Search

This work identifies 18 foundational challenges in assuring the alignment and safety of large language models (LLMs). These challenges are organized into three different categories: scientific understanding of LLMs, development and deployment methods, and sociotechnical challenges. Based on the identified challenges, we pose $200+$ concrete research questions.

A common way of learning to perform a task is to observe how it is carried out by experts. However, it is well known that for most tasks there is no unique way to perform them. This is especially noticeable the more complex the task is because factors such as the skill or the know-how of the expert may well affect the way she solves the task. In addition, learning from experts also suffers of having a small set of training examples generally coming from several experts (since experts are usually a limited and expensive resource), being all of them positive examples (i.e. examples that represent successful executions of the task). Traditional machine learning techniques are not useful in such scenarios, as they require extensive training data. Starting from very few executions of the task presented as activity sequences, we introduce a novel inductive approach for learning multiple models, with each one representing an alternative strategy of performing a task. By an iterative process based on generalisation and specialisation, we learn the underlying patterns that capture the different styles of performing a task exhibited by the examples. We illustrate our approach on two common activity recognition tasks: a surgical skills training task and a cooking domain. We evaluate the inferred models with respect to two metrics that measure how well the models represent the examples and capture the different forms of executing a task showed by the examples. We compare our results with the traditional process mining approach and show that a small set of meaningful examples is enough to obtain patterns that capture the different strategies that are followed to solve the tasks., Comment: 32 pages, Github repository, published paper, authors' version

In traditional machine teaching, a teacher wants to teach a concept to a learner, by means of a finite set of examples, the witness set. But concepts can have many equivalent representations. This redundancy strongly affects the search space, to the extent that teacher and learner may not be able to easily determine the equivalence class of each representation. In this common situation, instead of teaching concepts, we explore the idea of teaching representations. We work with several teaching schemas that exploit representation and witness size (Eager, Greedy and Optimal) and analyze the gains in teaching effectiveness for some representational languages (DNF expressions and Turing-complete P3 programs). Our theoretical and experimental results indicate that there are various types of redundancy, handled better by the Greedy schema introduced here than by the Eager schema, although both can be arbitrarily far away from the Optimal. For P3 programs we found that witness sets are usually smaller than the programs they identify, which is an illuminating justification of why machine teaching from examples makes sense at all., Comment: 16 pages, 3 figures, 3 tables

The Animal-AI Environment is a unique game-based research platform designed to serve both the artificial intelligence and cognitive science research communities. In this paper, we present Animal-AI 3, the latest version of the environment, outlining several major new features that make the game more engaging for humans and more complex for AI systems. New features include interactive buttons, reward dispensers, and player notifications, as well as an overhaul of the environment's graphics and processing for significant increases in agent training time and quality of the human player experience. We provide detailed guidance on how to build computational and behavioural experiments with Animal-AI 3. We present results from a series of agents, including the state-of-the-art Deep Reinforcement Learning agent (dreamer-v3), on newly designed tests and the Animal-AI Testbed of 900 tasks inspired by research in comparative psychology. Animal-AI 3 is designed to facilitate collaboration between the cognitive sciences and artificial intelligence. This paper serves as a stand-alone document that motivates, describes, and demonstrates Animal-AI 3 for the end user.

Comprehensive and accurate evaluation of general-purpose AI systems such as large language models allows for effective mitigation of their risks and deepened understanding of their capabilities. Current evaluation methodology, mostly based on benchmarks of specific tasks, falls short of adequately assessing these versatile AI systems, as present techniques lack a scientific foundation for predicting their performance on unforeseen tasks and explaining their varying performance on specific task items or user inputs. Moreover, existing benchmarks of specific tasks raise growing concerns about their reliability and validity. To tackle these challenges, we suggest transitioning from task-oriented evaluation to construct-oriented evaluation. Psychometrics, the science of psychological measurement, provides a rigorous methodology for identifying and measuring the latent constructs that underlie performance across multiple tasks. We discuss its merits, warn against potential pitfalls, and propose a framework to put it into practice. Finally, we explore future opportunities of integrating psychometrics with the evaluation of general-purpose AI systems.

Given rapid progress toward advanced AI and risks from frontier AI systems (advanced AI systems pushing the boundaries of the AI capabilities frontier), the creation and implementation of AI governance and regulatory schemes deserves prioritization and substantial investment. However, the status quo is untenable and, frankly, dangerous. A regulatory gap has permitted AI labs to conduct research, development, and deployment activities with minimal oversight. In response, frontier AI system evaluations have been proposed as a way of assessing risks from the development and deployment of frontier AI systems. Yet, the budding AI risk evaluation ecosystem faces significant coordination challenges, such as a limited diversity of evaluators, suboptimal allocation of effort, and perverse incentives. This paper proposes a solution in the form of an international consortium for AI risk evaluations, comprising both AI developers and third-party AI risk evaluators. Such a consortium could play a critical role in international efforts to mitigate societal-scale risks from advanced AI, including in managing responsible scaling policies and coordinated evaluation-based risk response. In this paper, we discuss the current evaluation ecosystem and its shortcomings, propose an international consortium for advanced AI risk evaluations, discuss issues regarding its implementation, discuss lessons that can be learnt from previous international institutions and existing proposals for international AI governance institutions, and, finally, we recommend concrete steps to advance the establishment of the proposed consortium: (i) solicit feedback from stakeholders, (ii) conduct additional research, (iii) conduct a workshop(s) for stakeholders, (iv) analyze feedback and create final proposal, (v) solicit funding, and (vi) create a consortium., Comment: 50 pages, 2 figures; updated w/ a few minor revisions based on feedback from SoLaR Workshop reviewers (on 5 page version)

We introduce the fundamental ideas and challenges of Predictable AI, a nascent research area that explores the ways in which we can anticipate key indicators of present and future AI ecosystems. We argue that achieving predictability is crucial for fostering trust, liability, control, alignment and safety of AI ecosystems, and thus should be prioritised over performance. While distinctive from other areas of technical and non-technical AI research, the questions, hypotheses and challenges relevant to Predictable AI were yet to be clearly described. This paper aims to elucidate them, calls for identifying paths towards AI predictability and outlines the potential impact of this emergent field., Comment: 11 pages excluding references, 4 figures, and 2 tables. Paper Under Review

As machine learning models become more general, we need to characterise them in richer, more meaningful ways. We describe a method to infer the cognitive profile of a system from diverse experimental data. To do so, we introduce measurement layouts that model how task-instance features interact with system capabilities to affect performance. These features must be triangulated in complex ways to be able to infer capabilities from non-populational data -- a challenge for traditional psychometric and inferential tools. Using the Bayesian probabilistic programming library PyMC, we infer different cognitive profiles for agents in two scenarios: 68 actual contestants in the AnimalAI Olympics and 30 synthetic agents for O-PIAAGETS, an object permanence battery. We showcase the potential for capability-oriented evaluation., Comment: 8 Pages + 14 pages of Appendices. 15 Figures. Submitted to AAAI 2024. Preprint

Language models have become very popular recently and many claims have been made about their abilities, including for commonsense reasoning. Given the increasingly better results of current language models on previous static benchmarks for commonsense reasoning, we explore an alternative dialectical evaluation. The goal of this kind of evaluation is not to obtain an aggregate performance value but to find failures and map the boundaries of the system. Dialoguing with the system gives the opportunity to check for consistency and get more reassurance of these boundaries beyond anecdotal evidence. In this paper we conduct some qualitative investigations of this kind of evaluation for the particular case of spatial reasoning (which is a fundamental aspect of commonsense reasoning). We conclude with some suggestions for future work both to improve the capabilities of language models and to systematise this kind of dialectical evaluation., Comment: 11 pages in main paper + 71 pages in appendix

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is vital that we understand the present and near-future capabilities and limitations of language models. To address this challenge, we introduce the Beyond the Imitation Game benchmark (BIG-bench). BIG-bench currently consists of 204 tasks, contributed by 450 authors across 132 institutions. Task topics are diverse, drawing problems from linguistics, childhood development, math, common-sense reasoning, biology, physics, social bias, software development, and beyond. BIG-bench focuses on tasks that are believed to be beyond the capabilities of current language models. We evaluate the behavior of OpenAI's GPT models, Google-internal dense transformer architectures, and Switch-style sparse transformers on BIG-bench, across model sizes spanning millions to hundreds of billions of parameters. In addition, a team of human expert raters performed all tasks in order to provide a strong baseline. Findings include: model performance and calibration both improve with scale, but are poor in absolute terms (and when compared with rater performance); performance is remarkably similar across model classes, though with benefits from sparsity; tasks that improve gradually and predictably commonly involve a large knowledge or memorization component, whereas tasks that exhibit "breakthrough" behavior at a critical scale often involve multiple steps or components, or brittle metrics; social bias typically increases with scale in settings with ambiguous context, but this can be improved with prompting., Comment: 27 pages, 17 figures + references and appendices, repo: https://github.com/google/BIG-bench

The progress of some AI paradigms such as deep learning is said to be linked to an exponential growth in the number of parameters. There are many studies corroborating these trends, but does this translate into an exponential increase in energy consumption? In order to answer this question we focus on inference costs rather than training costs, as the former account for most of the computing effort, solely because of the multiplicative factors. Also, apart from algorithmic innovations, we account for more specific and powerful hardware (leading to higher FLOPS) that is usually accompanied with important energy efficiency optimisations. We also move the focus from the first implementation of a breakthrough paper towards the consolidated version of the techniques one or two year later. Under this distinctive and comprehensive perspective, we study relevant models in the areas of computer vision and natural language processing: for a sustained increase in performance we see a much softer growth in energy consumption than previously anticipated. The only caveat is, yet again, the multiplicative factor, as future AI increases penetration and becomes more pervasive., Comment: For a revised version and its published version refer to: Desislavov, Radosvet, Fernando Mart\'inez-Plumed, and Jos\'e Hern\'andez-Orallo. Trends in AI inference energy consumption: Beyond the performance-vs-parameter laws of deep learning. Sustainable Computing: Informatics and Systems, Volume 38, April 2023. (https://doi.org/10.1016/j.suscom.2023.100857)

Recent research in machine teaching has explored the instruction of any concept expressed in a universal language. In this compositional context, new experimental results have shown that there exist data teaching sets surprisingly shorter than the concept description itself. However, there exists a bound for those remarkable experimental findings through teaching size and concept complexity that we further explore here. As concepts are rarely taught in isolation we investigate the best configuration of concepts to teach a given set of concepts, where those that have been acquired first can be reused for the description of new ones. This new notion of conditional teaching size uncovers new insights, such as the interposition phenomenon: certain prior knowledge generates simpler compatible concepts that increase the teaching size of the concept that we want to teach. This does not happen for conditional Kolmogorov complexity. Furthermore, we provide an algorithm that constructs optimal curricula based on interposition avoidance. This paper presents a series of theoretical results, including their proofs, and some directions for future work. New research possibilities in curriculum teaching in compositional scenarios are now wide open to exploration., Comment: 26 pages

Given the complexity of typical data science projects and the associated demand for human expertise, automation has the potential to transform the data science process. Key insights: * Automation in data science aims to facilitate and transform the work of data scientists, not to replace them. * Important parts of data science are already being automated, especially in the modeling stages, where techniques such as automated machine learning (AutoML) are gaining traction. * Other aspects are harder to automate, not only because of technological challenges, but because open-ended and context-dependent tasks require human interaction., Comment: 19 pages, 3 figures. v1 accepted for publication (April 2021) in Communications of the ACM

The Apperception Engine is an unsupervised learning system. Given a sequence of sensory inputs, it constructs a symbolic causal theory that both explains the sensory sequence and also satisfies a set of unity conditions. The unity conditions insist that the constituents of the theory - objects, properties, and laws - must be integrated into a coherent whole. Once a theory has been constructed, it can be applied to predict future sensor readings, retrodict earlier readings, or impute missing readings. In this paper, we evaluate the Apperception Engine in a diverse variety of domains, including cellular automata, rhythms and simple nursery tunes, multi-modal binding problems, occlusion tasks, and sequence induction intelligence tests. In each domain, we test our engine's ability to predict future sensor values, retrodict earlier sensor values, and impute missing sensory data. The engine performs well in all these domains, significantly outperforming neural net baselines and state of the art inductive logic programming systems. These results are significant because neural nets typically struggle to solve the binding problem (where information from different modalities must somehow be combined together into different aspects of one unified object) and fail to solve occlusion tasks (in which objects are sometimes visible and sometimes obscured from view). We note in particular that in the sequence induction intelligence tests, our system achieved human-level performance. This is notable because our system is not a bespoke system designed specifically to solve intelligence tests, but a general-purpose system that was designed to make sense of any sensory sequence., Comment: arXiv admin note: substantial text overlap with arXiv:1910.02227

This paper attempts to answer a central question in unsupervised learning: what does it mean to "make sense" of a sensory sequence? In our formalization, making sense involves constructing a symbolic causal theory that both explains the sensory sequence and also satisfies a set of unity conditions. The unity conditions insist that the constituents of the causal theory -- objects, properties, and laws -- must be integrated into a coherent whole. On our account, making sense of sensory input is a type of program synthesis, but it is unsupervised program synthesis. Our second contribution is a computer implementation, the Apperception Engine, that was designed to satisfy the above requirements. Our system is able to produce interpretable human-readable causal theories from very small amounts of data, because of the strong inductive bias provided by the unity conditions. A causal theory produced by our system is able to predict future sensor readings, as well as retrodict earlier readings, and impute (fill in the blanks of) missing sensory readings, in any combination. We tested the engine in a diverse variety of domains, including cellular automata, rhythms and simple nursery tunes, multi-modal binding problems, occlusion tasks, and sequence induction intelligence tests. In each domain, we test our engine's ability to predict future sensor values, retrodict earlier sensor values, and impute missing sensory data. The engine performs well in all these domains, significantly out-performing neural net baselines. We note in particular that in the sequence induction intelligence tests, our system achieved human-level performance. This is notable because our system is not a bespoke system designed specifically to solve intelligence tests, but a general-purpose system that was designed to make sense of any sensory sequence.

Recent advances in artificial intelligence have been strongly driven by the use of game environments for training and evaluating agents. Games are often accessible and versatile, with well-defined state-transitions and goals allowing for intensive training and experimentation. However, agents trained in a particular environment are usually tested on the same or slightly varied distributions, and solutions do not necessarily imply any understanding. If we want AI systems that can model and understand their environment, we need environments that explicitly test for this. Inspired by the extensive literature on animal cognition, we present an environment that keeps all the positive elements of standard gaming environments, but is explicitly designed for the testing of animal-like artificial cognition., Comment: 14 pages, 34 figures (update: reduce images size)

The causes underlying unfair decision making are complex, being internalised in different ways by decision makers, other actors dealing with data and models, and ultimately by the individuals being affected by these decisions. One frequent manifestation of all these latent causes arises in the form of missing values: protected groups are more reluctant to give information that could be used against them, delicate information for some groups can be erased by human operators, or data acquisition may simply be less complete and systematic for minority groups. As a result, missing values and bias in data are two phenomena that are tightly coupled. However, most recent techniques, libraries and experimental results dealing with fairness in machine learning have simply ignored missing data. In this paper, we claim that fairness research should not miss the opportunity to deal properly with missing data. To support this claim, (1) we analyse the sources of missing data and bias, and we map the common causes, (2) we find that rows containing missing values are usually fairer than the rest, which should not be treated as the uncomfortable ugly data that different techniques and libraries get rid of at the first occasion, and (3) we study the trade-off between performance and fairness when the rows with missing values are used (either because the technique deals with them directly or by imputation methods). We end the paper with a series of recommended procedures about what to do with missing data when aiming for fair decision making., Comment: Preprint submitted to Decision Support Systems Journal

Item response theory (IRT) can be applied to the analysis of the evaluation of results from AI benchmarks. The two-parameter IRT model provides two indicators (difficulty and discrimination) on the side of the item (or AI problem) while only one indicator (ability) on the side of the respondent (or AI agent). In this paper we analyse how to make this set of indicators dual, by adding a fourth indicator, generality, on the side of the respondent. Generality is meant to be dual to discrimination, and it is based on difficulty. Namely, generality is defined as a new metric that evaluates whether an agent is consistently good at easy problems and bad at difficult ones. With the addition of generality, we see that this set of four key indicators can give us more insight on the results of AI benchmarks. In particular, we explore two popular benchmarks in AI, the Arcade Learning Environment (Atari 2600 games) and the General Video Game AI competition. We provide some guidelines to estimate and interpret these indicators for other AI benchmarks and competitions., Comment: This report is a preliminary version of a related paper with title "Dual Indicators to Analyse AI Benchmarks: Difficulty, Discrimination, Ability and Generality", accepted for publication at IEEE Transactions on Games. Please refer to and cite the journal paper (https://doi.org/10.1109/TG.2018.2883773)

Given one or two examples, humans are good at understanding how to solve a problem independently of its domain, because they are able to detect what the problem is and to choose the appropriate background knowledge according to the context. For instance, presented with the string "8/17/2017" to be transformed to "17th of August of 2017", humans will process this in two steps: (1) they recognise that it is a date and (2) they map the date to the 17th of August of 2017. Inductive Programming (IP) aims at learning declarative (functional or logic) programs from examples. Two key advantages of IP are the use of background knowledge and the ability to synthesise programs from a few input/output examples (as humans do). In this paper we propose to use IP as a means for automating repetitive data manipulation tasks, frequently presented during the process of {\em data wrangling} in many data manipulation problems. Here we show that with the use of general-purpose declarative (programming) languages jointly with generic IP systems and the definition of domain-specific knowledge, many specific data wrangling problems from different application domains can be automatically solved from very few examples. We also propose an integrated benchmark for data wrangling, which we share publicly for the community., Comment: 24 pages

This paper presents a multidisciplinary task approach for assessing the impact of artificial intelligence on the future of work. We provide definitions of a task from two main perspectives: socio-economic and computational. We propose to explore ways in which we can integrate or map these perspectives, and link them with the skills or capabilities required by them, for humans and AI systems. Finally, we argue that in order to understand the dynamics of tasks, we have to explore the relevance of autonomy and generality of AI systems for the automation or alteration of the workplace., Comment: AEGAP2018 Workshop at ICML 2018, 7 pages, 1 table

This document contains the outcome of the first Human behaviour and machine intelligence (HUMAINT) workshop that took place 5-6 March 2018 in Barcelona, Spain. The workshop was organized in the context of a new research programme at the Centre for Advanced Studies, Joint Research Centre of the European Commission, which focuses on studying the potential impact of artificial intelligence on human behaviour. The workshop gathered an interdisciplinary group of experts to establish the state of the art research in the field and a list of future research challenges to be addressed on the topic of human and machine intelligence, algorithm's potential impact on human cognitive capabilities and decision making, and evaluation and regulation needs. The document is made of short position statements and identification of challenges provided by each expert, and incorporates the result of the discussions carried out during the workshop. In the conclusion section, we provide a list of emerging research topics and strategies to be addressed in the near future., Comment: Proceedings of 1st HUMAINT (Human Behaviour and Machine Intelligence) workshop, Barcelona, Spain, March 5-6, 2018, edited by European Commission, Seville, 2018, JRC111773 https://ec.europa.eu/jrc/communities/community/humaint/document/assessing-impact-machine-intelligence-human-behaviour-interdisciplinary. arXiv admin note: text overlap with arXiv:1409.3097 by other authors

We reframe the analysis of progress in AI by incorporating into an overall framework both the task performance of a system, and the time and resource costs incurred in the development and deployment of the system. These costs include: data, expert knowledge, human oversight, software resources, computing cycles, hardware and network facilities, and (what kind of) time. These costs are distributed over the life cycle of the system, and may place differing demands on different developers and users. The multidimensional performance and cost space we present can be collapsed to a single utility metric that measures the value of the system for different stakeholders. Even without a single utility function, AI advances can be generically assessed by whether they expand the Pareto surface. We label these types of costs as neglected dimensions of AI progress, and explore them using four case studies: Alpha* (Go, Chess, and other board games), ALE (Atari games), ImageNet (Image classification) and Virtual Personal Assistants (Siri, Alexa, Cortana, and Google Assistant). This broader model of progress in AI will lead to novel ways of estimating the potential societal use and impact of an AI system, and the establishment of milestones for future progress.

We investigate the teaching of infinite concept classes through the effect of the learning bias (which is used by the learner to prefer some concepts over others and by the teacher to devise the teaching examples) and the sampling bias (which determines how the concepts are sampled from the class). We analyse two important classes: Turing machines and finite-state machines. We derive bounds for the biased teaching dimension when the learning bias is derived from a complexity measure (Kolmogorov complexity and minimal number of states respectively) and analyse the sampling distributions that lead to finite expected biased teaching dimensions. We highlight the existing trade-off between the bound and the representativeness of the sample, and its implications for the understanding of what teaching rich concepts to machines entails.

We propose an extension of the Cross Industry Standard Process for Data Mining (CRISPDM) which addresses specific challenges of machine learning and data mining for context and model reuse handling. This new general context-aware process model is mapped with CRISP-DM reference model proposing some new or enhanced outputs.

This note revisits the concepts of task and difficulty. The notion of cognitive task and its use for the evaluation of intelligent systems is still replete with issues. The view of tasks as MDP in the context of reinforcement learning has been especially useful for the formalisation of learning tasks. However, this alternate interaction does not accommodate well for some other tasks that are usual in artificial intelligence and, most especially, in animal and human evaluation. In particular, we want to have a more general account of episodes, rewards and responses, and, most especially, the computational complexity of the algorithm behind an agent solving a task. This is crucial for the determination of the difficulty of a task as the (logarithm of the) number of computational steps required to acquire an acceptable policy for the task, which includes the exploration of policies and their verification. We introduce a notion of asynchronous-time stochastic tasks. Based on this interpretation, we can see what task difficulty is, what instance difficulty is (relative to a task) and also what task compositions and decompositions are., Comment: 30 pages

The application of cognitive mechanisms to support knowledge acquisition is, from our point of view, crucial for making the resulting models coherent, efficient, credible, easy to use and understandable. In particular, there are two characteristic features of intelligence that are essential for knowledge development: forgetting and consolidation. Both plays an important role in knowledge bases and learning systems to avoid possible information overflow and redundancy, and in order to preserve and strengthen important or frequently used rules and remove (or forget) useless ones. We present an incremental, long-life view of knowledge acquisition which tries to improve task after task by determining what to keep, what to consolidate and what to forget, overcoming The Stability-Plasticity dilemma. In order to do that, we rate rules by introducing several metrics through the first adaptation, to our knowledge, of the Minimum Message Length (MML) principle to a coverage graph, a hierarchical assessment structure which treats evidence and rules in a unified way. The metrics are not only used to forget some of the worst rules, but also to set a consolidation process to promote those selected rules to the knowledge base, which is also mirrored by a demotion system. We evaluate the framework with a series of tasks in a chess rule learning domain.

In this exploratory note we ask the question of what a measure of performance for all tasks is like if we use a weighting of tasks based on a difficulty function. This difficulty function depends on the complexity of the (acceptable) solution for the task (instead of a universal distribution over tasks or an adaptive test). The resulting aggregations and decompositions are (now retrospectively) seen as the natural (and trivial) interactive generalisation of the C-tests., Comment: 16 pages

Artificial intelligence develops techniques and systems whose performance must be evaluated on a regular basis in order to certify and foster progress in the discipline. We will describe and critically assess the different ways AI systems are evaluated. We first focus on the traditional task-oriented evaluation approach. We see that black-box (behavioural evaluation) is becoming more and more common, as AI systems are becoming more complex and unpredictable. We identify three kinds of evaluation: Human discrimination, problem benchmarks and peer confrontation. We describe the limitations of the many evaluation settings and competitions in these three categories and propose several ideas for a more systematic and robust evaluation. We then focus on a less customary (and challenging) ability-oriented evaluation approach, where a system is characterised by its (cognitive) abilities, rather than by the tasks it is designed to solve. We discuss several possibilities: the adaptation of cognitive tests used for humans and animals, the development of tests derived from algorithmic information theory or more general approaches under the perspective of universal psychometrics., Comment: 34 pages. This paper is largely superseded by the following paper: "Evaluation in artificial intelligence: from task-oriented to ability-oriented measurement" Journal of Artificial Intelligence Review (2016). doi:10.1007/s10462-016-9505-7, \url{http://dx.doi.org/10.1007/s10462-016-9505-7}. Please check and refer to the journal paper