Harnessing human and machine intelligence for planetary-level climate action

npj Climate Action, Journal Year: 2023, Volume and Issue: 2(1)

Published: Aug. 17, 2023

Abstract The ongoing global race for bigger and better artificial intelligence (AI) systems is expected to have a profound societal environmental impact by altering job markets, disrupting business models, enabling new governance welfare structures that can affect consensus climate action pathways. However, the current AI are trained on biased datasets could destabilize political agencies impacting change mitigation adaptation decisions compromise social stability, potentially leading tipping events. Thus, appropriate design of less system reflects both direct indirect effects societies planetary challenges question paramount importance. In this paper, we tackle data-centric knowledge generation in ways minimize AI. We argue need co-align with an epistemic web health more trustworthy decision-making. A human-in-the-loop be designed align three goals. First, it contribute supports action. Second, directly enable interventions through elements. Finally, reduce data injustices associated pretraining datasets.

Language: Английский

---

AI Governance and the Policymaking Process: Key Considerations for Reducing AI Risk

Big Data and Cognitive Computing, Journal Year: 2019, Volume and Issue: 3(2), P. 26 - 26

Published: May 8, 2019

This essay argues that a new subfield of AI governance should be explored examines the policy-making process and its implications for governance. A growing number researchers have begun working on question how to mitigate catastrophic risks transformative artificial intelligence, including what policies states adopt. However, this identifies preceding, meta-level problem space possible is affected by politics administrative mechanisms those are created implemented. creates set key considerations field influence action future policymakers. some theories policymaking process, they compare current work in governance, their at large ends identifying areas research.

Language: Английский

---

International Politics in the Age of Existential Threats

Journal of Global Security Studies, Journal Year: 2020, Volume and Issue: 6(3)

Published: May 16, 2020

Abstract Humans in the twenty-first century live under specter of anthropogenic existential threats to human civilization and survival. What is significance humanity’s capacity for self-destruction meaning “security” “survival” international politics? The argument that it constitutes a material “revolution” politics—that is, growing spectrum represents radical transformation context politics turns established truths about security survival on their heads. paper develops theoretical framework based historical materialism, especially proposition circumstances “forces destruction” determine viability different “modes protection”, political “units” “structures”, “security ideologies” politics. seeks demonstrate disjuncture (or “contradiction”) between (“forces destruction”); practices war, use military force, balance-of-power (“modes protection”); units nation-states structure anarchy (“political superstructure”); primacy “national security” doctrines “self-help” “power politics” (“security ideologies”). Specifically, humanityapos;s interdependence with respect calls into question centrality national In an age threats, better understood as humanity.

Language: Английский

---

Reward tampering problems and solutions in reinforcement learning: a causal influence diagram perspective

Synthese, Journal Year: 2021, Volume and Issue: 198(S27), P. 6435 - 6467

Published: May 19, 2021

Language: Английский

---

Robotics, AI, and Humanity

Springer eBooks, Journal Year: 2021, Volume and Issue: unknown

Published: Jan. 1, 2021

Language: Английский

---

‘Solving for X?’ Towards a problem-finding framework to ground long-term governance strategies for artificial intelligence

Futures, Journal Year: 2021, Volume and Issue: 126, P. 102672 - 102672

Published: Jan. 7, 2021

Language: Английский

---

Aligning Generative Language Models with Human Values

Findings of the Association for Computational Linguistics: NAACL 2022, Journal Year: 2022, Volume and Issue: unknown

Published: Jan. 1, 2022

Although current large-scale generative language models (LMs) can show impressive insights about factual knowledge, they do not exhibit similar success with respect to human values judgements (e.g., whether or the generations of an LM are moral). Existing methods learn either by directly mimicking behavior data, rigidly constraining generation space human-chosen tokens. These inherently limited in that consider contextual and abstract nature as a result often fail when dealing out-of-domain context sophisticated values.This paper proposes SENSEI, new reinforcement learning based method embed into each step generation. SENSEI deploys Actor-Critic framework, where Critic is reward distributor simulates assignment procedure humans, while Actor guides towards maximum direction. Compared five existing three alignment datasets, only achieves higher performance terms both automatic evaluations, but also shows improvements on robustness transfer unseen values.

Language: Английский

---

State of science: models and methods for understanding and enhancing teams and teamwork in complex sociotechnical systems

Ergonomics, Journal Year: 2021, Volume and Issue: 65(2), P. 161 - 187

Published: Dec. 6, 2021

This state of the science review brings together disparate literature effective strategies for enhancing and accelerating team performance. The evaluates synthesises models proposes recommended avenues future research. two major Input-Mediator-Output-Input (IMOI) framework Big Five dimensions teamwork were reviewed both will need significant development application to teams comprising non-human agents. Research suggests that a multi-method approach is appropriate measurements, such as integration methods from self-report, observer ratings, event-based measurement automated recordings. Simulations are most team-based training interventions. impact new technology autonomous agents discussed with respect changing nature teamwork. In particular, whether existing measures suitable support design, operation evaluation human-nonhuman future. Practitioner summary: recommends Team be adapted describe interaction agents, which what likely hold. interventions use simulation-based approaches.

Language: Английский

---

AI safety: state of the field through quantitative lens

Published: Sept. 28, 2020

Last decade has seen major improvements in the performance of artificial intelligence which driven wide-spread applications. Unforeseen effects such massad-option put notion AI safety into public eye. is a relatively new field research focused on techniques for building beneficial humans. While there exist survey papers safety, lack quantitative look at being conducted. The aspect gives data-driven insight about emerging trends, knowledge gaps and potential areas future research. In this paper, bibliometric analysis literature finds significant increase activity since 2015. Also, so that most technical issues are open, including: explainability its long-term utility, value alignment we have identified as important topic. Equally, severe concrete policies regarding AI. As expect to be one main driving forces changes, under need decide direction humanity's future.

Language: Английский

---

Multilayered review of safety approaches for machine learning-based systems in the days of AI

Journal of Systems and Software, Journal Year: 2021, Volume and Issue: 176, P. 110941 - 110941

Published: March 6, 2021

Language: Английский

---