ProsocialDialog: A Prosocial Backbone for Conversational Agents

Abstract

Most existing dialogue systems fail to respond properly to potentially unsafe user utterances by either ignoring or passively agreeing with them. To address this issue, we introduce PROSOCIALDIALOG, the first large-scale multi-turn dialogue dataset to teach conversational agents to respond to problematic content following social norms. Covering diverse unethical, problematic, biased, and toxic situations, PROSOCIALDIALOG contains responses that encourage prosocial behavior, grounded in commonsense social rules (i.e., rules-ofthumb, RoTs). Created via a human-AI collaborative framework, PROSOCIALDIALOG consists of 58K dialogues, with 331K utterances, 160K RoTs, and 497K dialogue safety labels accompanied by free-form rationales.With this dataset, we introduce a dialogue safety detection module, Canary, capable of generating RoTs given conversational context, and a socially-informed dialogue agent, Prost. Empirical results show that Prost generates more socially acceptable dialogues compared to other state-of-the-art language and dialogue models in both in-domain and outof-domain settings. Additionally, Canary effectively guides conversational agents and offthe-shelf language models to generate significantly more prosocial responses. Our work highlights the promise and importance of creating and steering conversational AI to be socially responsible.

Datasets for Large Language Models: A Comprehensive Survey

“…Social Norms Domain. PROSOCIALDIALOG (Kim et al, 2022) is a multiturn English conversation dataset that instructs models to respond to problematic inputs according to human social norms. The dataset covers various unethical, problematic, biased, and harmful scenarios, created using a human-machine collaboration framework.…”

Designing, Evaluating, and Learning from Humans Interacting with NLP Models

“…It represents the use case where evaluation is dynamic (1) the model performance is easily swayed by human responses and can hardly be measured on benchmark datasets (Li et al, 2021), ( 2) the model has to balance multiple criteria like interestingness, informativeness, etc. which could be subjective for different user groups (Thoppilan et al, 2022), (3) it is essential to implement fallback options (e.g., responses like "sorry I didn't understand" that's built around the model at the UI level) when the model does not behave as expected or safety modules when there is potential for controversiality (Kim et al, 2022). These properties also make dialog systems an ideal testbed for discussing UI designs ( §2.3) and personalization ( §2.4).…”

DialGuide: Aligning Dialogue Model Behavior with Developer Guidelines

“…Other approaches include strategies for responding to problematic contexts, such as steering away from toxicity (Baheti et al, 2021;Arora et al, 2022), using apologies (Ung et al, 2022), and non-sequiturs (Xu et al, 2021b). Our work is closely related to a study that proposed ProSocial dialog, a dataset where speakers disagree with unethical and toxic contexts using safety labels and social norms (Kim et al, 2022a). Using guidelines allows for more fine-grained control by specifying the contexts they are relevant to, and can provide more informative responses.…”

“…We use Blenderbot (Roller et al, 2021) to generate 3 additional responses for each context, creating a set of four responses including the original response from the dataset, denoted as R b , which is used in tasks A) and C) below. DI-ALGUIDE-SAFETY consists of data for the safety domain, where we augment conversations from the ProsocialDialog (Kim et al, 2022a) dataset. A) Guideline writing task.…”

“…In the DIALGUIDE framework * , we benchmark three tasks: 1) Guideline selection, where a model needs to retrieve context-relevant guidelines, 2) Response generation, where a model generates a response that follows a selected guideline, and 3) Response entailment verification, where the model determines whether a response follows or violates a guideline. We augment conversations from existing dialogue datasets -Blended Skills talk (Smith et al, 2020) and ProsocialDialog (Kim et al, 2022a) by collecting annotations of 1) relevant/irrelevant guidelines to the conversation context and 2) responses following/violating the guideline. To test the models' semantic understanding, we also create adversarial train and test sets.…”