Welcome to the sixth edition of the AI Index Report! This year, the report introduces more original data than any previous edition, including a new chapter on AI public opinion, a more thorough technical performance chapter, original analysis about large language and multimodal models, detailed trends in global AI legislation records, a study of the environmental impact of AI systems, and more. The AI Index Report tracks, collates, distills, and visualizes data related to artificial intelligence. Our mission is to provide unbiased, rigorously vetted, broadly sourced data in order for policymakers, researchers, executives, journalists, and the general public to develop a more thorough and nuanced understanding of the complex field of AI. The report aims to be the world's most credible and authoritative source for data and insights about AI.

Up to 150000 asteroids will be visible in the images of the ESA Euclid space telescope, and the instruments of Euclid offer multiband visual to near-infrared photometry and slitless spectra of these objects. Most asteroids will appear as streaks in the images. Due to the large number of images and asteroids, automated detection methods are needed. A non-machine-learning approach based on the StreakDet software was previously tested, but the results were not optimal for short and/or faint streaks. We set out to improve the capability to detect asteroid streaks in Euclid images by using deep learning. We built, trained, and tested a three-step machine-learning pipeline with simulated Euclid images. First, a convolutional neural network (CNN) detected streaks and their coordinates in full images, aiming to maximize the completeness (recall) of detections. Then, a recurrent neural network (RNN) merged snippets of long streaks detected in several parts by the CNN. Lastly, gradient-boosted trees (XGBoost) linked detected streaks between different Euclid exposures to reduce the number of false positives and improve the purity (precision) of the sample. The deep-learning pipeline surpasses the completeness and reaches a similar level of purity of a non-machine-learning pipeline based on the StreakDet software. Additionally, the deep-learning pipeline can detect asteroids 0.25-0.5 magnitudes fainter than StreakDet. The deep-learning pipeline could result in a 50% increase in the number of detected asteroids compared to the StreakDet software. There is still scope for further refinement, particularly in improving the accuracy of streak coordinates and enhancing the completeness of the final stage of the pipeline, which involves linking detections across multiple exposures.

In recent years, many interpretability methods have been proposed to help interpret the internal states of Transformer-models, at different levels of precision and complexity. Here, to analyze encoder-decoder Transformers, we propose a simple, new method: DecoderLens. Inspired by the LogitLens (for decoder-only Transformers), this method involves allowing the decoder to cross-attend representations of intermediate encoder layers instead of using the final encoder output, as is normally done in encoder-decoder models. The method thus maps previously uninterpretable vector representations to human-interpretable sequences of words or symbols. We report results from the DecoderLens applied to models trained on question answering, logical reasoning, speech recognition and machine translation. The DecoderLens reveals several specific subtasks that are solved at low or intermediate layers, shedding new light on the information flow inside the encoder component of this important class of models.

Neural Radiance Fields (NeRFs) have recently emerged as a powerful tool for 3D scene representation and rendering. These data-driven models can learn to synthesize high-quality images from sparse 2D observations, enabling realistic and interactive scene reconstructions. However, the growing usage of NeRFs in critical applications such as augmented reality, robotics, and virtual environments could be threatened by adversarial attacks. In this paper we present how generalizable NeRFs can be attacked by both low-intensity adversarial attacks and adversarial patches, where the later could be robust enough to be used in real world applications. We also demonstrate targeted attacks, where a specific, predefined output scene is generated by these attack with success.

Summarizing book-length documents (>100K tokens) that exceed the context window size of large language models (LLMs) requires first breaking the input document into smaller chunks and then prompting an LLM to merge, update, and compress chunk-level summaries. Despite the complexity and importance of this task, it has yet to be meaningfully studied due to the challenges of evaluation: existing book-length summarization datasets (e.g., BookSum) are in the pretraining data of most public LLMs, and existing evaluation methods struggle to capture errors made by modern LLM summarizers. In this paper, we present the first study of the coherence of LLM-based book-length summarizers implemented via two prompting workflows: (1) hierarchically merging chunk-level summaries, and (2) incrementally updating a running summary. We obtain 1193 fine-grained human annotations on GPT-4 generated summaries of 100 recently-published books and identify eight common types of coherence errors made by LLMs. Because human evaluation is expensive and time-consuming, we develop an automatic metric, BooookScore, that measures the proportion of sentences in a summary that do not contain any of the identified error types. BooookScore has high agreement with human annotations and allows us to systematically evaluate the impact of many other critical parameters (e.g., chunk size, base LLM) while saving $15K and 500 hours in human evaluation costs. We find that closed-source LLMs such as GPT-4 and Claude 2 produce summaries with higher BooookScore than the oft-repetitive ones generated by LLaMA 2. Incremental updating yields lower BooookScore but higher level of detail than hierarchical merging, a trade-off sometimes preferred by human annotators. We release code and annotations after blind review to spur more principled research on book-length summarization.

Large Language Models (LLMs) are quickly becoming ubiquitous, but the implications for social science research are not yet well understood. This paper asks whether LLMs can help us analyse large-N qualitative data from open-ended interviews, with an application to transcripts of interviews with Rohingya refugees in Cox's Bazaar, Bangladesh. We find that a great deal of caution is needed in using LLMs to annotate text as there is a risk of introducing biases that can lead to misleading inferences. We here mean bias in the technical sense, that the errors that LLMs make in annotating interview transcripts are not random with respect to the characteristics of the interview subjects. Training simpler supervised models on high-quality human annotations with flexible coding leads to less measurement error and bias than LLM annotations. Therefore, given that some high quality annotations are necessary in order to asses whether an LLM introduces bias, we argue that it is probably preferable to train a bespoke model on these annotations than it is to use an LLM for annotation.

Fairness problems in recommender systems often have a complexity in practice that is not adequately captured in simplified research formulations. A social choice formulation of the fairness problem, operating within a multi-agent architecture of fairness concerns, offers a flexible and multi-aspect alternative to fairness-aware recommendation approaches. Leveraging social choice allows for increased generality and the possibility of tapping into well-studied social choice algorithms for resolving the tension between multiple, competing fairness concerns. This paper explores a range of options for choice mechanisms in multi-aspect fairness applications using both real and synthetic data and shows that different classes of choice and allocation mechanisms yield different but consistent fairness / accuracy tradeoffs. We also show that a multi-agent formulation offers flexibility in adapting to user population dynamics.

The way users acquire information is undergoing a paradigm shift with the advent of ChatGPT. Unlike conventional search engines, ChatGPT retrieves knowledge from the model itself and generates answers for users. ChatGPT's impressive question-answering (QA) capability has attracted more than 100 million users within a short period of time but has also raised concerns regarding its reliability. In this paper, we perform the first large-scale measurement of ChatGPT's reliability in the generic QA scenario with a carefully curated set of 5,695 questions across ten datasets and eight domains. We find that ChatGPT's reliability varies across different domains, especially underperforming in law and science questions. We also demonstrate that system roles, originally designed by OpenAI to allow users to steer ChatGPT's behavior, can impact ChatGPT's reliability in an imperceptible way. We further show that ChatGPT is vulnerable to adversarial examples, and even a single character change can negatively affect its reliability in certain cases. We believe that our study provides valuable insights into ChatGPT's reliability and underscores the need for strengthening the reliability and security of large language models (LLMs).

The growth of hateful online content, or hate speech, has been associated with a global increase in violent crimes against minorities [23]. Harmful online content can be produced easily, automatically and anonymously. Even though, some form of auto-detection is already achieved through text classifiers in NLP, they can be fooled by adversarial attacks. To strengthen existing systems and stay ahead of attackers, we need better adversarial attacks. In this paper, we show that up to 70% of adversarial examples generated by adversarial attacks should be discarded because they do not preserve semantics. We address this core weakness and propose a new, fully supervised sentence embedding technique called Semantics-Preserving-Encoder (SPE). Our method outperforms existing sentence encoders used in adversarial attacks by achieving 1.2x - 5.1x better real attack success rate. We release our code as a plugin that can be used in any existing adversarial attack to improve its quality and speed up its execution.

Recent years have witnessed much interest in temporal reasoning over knowledge graphs (KG) for complex question answering (QA), but there remains a substantial gap in human capabilities. We explore how to generalize relational graph convolutional networks (RGCN) for temporal KGQA. Specifically, we propose a novel, intuitive and interpretable scheme to modulate the messages passed through a KG edge during convolution, based on the relevance of its associated time period to the question. We also introduce a gating device to predict if the answer to a complex temporal question is likely to be a KG entity or time and use this prediction to guide our scoring mechanism. We evaluate the resulting system, which we call TwiRGCN, on TimeQuestions, a recently released, challenging dataset for multi-hop complex temporal QA. We show that TwiRGCN significantly outperforms state-of-the-art systems on this dataset across diverse question types. Notably, TwiRGCN improves accuracy by 9--10 percentage points for the most difficult ordinal and implicit question types.