Rationalization

Ehsan, Upol; Harrison, Brent; Chan, Larry; Riedl, Mark O.

doi:10.1145/3278721.3278736

Cited by 67 publications

(11 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In order to avoid problems arising from ambiguity, we reflect on our definition. We follow Ehsan et al (2018) who define AI rationalization as a process of generating rationales of a model's behavior as if a human had performed the behavior.…”

Section: Discussion and Future Directionsmentioning

confidence: 99%

Natural Language Rationales with Full-Stack Visual Reasoning: From Pixels to Semantic Frames to Commonsense Graphs

Marasović¹,

Bhagavatula

Park

et al. 2020

Findings of the Association for Computational Linguistics: EMNLP 2020

View full text Add to dashboard Cite

Natural language rationales could provide intuitive, higher-level explanations that are easily understandable by humans, complementing the more broadly studied lower-level explanations based on gradients or attention weights. We present the first study focused on generating natural language rationales across several complex visual reasoning tasks: visual commonsense reasoning, visual-textual entailment, and visual question answering. The key challenge of accurate rationalization is comprehensive image understanding at all levels: not just their explicit content at the pixel level, but their contextual contents at the semantic and pragmatic levels. We present RATIONALE VT TRANSFORMER, an integrated model that learns to generate free-text rationales by combining pretrained language models with object recognition, grounded visual semantic frames, and visual commonsense graphs. Our experiments show that free-text rationalization is a promising research direction to complement model interpretability for complex visual-textual reasoning tasks. In addition, we find that integration of richer semantic and pragmatic visual features improves visual fidelity of rationales.

show abstract

Section: Discussion and Future Directionsmentioning

confidence: 99%

Natural Language Rationales with Full-Stack Visual Reasoning: From Pixels to Semantic Frames to Commonsense Graphs

Marasović¹,

Bhagavatula

Park

et al. 2020

Findings of the Association for Computational Linguistics: EMNLP 2020

View full text Add to dashboard Cite

show abstract

“…Regarding the investigation of best explanations, some studies rely on user evaluation, namely users' subjective opinions expressed in surveys or interviews. Different types of measurements have been proposed, measuring user satisfaction [10], acceptance [11], trust [12] or the goodness of an explanation [13]. Miller et al [14] state that humans are more likely to accept explanations that are consistent with their prior beliefs.…”

Section: Introductionmentioning

confidence: 99%

Comparing User Perception of Explanations Developed with XAI Methods

Aechtner

Cabrera

Katwal

et al. 2022

2022 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE)

View full text Add to dashboard Cite

published version features the final layout of the paper including the volume, issue and page numbers. Link to publication General rightsCopyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.• Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal.If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the "Taverne" license above, please follow below link for the End User

show abstract

“…Scientists and designers need to improve our understanding of how well people can distinguish between actual humans and current state-of-the-art AI on the basis of behavior within complex environments. As AI begins to incorporate humanlike behavioral traits and motivations, the result may be that the information from these actors is more comprehensible and satisfying (Romero et al, 2017;Ehsan et al, 2018), yet there will be the potential for other humans to use the technology in ways that pose serious threats to our society (Zaleski, 2016;Schwartz, 2018;Huhn, 2019;Wagner and Blewer, 2019).…”

Section: Discussionmentioning

confidence: 99%

“…Videogames provide the perfect environment to study how humans interact with AI because they are already developed for rich social interactions, and many provide the opportunity for researchers to build systems to capture behavioral data from within the game through custom modifications. Videogames have been used both as a platform for training and evaluating AI (Laird and VanLent, 2001;Mnih et al, 2015;Bard et al, 2020), and as a way to investigate how human and AI performances are perceived by measuring human behavior and subjective experience (Ehsan et al, 2018;Tulk et al, 2018) as well as physiological measures (Lim and Reeves, 2009).…”

Section: Introductionmentioning

confidence: 99%

Behavioral Cues of Humanness in Complex Environments: How People Engage With Human and Artificially Intelligent Agents in a Multiplayer Videogame

2020

View full text Add to dashboard Cite

The development of AI that can socially engage with humans is exciting to imagine, but such advanced algorithms might prove harmful if people are no longer able to detect when they are interacting with non-humans in online environments. Because we cannot fully predict how socially intelligent AI will be applied, it is important to conduct research into how sensitive humans are to behaviors of humans compared to those produced by AI. This paper presents results from a behavioral Turing Test, in which participants interacted with a human, or a simple or “social” AI within a complex videogame environment. Participants (66 total) played an open world, interactive videogame with one of these co-players and were instructed that they could interact non-verbally however they desired for 30 min, after which time they would indicate their beliefs about the agent, including three Likert measures of how much participants trusted and liked the co-player, the extent to which they perceived them as a “real person,” and an interview about the overall perception and what cues participants used to determine humanness. T -tests, Analysis of Variance and Tukey's HSD was used to analyze quantitative data, and Cohen's Kappa and χ 2 was used to analyze interview data. Our results suggest that it was difficult for participants to distinguish between humans and the social AI on the basis of behavior. An analysis of in-game behaviors, survey data and qualitative responses suggest that participants associated engagement in social interactions with humanness within the game.

show abstract

Rationalization

Cited by 67 publications

References 11 publications

Natural Language Rationales with Full-Stack Visual Reasoning: From Pixels to Semantic Frames to Commonsense Graphs

Natural Language Rationales with Full-Stack Visual Reasoning: From Pixels to Semantic Frames to Commonsense Graphs

Comparing User Perception of Explanations Developed with XAI Methods

Behavioral Cues of Humanness in Complex Environments: How People Engage With Human and Artificially Intelligent Agents in a Multiplayer Videogame

Contact Info

Product

Resources

About