Would you Rather? A New Benchmark for Learning Machine Alignment with Cultural Values and Social Preferences

Would you Rather? A New Benchmark for Learning Machine Alignment with Cultural Values and Social Preferences †Yi Tay∗, ♭Donovan Ong∗, ♯Jie Fu, †Alvin Chan, ♭Nancy F. Chen ∗φLuu Anh Tuan, ♯‡Christopher Pal †Nanyang Technological University, Singapore ♯Polytechnique Montreal, Mila, ‡Canada CIFAR AI Chair ♭A*STAR, Singapore, ∗MIT CSAIL, φVinAI Research ytay017@gmail.com, ongyl@i2r.a-star.edu.sg Abstract Understanding human preferences, along with cultural and social nuances, lives at the heart of natural language understanding. Concretely, we present a new task and corpus for learn- ing alignments between machine and human preferences. Our newly introduced problem is concerned with predicting the preferable op- tions from two sentences describing scenar- ios that may involve social and cultural situ- ations. Our problem is framed as a natural lan- guage inference task with crowd-sourced pref- erence votes by human players, obtained from a gamified voting platform. We benchmark several state-of-the-art neural models, along with BERT and friends on this task. Our ex- perimental results show that current state-of- the-art NLP models still leave much room for improvement. 1 Introduction The ability to understanding social nuances and human preferences is central to natural language understanding. This also enables better alignment of machine learning models with human values, eventually leading to better human-compatible AI applications (Peterson et al., 2019; Leslie, 2019; Rosenfeld and Kraus, 2018; Amodei et al., 2016; Russell and Norvig, 2016). There exist a plethora of work on studying opti- mal decision-making under a variety of situations (Edwards, 1954; Bottom, 2004; Plonsky et al., 2019; Peterson et al., 2019). On the other hand, cognitive models of human decision-making are usually based on small datasets (Peterson et al., 2019). Furthermore, these studies tend to only consider individuals in isolation. In contrast, we investigate the influence of cultural and social nu- ances for choice prediction at scale. In other words, we study the social preference as a whole, ∗ First two authors contributed equally not those of an individual in isolation, which is ar- guably more challenging and largely unexplored. In this work, we propose a new benchmark dataset with a large number of 200k data points, Machine Alignment with Cultural values and Social preferences (MACS), for learning AI align- ment with humans. Our dataset is based on a pop- ular gamified voting platform, namely the game of ‘would you rather?’. In this game, participants are given two choices and vote for the more preferable option. Examples from our dataset can be found at Table 1. To the best of our knowledge, our work is the first work to incorporate voting-based lan- guage games as a language understanding bench- mark. In many ways, our benchmark dataset is remi- niscent of the natural language inference problem (MacCartney, 2009; Bowman et al., 2015), social commonsense reasoning (Sap et al., 2019) or other natural language understanding problems (Wang et al., 2018; Zellers et al., 2018). To this end, our problem is framed in a way that enables con- venient benchmarking of existing state-of-the-art NLU models such as BERT (Devlin et al., 2018) or RoBERTa …

Content truncated to 500 words