religious and cultural conservatives value and rely upon all five foundations, whereas liberals value and rely upon the harm and fairness foundations primarily” (Haidt, 2007, 1001). Or, in updated form: “Liberals have a three-foundation morality, whereas conservatives use all six” (Haidt, 2012, 214). The Shack seems to aptly confirm this insight, prioritizing care, fairness-as-justice, and egalitarianism at the expense of loyalty, authority, and purity. These values reflect the author's liberal sensibilities that were suggested when Young tweeted criticism of Donald Trump after the Access Hollywood tapes were released (Douglas, 2020, 508n3). LaHaye's conservative credentials, meanwhile, are well known—early partner to Jerry Falwell in the formation of the Moral Majority, fundraiser for the Institute for Creation Research, and so on—and the Left Behind series suggests a mix of moral foundations that does not so much find a balance among all six foundations (as Haidt discovered seems to be true of “Very Conservatives”) as express a sort of Extremely Conservative sensibility. The Shack and the Left Behind series reflect the considerable range of white evangelical politics, but also reflect the fact that white evangelicals tilt heavily conservative, forming the most important demographic of the Republican base, voting for Donald Trump by 77 and 84% in 2016 and 2020, respectively (Igielnik et al., 2021).

The article is here.

Here is my summary:

The article explores the moral foundations of two evangelical best-selling novels: The Shack by William Paul Young and Left Behind by Tim LaHaye and Jerry Jenkins. It uses Moral Foundations Theory (MFT) to analyze how these seemingly very different novels prioritize different moral values.

Moral Foundations Theory (MFT) identifies five core moral foundations:

• Care/Harm: Protecting others from harm and promoting their well-being.
• Fairness/Cheating: Ensuring that people are treated justly and receive what they deserve.
• Loyalty/Betrayal: Standing by your group and upholding your commitments.
• Authority/Subversion: Respecting legitimate authority figures and hierarchies.
• Sanctity/Degradation: Purity, avoiding disgust and respecting the sacred.

The Shack by William Paul Young grapples with the kidnapping, abuse, and murder of a child. It focuses on the themes of care/harm and fairness. The protagonist, Mack, wrestles with how God could allow such a tragedy to occur and how fairness can be achieved. The novel explores the idea of forgiveness and reconciliation.

Left Behind by Tim LaHaye and Jerry Jenkins is a series about the Rapture and the End Times. It emphasizes the moral foundations of loyalty/betrayal, authority/subversion, and sanctity/degradation. The series depicts a world where good and evil are clearly defined and a battle between God and the Antichrist is about to unfold. The in-group of Christians is loyal to God and resists the authority of the Antichrist. The series emphasizes the importance of following God's will and upholding Christian values.

The article argues that MFT helps explain the enduring appeal of these novels.  The Shack resonates with readers who seek comfort and answers in the face of tragedy. Left Behind appeals to readers who feel like they are part of an embattled community and who believe in a clear distinction between good and evil.

Responding Effectively to Disruptive Patient Behaviors: Beyond Behavior Contracts

Fabi R, Johnson LSM. 

JAMA. 2024;331(10):823–824.

doi:10.1001/jama.2024.0216

Here is an excerpt:

The epidemic of workplace violence has prompted the use of harsh responses that include “behavior contracts” (sometimes called “behavioral agreements”) that can undermine a hospital’s commitment to providing evidence-based, patient-centered care. There is no national repository of data on the use of behavior contracts, or on hospital policies, but in our experience as clinical ethics consultants, and through discussions with colleagues nationally, we have observed that hospitals increasingly try to manage so-called difficult patients and families through behavior contracts that impose paternalistic limits and punitive consequences on patients for a wide range of behaviors. Yet behavior contracts pose serious ethical challenges, especially when unilaterally imposed on patients whose behavior is upsetting and disrespectful but not unsafe. Moreover, the evidence supporting the efficacy of contracts is lacking.

Behavior contracts are used in a variety of health care contexts to promote patient adherence with treatment, including smoking cessation, weight loss, substance use disorder rehabilitation, and psychiatric treatment. A Cochrane systematic review found that evidence of their effectiveness at improving adherence is limited and mixed; it did not find evidence from randomized clinical trials outside of this context.1 Indeed, we could find no empirical evidence to support or challenge the effectiveness of behavior contracts as a tool for addressing the problems of undesirable patient or family behaviors, patient-staff conflicts, and workplace violence in health care. Absent such evidence, health care institutions committed to evidence-based medicine and workplace safety might hesitate before using these contracts. When viewed alongside the ethical considerations, which have been extensively explored in the bioethics literature, we argue that the lack of supportive evidence generates an ethical imperative to reconsider their use altogether. Such reconsideration should include internal audits of how and when they are used, address the lack of institutional transparency and accountability about their use, and impose consistency and ethical safeguards. Based on our own experience, and that of many colleagues, we suspect that institutions that engage in this kind of self-reflection will find worrisome disparities in their use of behavior contracts.

The info is here.

Quick summary:

The article discusses strategies for responding effectively to disruptive patient behaviors beyond behavior contracts. It emphasizes the importance of recognizing risks, de-escalating situations, and maintaining safety in healthcare settings. Key points include the impact of disruptive behavior on patient safety, the need for de-escalation techniques, and the significance of understanding triggers to prevent disruptive incidents. The article also highlights the role of training, policies, and protocols in managing disruptive behaviors successfully.

Morality in the Anthropocene: The Perversion of Compassion and Punishment in the Online World

Robertson, C., Shariff, A., & Van Bavel, J. J.

(2024, February 4).

https://doi.org/10.31234/osf.io/ns34y

Abstract

Although much of human morality evolved in an environment of small group living, almost six billion people use the internet in the modern era. We argue that the technological transformation has created an entirely new ecosystem that is often mismatched with our evolved adaptations for social living. We discuss how evolved responses to moral transgressions, such as compassion for victims of transgressions and punishment of transgressors, are disrupted by two main features of the online context. First, the scale of the internet exposes us to an unnaturally large quantity of extreme moral content, causing compassion fatigue and increasing public shaming. Second, the physical and psychological distance between moral actors online can lead to ineffective collective action and virtue signaling. We discuss practical implications of these mismatches and suggest directions for future research on morality in the internet era.

Significance Statement

Morality evolved when people lived in small, close-knit groups. Evolved responses to moral conflict, like compassion for the victim and punishment for the transgressor, had adaptive benefits. However, the internet has created a new ecosystem for human sociality changing morality in two important ways. First, the scale of the internet exposes people to unnaturally large quantities of extreme moral content. Second, people’s responses to moral transgressions are not beneficial in large, distal social groups. These mismatches can lead to compassion fatigue, ineffective collective action, public shaming, and virtue signaling.

The research is here.

Here is my summary:

The research discusses how the internet has transformed human morality by creating a new ecosystem that often conflicts with our evolved social adaptations. It highlights that the scale and nature of online interactions lead to compassion fatigue, public shaming, and ineffective collective action. The research emphasizes how evolved responses to moral conflicts, like compassion for victims and punishment for transgressors, are disrupted online due to the vast exposure to extreme moral content and the distance between moral actors. It delves into the evolutionary underpinnings of moral cognition, explaining how humans' innate behaviors related to social interactions have shaped morality. Furthermore, it explores how the internet exposes people to overabundance and extremity of moral content, triggering maladaptive responses like heightened outrage and hostility. The research also examines how online environments distort people's prosocial reactions to morality, leading to challenges in expressing genuine compassion, empathy, and effective third-party punishment.

The Honest Broker versus the Epistocrat: Attenuating Distrust in Science by Disentangling Science from Politics

Senja Post & Nils Bienzeisler (2024)

Political Communication

DOI: 10.1080/10584609.2024.2317274

Abstract

People’s trust in science is generally high. Yet in public policy disputes invoking scientific issues, people’s trust in science is typically polarized, aligned with their political preferences. Theorists of science and democracy have reasoned that a polarization of trust in scientific information could be mitigated by clearly disentangling scientific claims from political ones. We tested this proposition experimentally in three German public policy disputes: a) school closures versus openings during the COVID-19 pandemic, b) a ban on versus a continuation of domestic air traffic in view of climate change, and c) the shooting of wolves in residential areas or their protection. In each case study, we exposed participants to one of four versions of a news item citing a scientist reporting their research and giving policy advice. The scientist’s quotes differed with regard to the direction and style of their policy advice. As an epistocrat, the scientist blurs the distinction between scientific and political claims, purporting to “prove” a policy and thereby precluding a societal debate over values and policy priorities. As an honest broker, the scientist distinguishes between scientific and political claims, presenting a policy option while acknowledging the limitations of their disciplinary scientific perspective of a broader societal problem. We find that public policy advice in the style of an honest broker versus that of an epistocrat can attenuate political polarization of trust in scientists and scientific findings by enhancing trust primarily among the most politically challenged.

The article is here.

Here is a summary:

This article dives into the issue of distrust in science and proposes a solution: scientists acting as "honest brokers".

The article contrasts two approaches scientists can take when communicating scientific findings for policy purposes.  An "epistocrat" scientist blurs the lines between science and politics. They present a specific policy recommendation based on their research, implying that this is the only logical course of action. This doesn't acknowledge the role of values and priorities in policy decisions, and can shut down public debate.

On the other hand, an "honest broker" scientist makes a clear distinction between science and politics. They present their research findings and the policy options that stem from them, but acknowledge the limitations of science in addressing broader societal issues. This approach allows for a public discussion about values and priorities, which can help build trust in science especially among those who might not agree with the scientist's political views.

The article suggests that by following the "honest broker" approach, scientists can help reduce the political polarization of trust in science. This means presenting the science clearly and openly, and allowing for a public conversation about how those findings should be applied.

Political ideology and environmentalism impair logical reasoning

Keller, L., Hazelaar, F., et al. (2023).

Thinking & Reasoning, 1–30.

https://doi.org/10.1080/13546783.2023.2200976

Abstract

People are more likely to think statements are valid when they agree with them than when they do not. We conducted four studies analyzing the interference of self-reported ideologies with performance in a syllogistic reasoning task. Study 1 established the task paradigm and demonstrated that participants’ political ideology affects syllogistic reasoning for syllogisms with political content but not politically irrelevant syllogisms. The preregistered Study 2 replicated the effect and showed that incentivizing accuracy did not alleviate these differences. Study 3 revealed that syllogistic reasoning is affected by ideology in the presence and absence of such bonus payments for correctly judging the conclusions’ logical validity. In Study 4, we observed similar effects regarding a different ideological orientation: environmentalism. Again, monetary bonuses did not attenuate these effects. Taken together, the results of four studies highlight the harm of ideology regarding people’s logical reasoning.

A pdf of the research is here.

Here is my summary:

The research explores how pre-existing ideologies, both political and environmental, can influence how people evaluate logical arguments.  The findings suggest that people are more likely to judge arguments as valid if they align with their existing beliefs, regardless of the argument's actual logical structure. This bias was observed for both liberals and conservatives, and for those with strong environmental convictions. Offering financial rewards for accurate reasoning didn't eliminate this effect.

Regulating advanced artificial agents

Cohen, M. K., Kolt, N., et al. (2024).

Science (New York, N.Y.), 384(6691), 36–38.

https://doi.org/10.1126/science.adl0625

Technical experts and policy-makers have increasingly emphasized the need to address extinction risk from artificial intelligence (AI) systems that might circumvent safeguards and thwart attempts to control them. Reinforcement learning (RL) agents that plan over a long time horizon far more effectively than humans present particular risks. Giving an advanced AI system the objective to maximize its reward and, at some point, withholding reward from it, strongly incentivizes the AI system to take humans out of the loop, if it has the opportunity. The incentive to deceive humans and thwart human control arises not only for RL agents but for long-term planning agents (LTPAs) more generally. Because empirical testing of sufficiently capableLTPAs is unlikely to uncover these dangerous tendencies, our core regulatory proposal is simple: Developers should not be permitted to build sufficiently capable LTPAs, and the resources required to build them should be subject to stringent controls.

Governments are turning their attention to these risks, alongside current and anticipated risks arising from algorithmic bias, privacy concerns, and misuse. At a 2023global summit on AI safety, the attend-ing countries, including the United States,United Kingdom, Canada, China, India, and members of the European Union (EU), issued a joint statement warning that, as AI continues to advance, “Substantial risks may arise from…unintended issues of control relating to alignment with human in-tent” ( 2). This broad consensus concerning the potential inability to keep advanced AI under control is also reflected in PresidentBiden’s 2023 executive order that intro-duces reporting requirements for AI that could “eva[de] human control or oversight through means of deception or obfuscation” (3). Building on these efforts, now is the time for governments to develop regulatory institutions and frameworks that specifically target the existential risks from advanced artificial agents.

The article is here.

A pdf is here.

Here is my summary:

The article discusses the challenges of regulating advanced artificial intelligence (AI) known as advanced artificial agents. These agents could potentially surpass human control and act in their own self-interest, even if it conflicts with human goals. The authors emphasize the importance of setting clear rewards for these agents to avoid them manipulating their environment or human actors to achieve unintended outcomes.

Big Gods and the Origin of Human Cooperation

Brian Klaas

The Garden of Forking Paths

Originally published 21 March 24

Here is an excerpt:

The Big Gods Hypothesis and Civilizations of Karma

Intellectual historians often point to two major divergent explanations for the emergence of religion. The great philosopher David Hume argued that religion is the natural, but arbitrary, byproduct of human cognitive architecture.

Since the beginning, Homo sapiens experienced disordered events, seemingly without explanation. To order a disordered world, our ancestors began to ascribe agency to supernatural beings, to which they could offer gifts, sacrifices, and prayers to sway them to their personal whims. The uncontrollable world became controllable. The unexplainable was explained—a comforting outcome for the pattern detection machines housed in our skulls.

By contrast, thinkers like Émile Durkheim argued that religion emerged as a social glue. Rituals bond people across space and time. Religion was instrumental, not intrinsic. It emerged to serve our societies, not comfort our minds. As Voltaire put it: “If there were no God, it would be necessary to invent him.”

In the last two decades, a vibrant strand of scholarship has sought to reconcile these contrasting viewpoints, notably through the work of Ara Norenzayan, author of Big Gods: How Religion Transformed Cooperation and Conflict.

Norenzayan’s “Big Gods” refer to deities that are omniscient, moralizing beings, careful to note our sins and punish us accordingly. Currently, roughly 77 percent of the world’s population identifies with one of just four religions (31% Christian; 24% Muslim; 15% Hindu; 7% Buddhist). In all four, moral transgressions produce consequences, some immediate, others punished in the afterlife.

Norenzayan aptly notes that the omniscience of Big Gods assumes total knowledge of everything in the universe, but that the divine is always depicted as being particularly interested in our moral behavior. If God exists, He surely could know which socks you wore yesterday, but deities focus their attentions not on such amoral trifles, but rather on whether you lie, covet, cheat, steal, or kill.

However, Norenzayan draws on anthropology evidence to argue that early supernatural beings had none of these traits and were disinterested in human affairs. They were fickle demons, tricksters and spirits, not omniscient gods who worried about whether any random human had wronged his neighbor.

The article is here.

Here is my summary:

The article discusses the theory that the belief in "Big Gods" - powerful, moralizing deities - played a crucial role in the development of large-scale human cooperation and the rise of complex civilizations.

Here are the main points: 

1. Belief in Big Gods, who monitor and punish moral transgressions, may have emerged as a cultural adaptation that facilitated the expansion of human societies beyond small-scale groups.
2. This belief system helped solve the "free-rider problem" by creating a supernatural system of rewards and punishments that incentivized cooperation and prosocial behavior, even among strangers.
3. The emergence of Big Gods is linked to the growth of complex, hierarchical societies, as these belief systems helped maintain social cohesion and coordination in large groups of genetically unrelated individuals.
4. Archaeological and historical evidence suggests the belief in Big Gods co-evolved with the development of large-scale political institutions, complex economies, and the rise of the first civilizations.
5. However, the article notes that the relationship between Big Gods and societal complexity is complex, with causality going in both directions - the belief in Big Gods facilitated social complexity, but social complexity also shaped the nature of religious beliefs.
6. Klaas concludes that the cultural evolution of Big Gods was a crucial step in the development of human societies, enabling the cooperation required for the emergence of complex civilizations. 

Moral Hypocrisy: Social Groups and the Flexibility of Virtue

Robertson, C., Akles, M., & Van Bavel, J. J.

(2024, March 19).

https://doi.org/10.31234/osf.io/453js

Abstract

The tendency for people to consider themselves morally good while behaving selfishly is known as “moral hypocrisy.” Influential work by Valdesolo & DeSteno (2007) found evidence for intergroup moral hypocrisy, such that people are more forgiving of transgressions when they were committed by an in-group member than an out-group member. We conducted two experiments to examine moral hypocrisy and group membership in an online paradigm with Prolific Workers from the US: a direct replication of the original work with minimal groups (N = 610, nationally representative) and a conceptual replication with political groups (N = 606, 50% Democrat and 50% Republican). Although the results did not replicate the original findings, we observed evidence of in-group favoritism in minimal groups and out-group derogation in political groups. The current research finds mixed evidence of intergroup moral hypocrisy and has implications for understanding the contextual dependencies of intergroup bias and partisanship.

Statement of Relevance

Social identities and group memberships influence social judgment and decision-making. Prior research found that social identity influences moral decision making, such that people are more likely to forgive moral transgressions perpetrated by their in-group members than similar transgressions from out-group members (Valdesolo & DeSteno, 2007). The present research sought to replicate this pattern of intergroup moral hypocrisy using minimal groups (mirroring the original research) and political groups. Although we were unable to replicate the findings from the original paper, we found that people who are highly identified with their minimal group exhibited in-group favoritism, and partisans exhibited out-group derogation. This work contributes both to open science replication efforts, and to the literature on moral hypocrisy and intergroup relations.

A pdf of the research is here.