Generative AI systems produce a range of ethical and social risks. Evaluation of these risks is a critical step on the path to ensuring the safety of these systems. However, evaluation requires the availability of validated and established measurement approaches and tools. In this paper, we provide an empirical review of the methods and tools that are available for evaluating known safety of generative AI systems to date. To this end, we review more than 200 safety-related evaluations that have been applied to generative AI systems. We categorise each evaluation along multiple axes to create a detailed snapshot of the safety evaluation landscape to date. We release this data for researchers and AI safety practitioners (https://bitly.ws/3hUzu). Analysing the current safety evaluation landscape reveals three systemic ”evaluation gaps”. First, a ”modality gap” emerges as few safety evaluations exist for non-text modalities. Second, a ”risk coverage gap” arises as evaluations for several ethical and social risks are simply lacking. Third, a ”context gap” arises as most safety evaluations are model-centric and fail to take into account the broader context in which AI systems operate. Devising next steps for safety practitioners based on these findings, we present tactical ”low-hanging fruit” steps towards closing the identified evaluation gaps and their limitations. We close by discussing the role and limitations of safety evaluation to ensure the safety of generative AI systems.

Credit assignment in reinforcement learning is the problem of measuring an action’s influence on future rewards. In particular, this requires separating skill from luck, i.e. disentangling the effect of an action on rewards from that of external factors and subsequent actions. To achieve this, we adapt the notion of counterfactuals from causality theory to a model-free RL setup. The key idea is to condition value functions on future events, by learning to extract relevant information from a trajectory. We formulate a family of policy gradient algorithms that use these future-conditional value functions as baselines or critics, and show that they are provably low variance. To avoid the potential bias from conditioning on future information, we constrain the hindsight information to not contain information about the agent’s actions. We demonstrate the efficacy and validity of our algorithm on a number of illustrative and challenging problems.

We propose a general approach to fairness based on transporting distributions corresponding to different sensitive attributes to a common distribution. We use optimal transport theory to derive target distributions and methods that allow us to achieve fairness with minimal changes to the unfair model. Our approach is applicable to both classification and regression problems, can enforce different notions of fairness, and enable us to achieve a Pareto-optimal trade-off between accuracy and fairness. We demonstrate that it outperforms previous approaches in several benchmark fairness datasets.

In this work, we take a fresh look at some old and new algorithms for off-policy, return-based reinforcement learning. Expressing these in a common form, we derive a novel algorithm, Retrace(lambda), with three desired properties: (1) it has low variance; (2) it safely uses samples collected from any behaviour policy, whatever its degree of "off-policyness"; and (3) it is efficient as it makes the best use of samples collected from near on-policy behaviour policies. We analyse the contractive nature of the related operator under both off-policy policy evaluation and control settings and derive online sample-based algorithms. We believe this is the first return-based off-policy control algorithm converging a.s. to Q* without the GLIE assumption (Greedy in the Limit with Infinite Exploration). As a corollary, we prove the convergence of Watkins' Q(lambda), which was an open problem since 1989. We illustrate the benefits of Retrace(lambda) on a standard suite of Atari 2600 games.

We propose a novel step toward the unsupervised segmentation of whole objects by combining ldquohintsrdquo of partial scene segmentation offered by multiple soft, binary mattes. These mattes are implied by a set of hypothesized object boundary fragments in the scene. Rather than trying to find or define a single ldquobestrdquo segmentation, we generate multiple segmentations of an image. This reflects contemporary methods for unsupervised object discovery from groups of images, and it allows us to define intuitive evaluation metrics for our sets of segmentations based on the accurate and parsimonious delineation of scene objects. Our proposed approach builds on recent advances in spectral clustering, image matting, and boundary detection. It is demonstrated qualitatively and quantitatively on a dataset of scenes and is suitable for current work in unsupervised object discovery without top-down knowledge.

Less than 3% of the 145 million blind people living in developing countries are literate. This low literacy rate is partly due to the lack of trained teachers and the challenges associated with learning to write Braille on a traditional slate and stylus. These challenges include writing from right to left, writing mirrored images of letters, and receiving significantly delayed feedback. Extensive conversations with the Mathru School for the Blind near Bangalore, India, revealed the need for a robust, low-power, low-cost Braille writing tutor. We present an iterative and participatory design process resulting in the creation and refinement of a prototype Braille writing tutor system. This system uses a novel input device to capture a student's activity on a slate using a stylus and uses a range of techniques to teach Braille writing skills to both beginner and advanced students. We report on lessons learned from the implementation of this project and from a six-week pilot study at the Mathru school, and outline future directions for improvement.