Reinforcement learning with rare significant events

Ecoffet, Paul; Fontbonne, Nicolas; André, Jean‐Baptiste; Bredèche, Nicolas

doi:10.1145/3449726.3459462

Cited by 3 publications

(2 citation statements)

References 3 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A related principle is that of uncertainty: There may be different conceptions of loving action that are credible but differ in recommendations for particular situations. Similar to the idea of acting under moral uncertainty [104,105], an ML system could be given multiple independent implementations of machine love, and incentivized to act towards where those implementations agree, and away from what any implementation suggests is deeply harmful.…”

Section: Concerns About Unintended Impacts From Optimizationmentioning

confidence: 99%

Machine Love

Lehman¹

2023

Preprint

View full text Add to dashboard Cite

From social media to recommendation and search engines, the impact of machine learning (ML) upon our lives and society continues to grow. While ML generates much economic value, many of us have problematic relationships with social media and other ML-powered applications. One reason is that ML often optimizes for what we want in the moment, which is easy to quantify but at odds with what is known scientifically about human flourishing. Thus, through its impoverished models of us, ML currently falls far short of its exciting potential, which is for it to help us to reach ours. While there is no consensus on defining human flourishing, from diverse perspectives across psychology, philosophy, and spiritual traditions, love is understood to be one of its primary catalysts. Motivated by this view, this paper explores whether there is a useful conception of love fitting for machines to embody. While such machine love might seem like an error of category, historically it has been generative to explore whether some kernel of a nebulous concept, such as life or intelligence, can be thoughtfully abstracted and reimagined in a different medium, as in the fields of machine intelligence or artificial life. This paper forwards a candidate conception of machine love, inspired in particular by work in positive psychology and psychotherapy: to provide unconditional support enabling humans to autonomously pursue their own growth and development. While many alternate conceptions are possible, this one benefits from not requiring machines to simulate emotional affect or relationships, instead focusing on biasing an ML system's actions to support our growth. Through proof of concept experiments, this paper aims to highlight the need for richer models of human flourishing in ML, provide an example framework through which positive psychology can be combined with ML to realize a rough conception of machine love, and demonstrate that current language models begin to enable embodying qualitative humanistic principles. The conclusion is that though at present ML may often serve to addict, distract, or divide us, an alternative path may be opening up: We may align ML to support our growth, through it helping us to align ourselves towards our highest aspirations. * Much of this work was completed during a residency at Stochastic Labs.Preprint. Under review.

show abstract

Section: Concerns About Unintended Impacts From Optimizationmentioning

confidence: 99%

Machine Love

Lehman¹

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Currently, there are no designs for general ethics modules to be added to any AGI, or, more specifically, ones that try to implement utilitarian ethics. There have been other possible solutions to the alignment problem suggested, models that use AI to learn from ethical environments (Rodriguez-Soto et al, 2021), models to balance competing moral considerations under moral uncertainty (Ecoffet & Lehman, 2021), and attempts to build ethics into robots (Anderson & Anderson, 2010;Winfield et al, 2014;Vanderelst & Winfield, 2018;Bremner et al, 2019).…”

Section: Related Workmentioning

confidence: 99%