We test the hypothesis that language models trained with reinforcement learning from human feedback (RLHF) have the capability to "morally self-correct" -- to avoid producing harmful outputs -- if instructed to do so. We find strong evidence in support of this hypothesis across three different experiments, each of which reveal different facets of moral self-correction. We find that the capability for moral self-correction emerges at 22B model parameters, and typically improves with increasing model size and RLHF training. We believe that at this level of scale, language models obtain two capabilities that they can use for moral self-correction: (1) they can follow instructions and (2) they can learn complex normative concepts of harm like stereotyping, bias, and discrimination. As such, they can follow instructions to avoid certain kinds of morally harmful outputs. We believe our results are cause for cautious optimism regarding the ability to train language models to abide by ethical principles.
Language models (LMs) are pretrained to imitate internet text, including content that would violate human preferences if generated by an LM: falsehoods, offensive comments, personally identifiable information, low-quality or buggy code, and more. Here, we explore alternative objectives for pretraining LMs in a way that also guides them to generate text aligned with human preferences. We benchmark five objectives for pretraining with human feedback across three tasks and study how they affect the trade-off between alignment and capabilities of pretrained LMs. We find a Pareto-optimal and simple approach among those we explored: conditional training, or learning distribution over tokens conditional on their human preference scores given by a reward model. Conditional training reduces the rate of undesirable content by up to an order of magnitude, both when generating without a prompt and with an adversarially-chosen prompt. Moreover, conditional training maintains the downstream task performance of standard LM pretraining, both before and after task-specific finetuning. Pretraining with human feedback results in much better preference satisfaction than standard LM pretraining followed by finetuning with feedback, i.e., learning and then unlearning undesirable behavior. Our results suggest that we should move beyond imitation learning when pretraining LMs and incorporate human preferences from the start of training.
Naturally-occurring information-seeking questions often contain questionable assumptions -- assumptions that are false or unverifiable. Questions containing questionable assumptions are challenging because they require a distinct answer strategy that deviates from typical answers to information-seeking questions. For instance, the question "When did Marie Curie discover Uranium?" cannot be answered as a typical when question without addressing the false assumption "Marie Curie discovered Uranium". In this work, we propose (QA)$^2$ (Question Answering with Questionable Assumptions), an open-domain evaluation dataset consisting of naturally-occurring search engine queries that may or may not contain questionable assumptions. To be successful on (QA)$^2$, systems must be able to detect questionable assumptions and also be able to produce adequate responses for both typical information-seeking questions and ones with questionable assumptions. We find that current models do struggle with handling questionable assumptions -- the best performing model achieves 59% human rater acceptability on abstractive QA with (QA)$^2$ questions, leaving substantial headroom for progress.
As language models (LMs) scale, they develop many novel behaviors, good and bad, exacerbating the need to evaluate how they behave. Prior work creates evaluations with crowdwork (which is time-consuming and expensive) or existing data sources (which are not always available). Here, we automatically generate evaluations with LMs. We explore approaches with varying amounts of human effort, from instructing LMs to write yes/no questions to making complex Winogender schemas with multiple stages of LM-based generation and filtering. Crowdworkers rate the examples as highly relevant and agree with 90-100% of labels, sometimes more so than corresponding human-written datasets. We generate 154 datasets and discover new cases of inverse scaling where LMs get worse with size. Larger LMs repeat back a dialog user's preferred answer ("sycophancy") and express greater desire to pursue concerning goals like resource acquisition and goal preservation. We also find some of the first examples of inverse scaling in RL from Human Feedback (RLHF), where more RLHF makes LMs worse. For example, RLHF makes LMs express stronger political views (on gun rights and immigration) and a greater desire to avoid shut down. Overall, LM-written evaluations are high-quality and let us quickly discover many novel LM behaviors.
As AI systems become more capable, we would like to enlist their help to supervise other AIs. We experiment with methods for training a harmless AI assistant through self-improvement, without any human labels identifying harmful outputs. The only human oversight is provided through a list of rules or principles, and so we refer to the method as 'Constitutional AI'. The process involves both a supervised learning and a reinforcement learning phase. In the supervised phase we sample from an initial model, then generate self-critiques and revisions, and then finetune the original model on revised responses. In the RL phase, we sample from the finetuned model, use a model to evaluate which of the two samples is better, and then train a preference model from this dataset of AI preferences. We then train with RL using the preference model as the reward signal, i.e. we use 'RL from AI Feedback' (RLAIF). As a result we are able to train a harmless but non-evasive AI assistant that engages with harmful queries by explaining its objections to them. Both the SL and RL methods can leverage chain-of-thought style reasoning to improve the human-judged performance and transparency of AI decision making. These methods make it possible to control AI behavior more precisely and with far fewer human labels.
Developing safe and useful general-purpose AI systems will require us to make progress on scalable oversight: the problem of supervising systems that potentially outperform us on most skills relevant to the task at hand. Empirical work on this problem is not straightforward, since we do not yet have systems that broadly exceed our abilities. This paper discusses one of the major ways we think about this problem, with a focus on ways it can be studied empirically. We first present an experimental design centered on tasks for which human specialists succeed but unaided humans and current general AI systems fail. We then present a proof-of-concept experiment meant to demonstrate a key feature of this experimental design and show its viability with two question-answering tasks: MMLU and time-limited QuALITY. On these tasks, we find that human participants who interact with an unreliable large-language-model dialog assistant through chat -- a trivial baseline strategy for scalable oversight -- substantially outperform both the model alone and their own unaided performance. These results are an encouraging sign that scalable oversight will be tractable to study with present models and bolster recent findings that large language models can productively assist humans with difficult tasks.
The use of language-model-based question-answering systems to aid humans in completing difficult tasks is limited, in part, by the unreliability of the text these systems generate. Using hard multiple-choice reading comprehension questions as a testbed, we assess whether presenting humans with arguments for two competing answer options, where one is correct and the other is incorrect, allows human judges to perform more accurately, even when one of the arguments is unreliable and deceptive. If this is helpful, we may be able to increase our justified trust in language-model-based systems by asking them to produce these arguments where needed. Previous research has shown that just a single turn of arguments in this format is not helpful to humans. However, as debate settings are characterized by a back-and-forth dialogue, we follow up on previous results to test whether adding a second round of counter-arguments is helpful to humans. We find that, regardless of whether they have access to arguments or not, humans perform similarly on our task. These findings suggest that, in the case of answering reading comprehension questions, debate is not a helpful format.
We present the results of the NLP Community Metasurvey. Run from May to June 2022, the survey elicited opinions on controversial issues, including industry influence in the field, concerns about AGI, and ethics. Our results put concrete numbers to several controversies: For example, respondents are split almost exactly in half on questions about the importance of artificial general intelligence, whether language models understand language, and the necessity of linguistic structure and inductive bias for solving NLP problems. In addition, the survey posed meta-questions, asking respondents to predict the distribution of survey responses. This allows us not only to gain insight on the spectrum of beliefs held by NLP researchers, but also to uncover false sociological beliefs where the community's predictions don't match reality. We find such mismatches on a wide range of issues. Among other results, the community greatly overestimates its own belief in the usefulness of benchmarks and the potential for scaling to solve real-world problems, while underestimating its own belief in the importance of linguistic structure, inductive bias, and interdisciplinary science.