Are Large Language Models Moral Hypocrites? A Study Based on Moral Foundations (artigo / 2024)

 

Large language models (LLMs) have taken centre stage in debates on Artificial Intelligence. Yet there remains a gap in how to assess LLMs' conformity to important human values. In this paper, we investigate whether state-of-the-art LLMs, GPT-4 and Claude 2.1 (Gemini Pro and LLAMA 2 did not generate valid results) are moral hypocrites. We employ two research instruments based on the Moral Foundations Theory: (i) the Moral Foundations Questionnaire (MFQ), which investigates which values are considered morally relevant in abstract moral judgements; and (ii) the Moral Foundations Vignettes (MFVs), which evaluate moral cognition in concrete scenarios related to each moral foundation. 

We characterise conflicts in values between these different abstractions of moral evaluation as hypocrisy. We found that both models displayed reasonable consistency within each instrument compared to humans, but they displayed contradictory and hypocritical behaviour when we compared the abstract values present in the MFQ to the evaluation of concrete moral violations of the MFV.

[ PDF ]

© Como citar este artigo:

Nunes, J. L., Almeida, G. F. C. F., Araujo, M. de, & Barbosa, S. D. J. (2024). Are Large Language Models Moral Hypocrites? A Study Based on Moral Foundations. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, 7(1), 1074-1087. https://doi.org/10.1609/aies.v7i1.31704 

Futuro e desigualdade: O que é geoengenharia do clima e como ela pode influir na chamada justiça intergeracional (artigo / 2024)

 

Artigo para The Conversation Brasil.

Trechos:

"A catástrofe climática que atingiu o Rio Grande do Sul, entre abril e junho de 2024, precipitou um intenso debate sobre as consequências das mudanças climáticas no Brasil. Muito se fala agora sobre o modo desigual como as mudanças climáticas atingem populações mais vulneráveis."

"Grupos sociais que já eram vítimas de injustiças sociais por conta de graves desigualdades econômicas se veem agora ainda mais fragilizados diante de desastres climáticos. Sem políticas climáticas específicas, problemas de injustiça social tendem a se agravar no Brasil." 

"O que fazer? Que políticas para a promoção da justiça social são necessárias, isso ninguém negaria. O que nem todo mundo percebe, porém, é que desigualdades sociais e econômicas constituem apenas um aspecto das questões de justiça que emergem em decorrência das mudanças climáticas." 

"Desigualdades de poder na relação entre gerações são igualmente importantes. No entanto, pouca atenção tem sido dada a questões de justiça intergeracional nos debates sobre mudanças climáticas no Brasil. Isso é estranho, pois a injustiça intergeracional é tão preocupante quanto a injustiça social."

[ artigo completo ]

© Como citar este artigo: 

Araujo, Marcelo de; Fior, Pedro; Menezes, Pedro. 2024. "Futuro e desigualdade: o que é geoengenharia do clima e como ela pode influir na chamada justiça intergeracional". In The Conversation Brasil, 23 de agosto de 2024. https://theconversation.com/futuro-e-desigualdade-o-que-e-geoengenharia-do-clima-e-como-ela-pode-influir-na-chamada-justica-intergeracional-237119

Republicado por:

Globo. Um Só Planeta

Bios. Ética & Ciência a Serviço da Vida, v.2, n. 11

History repeats itself. Next time as tragedy: The ethics of climate adaptation for the overshoot generation (pre-print article / 2024)

In spite of the widespread recognition that effective climate policies require both mitigation and adaptation, the history of the climate ethics debate shows that the balance between adaptation and mitigation policies has been anything but stable. Although some researchers did indeed support a well-balanced mix of mitigation and adaptation strategies in the Villach (1985) and Toronto (1988) conferences, throughout the 1980s most people favoured adaptation over mitigation. This contrasts with preference for mitigation over adaptation in the 1990s and early 2000s, almost to the point of entirely discrediting the relevance of climate adaptation policies. It was not until the early 2010s that adaptation and mitigation came to be seen as equally important to address climate change. 

In this paper, we argue that by mid-century adaptation policies are likely to be seen, again, as more urgent than mitigation. The reason for this is that most mitigation pathways to the Paris Agreement involve a temperature overshoot, which will constrain the generation living under the overshoot period – the “overshoot generation” – to prioritize adaptation policies as a matter of survival. Even on the assumption that the overshoot generation will emit less than the current and the past generations have emitted, the overshoot generation will have to deploy massively as-yet unproven carbon removal technologies in order to mitigate the unmitigated emissions of the past. We call this process “retroactive mitigation”. We conclude the paper with four policy recommendations for the benefit of the overshoot generation.

[ PDF ]

© Como citar este artigo:
Araujo, Marcelo de and Fior Mota de Andrade, Pedro, "History repeats itself. Next time as tragedy: The ethics of climate adaptation for the overshoot generation" (June 11, 2024). Available at SSRN: https://ssrn.com/abstract=4862217 or http://dx.doi.org/10.2139/ssrn.4862217.

Exploring the psychology of LLMs’ moral and legal reasoning (article / 2024)

 

Large language models (LLMs) exhibit expert-level performance in tasks across a wide range of different domains. Ethical issues raised by LLMs and the need to align future versions makes it important to know how state of the art models reason about moral and legal issues. In this paper, we employ the methods of experimental psychology to probe into this question. We replicate eight studies from the experimental literature with instances of Google's Gemini Pro, Anthropic's Claude 2.1, OpenAI's GPT-4, and Meta's Llama 2 Chat 70b. 

We find that alignment with human responses shifts from one experiment to another, and that models differ amongst themselves as to their overall alignment, with GPT-4 taking a clear lead over all other models we tested. Nonetheless, even when LLM-generated responses are highly correlated to human responses, there are still systematic differences, with a tendency for models to exaggerate effects that are present among humans, in part by reducing variance. This recommends caution with regards to proposals of replacing human participants with current state-of-the-art LLMs in psychological research and highlights the need for further research about the distinctive aspects of machine psychology.

[ PDF ]

© Como citar este artigo:

Almeida, Guilherme F.C.F., José Luiz Nunes, Neele Engelmann, Alex Wiegmann, and Marcelo de Araujo. 2024. "Exploring the psychology of LLMs’ moral and legal reasoning". Artificial Intelligence 333 (August):104145. https://doi.org/10.1016/j.artint.2024.104145.

Epistemology goes AI: A study of GPT-3’s capacity to generate consistent and coherent ordered sets of propositions on a Single-Input-Multiple-Outputs Basis (artigo / 2024))

 

The more we rely on digital assistants, online search engines, and AI systems to revise our system of beliefs and increase our body of knowledge, the less we are able to resort to some independent criterion, unrelated to further digital tools, in order to asses the epistemic reliability of the outputs delivered by them. This raises some important questions to epistemology in general and pressing questions to applied to epistemology in particular. 

In this paper, we propose an experimental method for the assessment of GPT-3’s capacity to generate consistent and coherent sets of outputs. When several outputs to one and the same input are very repetitive they tend to be consistent with each other, that is they do not contradict each other. But consistency does not make the set of outputs as a whole more informative than the outputs considered individually. We argue that the less informative a set of outputs is, the less coherent it is. We establish a conceptual distinction between consistency and coherence in the light of what some epistemologists refer to as a coherence theories of truth and justification. 

While much attention has been given to GPT-3’s capacity to produce internally coherent individual outputs, we argue, instead, that more attention should be given to its capacity to produce consistent and coherent outputs generated on a single-input-multiple-outputs basis.

[ PDF ]

© Como citar este artigo:

de Araujo, M., de Almeida, G. & Nunes, J.L. (2024) "Epistemology goes AI: A study of GPT-3’s capacity to generate consistent and coherent ordered sets of propositions on a single-input-multiple-outputs basis". Minds & Machines 34, 2 (2024). https://doi.org/10.1007/s11023-024-09660-6 

Promoting science communication for the purpose of pandemic preparedness and response: An assessment of the relevance of pre-COVID pandemic “early warnings” (artigo / 2024)

Given the abrupt global disruption caused by SARS-CoV-2, one might think that the COVID pandemic was an unpredictable event. But in the years leading up to the emergence of the COVID pandemic, several documents had already been warning of the increasing occurrences of new disease outbreaks with pandemic potential and lack of corresponding policies to promote pandemic preparedness and response. 

In this article, we call these documents “early warnings”. We argue that a survey of early warnings can help science communicators to promote the public understanding of evidence-based pandemic preparedness and response policies at local or international level. Our proposal differs from other approaches to pandemic preparedness and response in that it highlights the relevance of documents published before the COVID outbreak. We show that the early warnings did not become outdated after the COVID outbreak, but, rather, that they are even more pressing now.

© Como citar este artigo:

Araujo, Marcelo de and Costa, Daniel de Vasconcelos. "Promoting Science Communication for the Purpose of Pandemic Preparedness and Response: An Assessment of the Relevance of Pre-COVID Pandemic early warnings" Human Affairs, vol. 34, no. 2, 2024, pp. 269-294. https://doi.org/10.1515/humaff-2023-0083

Urgentes e necessárias, políticas de adaptação climática enfrentam dilemas éticos no Brasil e no mundo (artigo / 2024)

 

Artigo para The Conversation Brasil.

Trechos:

"A solução para se conter o aquecimento global é a mesma para todos os países: reduzir as emissões de gases de efeito estufa (GEE), especialmente o dióxido de carbono (CO2). Esta é uma medida de mitigação. Mas ainda que todos os países resolvessem implementar imediatamente políticas de mitigação radicais, a crise climática não seria resolvida de um dia para o outro. Para se lidar com as mudanças climáticas são necessárias também políticas de adaptação." 

"Diferentemente das políticas para mitigação, as estratégias de adaptação devem ser específicas para cada cenário regional. Isto significa que diferentes países e regiões exigirão políticas de adaptação distintas. Por exemplo, no contexto brasileiro, a escassez de água na Região Nordeste pode contrastar diretamente com o excesso de chuvas na Serra Fluminense, resultando na necessidade de ações adaptativas específicas para cada realidade." 

"Da interação entre políticas de adaptação e políticas de mitigação surgem dilemas éticos que ainda não foram devidamente discutidos no Brasil."

[ artigo ]

© Como citar este artigo:
Araujo, Marcelo de; Fior, Pedro. 2024. "Urgentes e necessárias, políticas de adaptação climática enfrentam dilemas éticos no Brasil e no mundo". In The Conversation Brasil, 5 de fevereiro de 2024. https://theconversation.com/urgentes-e-necessarias-politicas-de-adaptacao-climatica-enfrentam-dilemas-eticos-no-brasil-e-no-mundo-222130