The Alan Turing Institute

This article reports the findings of AI4People, an Atomium-EISMD initiative designed to lay the foundations for a "Good AI Society". We introduce the core opportunities and risks of AI for society; present a synthesis of five ethical principles that should undergird its development and adoption; and offer 20 concrete recommendations-to assess, to develop, to incentivise, and to support good AI-which in some cases may be undertaken directly by national or supranational policy makers, while in others may be led by other stakeholders. If adopted, these recommendations would serve as a firm foundation for the establishment of a Good AI Society.

In information societies, operations, decisions and choices previously left to humans are increasingly delegated to algorithms, which may advise, if not decide, about how data should be interpreted and what actions should be taken as a result. More and more often, algorithms mediate social processes, business transactions, governmental decisions, and how we perceive, understand, and interact among ourselves and with the environment. Gaps between the design and operation of algorithms and our understanding of their ethical implications can have severe consequences affecting individuals as well as groups and whole societies. This paper makes three contributions to clarify the ethical importance of algorithmic mediation. It provides a prescriptive map to organise the debate. It reviews the current discussion of ethical aspects of algorithms. And it assesses the available literature in order to identify areas requiring further work to develop the ethics of algorithms.

The TRIPOD (Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis) statement was published in 2015 to provide the minimum reporting recommendations for studies developing or evaluating the performance of a prediction model. Methodological advances in the field of prediction have since included the widespread use of artificial intelligence (AI) powered by machine learning methods to develop prediction models. An update to the TRIPOD statement is thus needed. TRIPOD+AI provides harmonised guidance for reporting prediction model studies, irrespective of whether regression modelling or machine learning methods have been used. The new checklist supersedes the TRIPOD 2015 checklist, which should no longer be used. This article describes the development of TRIPOD+AI and presents the expanded 27 item checklist with more detailed explanation of each reporting recommendation, and the TRIPOD+AI for Abstracts checklist. TRIPOD+AI aims to promote the complete, accurate, and transparent reporting of studies that develop a prediction model or evaluate its performance. Complete reporting will facilitate study appraisal, model evaluation, and model implementation.

Abstract In this commentary, we discuss the nature of reversible and irreversible questions, that is, questions that may enable one to identify the nature of the source of their answers. We then introduce GPT-3, a third-generation, autoregressive language model that uses deep learning to produce human-like texts, and use the previous distinction to analyse it. We expand the analysis to present three tests based on mathematical, semantic (that is, the Turing Test), and ethical questions and show that GPT-3 is not designed to pass any of them. This is a reminder that GPT-3 does not do what it is not supposed to do, and that any interpretation of GPT-3 as the beginning of the emergence of a general form of artificial intelligence is merely uninformed science fiction. We conclude by outlining some of the significant consequences of the industrialisation of automatic and cheap production of good, semantic artefacts.

Abstract Over the past few decades, neuroimaging has become a ubiquitous tool in basic research and clinical studies of the human brain. However, no reference standards currently exist to quantify individual differences in neuroimaging metrics over time, in contrast to growth charts for anthropometric traits such as height and weight 1 . Here we assemble an interactive open resource to benchmark brain morphology derived from any current or future sample of MRI data ( http://www.brainchart.io/ ). With the goal of basing these reference charts on the largest and most inclusive dataset available, acknowledging limitations due to known biases of MRI studies relative to the diversity of the global population, we aggregated 123,984 MRI scans, across more than 100 primary studies, from 101,457 human participants between 115 days post-conception to 100 years of age. MRI metrics were quantified by centile scores, relative to non-linear trajectories 2 of brain structural changes, and rates of change, over the lifespan. Brain charts identified previously unreported neurodevelopmental milestones 3 , showed high stability of individuals across longitudinal assessments, and demonstrated robustness to technical and methodological differences between primary studies. Centile scores showed increased heritability compared with non-centiled MRI phenotypes, and provided a standardized measure of atypical brain structure that revealed patterns of neuroanatomical variation across neurological and psychiatric disorders. In summary, brain charts are an essential step towards robust quantification of individual variation benchmarked to normative trajectories in multiple, commonly used neuroimaging phenotypes.

The complexity of many biological, social and technological systems stems from the richness of the interactions among their units. Over the past decades, a variety of complex systems has been successfully described as networks whose interacting pairs of nodes are connected by links. Yet, from human communications to chemical reactions and ecological systems, interactions can often occur in groups of three or more nodes and cannot be described simply in terms of dyads. Until recently little attention has been devoted to the higher-order architecture of real complex systems. However, a mounting body of evidence is showing that taking the higher-order structure of these systems into account can enhance our modeling capacities and help us understand and predict their dynamical behavior. Here we present a complete overview of the emerging field of networks beyond pairwise interactions. We discuss how to represent higher-order interactions and introduce the different frameworks used to describe higher-order systems, highlighting the links between the existing concepts and representations. We review the measures designed to characterize the structure of these systems and the models proposed to generate synthetic structures, such as random and growing bipartite graphs, hypergraphs and simplicial complexes. We introduce the rapidly growing research on higher-order dynamical systems and dynamical topology, discussing the relations between higher-order interactions and collective behavior. We focus in particular on new emergent phenomena characterizing dynamical processes, such as diffusion, synchronization, spreading, social dynamics and games, when extended beyond pairwise interactions. We conclude with a summary of empirical applications, and an outlook on current modeling and conceptual frontiers.

The recent boom in microfluidics and combinatorial indexing strategies, combined with low sequencing costs, has empowered single-cell sequencing technology. Thousands-or even millions-of cells analyzed in a single experiment amount to a data revolution in single-cell biology and pose unique data science problems. Here, we outline eleven challenges that will be central to bringing this emerging field of single-cell data science forward. For each challenge, we highlight motivating research questions, review prior work, and formulate open problems. This compendium is for established researchers, newcomers, and students alike, highlighting interesting and rewarding problems for the coming years.

Across scientific disciplines, there is a rapidly growing recognition of the need for more statistically robust, transparent approaches to data visualization. Complementary to this, many scientists have called for plotting tools that accurately and transparently convey key aspects of statistical effects and raw data with minimal distortion. Previously common approaches, such as plotting conditional mean or median barplots together with error-bars have been criticized for distorting effect size, hiding underlying patterns in the raw data, and obscuring the assumptions upon which the most commonly used statistical tests are based. Here we describe a data visualization approach which overcomes these issues, providing maximal statistical information while preserving the desired 'inference at a glance' nature of barplots and other similar visualization devices. These "raincloud plots" can visualize raw data, probability density, and key summary statistics such as median, mean, and relevant confidence intervals in an appealing and flexible format with minimal redundancy. In this tutorial paper, we provide basic demonstrations of the strength of raincloud plots and similar approaches, outline potential modifications for their optimal use, and provide open-source code for their streamlined implementation in R, Python and Matlab ( https://github.com/RainCloudPlots/RainCloudPlots). Readers can investigate the R and Python tutorials interactively in the browser using Binder by Project Jupyter.

The complexity of many biological, social and technological systems stems from the richness of the interactions among their units. Over the past decades, a variety of complex systems has been successfully described as networks whose interacting pairs of nodes are connected by links. Yet, from human communications to chemical reactions and ecological systems, interactions can often occur in groups of three or more nodes and cannot be described simply in terms of dyads. Until recently little attention has been devoted to the higher-order architecture of real complex systems. However, a mounting body of evidence is showing that taking the higher-order structure of these systems into account can enhance our modeling capacities and help us understand and predict their dynamical behavior. Here we present a complete overview of the emerging field of networks beyond pairwise interactions. We discuss how to represent higher-order interactions and introduce the different frameworks used to describe higher-order systems, highlighting the links between the existing concepts and representations. We review the measures designed to characterize the structure of these systems and the models proposed to generate synthetic structures, such as random and growing bipartite graphs, hypergraphs and simplicial complexes. We introduce the rapidly growing research on higher-order dynamical systems and dynamical topology, discussing the relations between higher-order interactions and collective behavior. We focus in particular on new emergent phenomena characterizing dynamical processes, such as diffusion, synchronization, spreading, social dynamics and games, when extended beyond pairwise interactions. We conclude with a summary of empirical applications, and an outlook on current modeling and conceptual frontiers.

In recent months, researchers,1 government bodies,2 and the media3 have claimed that a ‘right to explanation’ of decisions made by automated and artificially intelligent algorithmic systems is legally mandated by the forthcoming European Union General Data Protection Regulation4 2016/679 (GDPR). The right to explanation is viewed as a promising mechanism in the broader pursuit by government and industry for accountability and transparency in algorithms, artificial intelligence, robotics, and other automated systems.5 Automated systems can have many unintended and unexpected effects.6 Public assessment of the extent and source of these problems is often difficult,7 owing to the use of complex and opaque algorithmic mechanisms.8 The alleged right to explanation would require data controllers to explain how such mechanisms reach decisions. Significant hype has been mounting over the empowering effects of such a legally enforceable right for data subjects, and the disruption of data intensive industries, which would be forced to explain how complex and perhaps inscrutable automated methods work in practice.

is to deeply root robotics research in science while developing novel robotic platforms that will enable new scientific discoveries. Of our 10 grand challenges, the first 7 represent underpinning technologies that have a wider impact on all application areas of robotics. For the next two challenges, we have included social robotics and medical robotics as application-specific areas of development to highlight the substantial societal and health impacts that they will bring. Finally, the last challenge is related to responsible innovation and how ethics and security should be carefully considered as we develop the technology further.

The CONSORT 2010 statement provides minimum guidelines for reporting randomized trials. Its widespread use has been instrumental in ensuring transparency in the evaluation of new interventions. More recently, there has been a growing recognition that interventions involving artificial intelligence (AI) need to undergo rigorous, prospective evaluation to demonstrate impact on health outcomes. The CONSORT-AI (Consolidated Standards of Reporting Trials-Artificial Intelligence) extension is a new reporting guideline for clinical trials evaluating interventions with an AI component. It was developed in parallel with its companion statement for clinical trial protocols: SPIRIT-AI (Standard Protocol Items: Recommendations for Interventional Trials-Artificial Intelligence). Both guidelines were developed through a staged consensus process involving literature review and expert consultation to generate 29 candidate items, which were assessed by an international multi-stakeholder group in a two-stage Delphi survey (103 stakeholders), agreed upon in a two-day consensus meeting (31 stakeholders) and refined through a checklist pilot (34 participants). The CONSORT-AI extension includes 14 new items that were considered sufficiently important for AI interventions that they should be routinely reported in addition to the core CONSORT 2010 items. CONSORT-AI recommends that investigators provide clear descriptions of the AI intervention, including instructions and skills required for use, the setting in which the AI intervention is integrated, the handling of inputs and outputs of the AI intervention, the human-AI interaction and provision of an analysis of error cases. CONSORT-AI will help promote transparency and completeness in reporting clinical trials for AI interventions. It will assist editors and peer reviewers, as well as the general readership, to understand, interpret and critically appraise the quality of clinical trial design and risk of bias in the reported outcomes.

As ever more policy-makers, governments and organisations turn to the gig economy and digital labour as an economic development strategy to bring jobs to places that need them, it becomes important to understand better how this might influence the livelihoods of workers. Drawing on a multi-year study with digital workers in Sub-Saharan Africa and South-east Asia, this article highlights four key concerns for workers: bargaining power, economic inclusion, intermediated value chains, and upgrading. The article shows that although there are important and tangible benefits for a range of workers, there are also a range of risks and costs that unduly affect the livelihoods of digital workers. Building on those concerns, it then concludes with a reflection on four broad strategies - certification schemes, organising digital workers, regulatory strategies and democratic control of online labour platforms - that could be employed to improve conditions and livelihoods for digital workers.

BACKGROUND: Directed acyclic graphs (DAGs) are an increasingly popular approach for identifying confounding variables that require conditioning when estimating causal effects. This review examined the use of DAGs in applied health research to inform recommendations for improving their transparency and utility in future research. METHODS: Original health research articles published during 1999-2017 mentioning 'directed acyclic graphs' (or similar) or citing DAGitty were identified from Scopus, Web of Science, Medline and Embase. Data were extracted on the reporting of: estimands, DAGs and adjustment sets, alongside the characteristics of each article's largest DAG. RESULTS: A total of 234 articles were identified that reported using DAGs. A fifth (n = 48, 21%) reported their target estimand(s) and half (n = 115, 48%) reported the adjustment set(s) implied by their DAG(s). Two-thirds of the articles (n = 144, 62%) made at least one DAG available. DAGs varied in size but averaged 12 nodes [interquartile range (IQR): 9-16, range: 3-28] and 29 arcs (IQR: 19-42, range: 3-99). The median saturation (i.e. percentage of total possible arcs) was 46% (IQR: 31-67, range: 12-100). 37% (n = 53) of the DAGs included unobserved variables, 17% (n = 25) included 'super-nodes' (i.e. nodes containing more than one variable) and 34% (n = 49) were visually arranged so that the constituent arcs flowed in the same direction (e.g. top-to-bottom). CONCLUSION: There is substantial variation in the use and reporting of DAGs in applied health research. Although this partly reflects their flexibility, it also highlights some potential areas for improvement. This review hence offers several recommendations to improve the reporting and use of DAGs in future research.

Heart failure (HF) is a leading cause of morbidity and mortality worldwide. A small proportion of HF cases are attributable to monogenic cardiomyopathies and existing genome-wide association studies (GWAS) have yielded only limited insights, leaving the observed heritability of HF largely unexplained. We report results from a GWAS meta-analysis of HF comprising 47,309 cases and 930,014 controls. Twelve independent variants at 11 genomic loci are associated with HF, all of which demonstrate one or more associations with coronary artery disease (CAD), atrial fibrillation, or reduced left ventricular function, suggesting shared genetic aetiology. Functional analysis of non-CAD-associated loci implicate genes involved in cardiac development (MYOZ1, SYNPO2L), protein homoeostasis (BAG3), and cellular senescence (CDKN1A). Mendelian randomisation analysis supports causal roles for several HF risk factors, and demonstrates CAD-independent effects for atrial fibrillation, body mass index, and hypertension. These findings extend our knowledge of the pathways underlying HF and may inform new therapeutic strategies.

IMPORTANCE: Genetic factors increase risk of dementia, but the extent to which this can be offset by lifestyle factors is unknown. OBJECTIVE: To investigate whether a healthy lifestyle is associated with lower risk of dementia regardless of genetic risk. DESIGN, SETTING, AND PARTICIPANTS: A retrospective cohort study that included adults of European ancestry aged at least 60 years without cognitive impairment or dementia at baseline. Participants joined the UK Biobank study from 2006 to 2010 and were followed up until 2016 or 2017. EXPOSURES: A polygenic risk score for dementia with low (lowest quintile), intermediate (quintiles 2 to 4), and high (highest quintile) risk categories and a weighted healthy lifestyle score, including no current smoking, regular physical activity, healthy diet, and moderate alcohol consumption, categorized into favorable, intermediate, and unfavorable lifestyles. MAIN OUTCOMES AND MEASURES: Incident all-cause dementia, ascertained through hospital inpatient and death records. RESULTS: A total of 196 383 individuals (mean [SD] age, 64.1 [2.9] years; 52.7% were women) were followed up for 1 545 433 person-years (median [interquartile range] follow-up, 8.0 [7.4-8.6] years). Overall, 68.1% of participants followed a favorable lifestyle, 23.6% followed an intermediate lifestyle, and 8.2% followed an unfavorable lifestyle. Twenty percent had high polygenic risk scores, 60% had intermediate risk scores, and 20% had low risk scores. Of the participants with high genetic risk, 1.23% (95% CI, 1.13%-1.35%) developed dementia compared with 0.63% (95% CI, 0.56%-0.71%) of the participants with low genetic risk (adjusted hazard ratio, 1.91 [95% CI, 1.64-2.23]). Of the participants with a high genetic risk and unfavorable lifestyle, 1.78% (95% CI, 1.38%-2.28%) developed dementia compared with 0.56% (95% CI, 0.48%-0.66%) of participants with low genetic risk and favorable lifestyle (hazard ratio, 2.83 [95% CI, 2.09-3.83]). There was no significant interaction between genetic risk and lifestyle factors (P = .99). Among participants with high genetic risk, 1.13% (95% CI, 1.01%-1.26%) of those with a favorable lifestyle developed dementia compared with 1.78% (95% CI, 1.38%-2.28%) with an unfavorable lifestyle (hazard ratio, 0.68 [95% CI, 0.51-0.90]). CONCLUSIONS AND RELEVANCE: Among older adults without cognitive impairment or dementia, both an unfavorable lifestyle and high genetic risk were significantly associated with higher dementia risk. A favorable lifestyle was associated with a lower dementia risk among participants with high genetic risk.

Machine learning can impact people with legal or ethical consequences when it is used to automate decisions in areas such as insurance, lending, hiring, and predictive policing. In many of these scenarios, previous decisions have been made that are unfairly biased against certain subpopulations, for example those of a particular race, gender, or sexual orientation. Since this past data may be biased, machine learning predictors must account for this to avoid perpetuating or creating discriminatory practices. In this paper, we develop a framework for modeling fairness using tools from causal inference. Our definition of counterfactual fairness captures the intuition that a decision is fair towards an individual if it is the same in (a) the actual world and (b) a counterfactual world where the individual belonged to a different demographic group. We demonstrate our framework on a real-world problem of fair prediction of success in law school.

OBJECTIVE: To establish whether there is any change in mortality from infection with a new variant of SARS-CoV-2, designated a variant of concern (VOC-202012/1) in December 2020, compared with circulating SARS-CoV-2 variants. DESIGN: Matched cohort study. SETTING: Community based (pillar 2) covid-19 testing centres in the UK using the TaqPath assay (a proxy measure of VOC-202012/1 infection). PARTICIPANTS: 54 906 matched pairs of participants who tested positive for SARS-CoV-2 in pillar 2 between 1 October 2020 and 29 January 2021, followed-up until 12 February 2021. Participants were matched on age, sex, ethnicity, index of multiple deprivation, lower tier local authority region, and sample date of positive specimens, and differed only by detectability of the spike protein gene using the TaqPath assay. MAIN OUTCOME MEASURE: Death within 28 days of the first positive SARS-CoV-2 test result. RESULTS: The mortality hazard ratio associated with infection with VOC-202012/1 compared with infection with previously circulating variants was 1.64 (95% confidence interval 1.32 to 2.04) in patients who tested positive for covid-19 in the community. In this comparatively low risk group, this represents an increase in deaths from 2.5 to 4.1 per 1000 detected cases. CONCLUSIONS: The probability that the risk of mortality is increased by infection with VOC-202012/01 is high. If this finding is generalisable to other populations, infection with VOC-202012/1 has the potential to cause substantial additional mortality compared with previously circulating variants. Healthcare capacity planning and national and international control policies are all impacted by this finding, with increased mortality lending weight to the argument that further coordinated and stringent measures are justified to reduce deaths from SARS-CoV-2.

BACKGROUND: Coronary artery disease (CAD) has substantial heritability and a polygenic architecture. However, the potential of genomic risk scores to help predict CAD outcomes has not been evaluated comprehensively, because available studies have involved limited genomic scope and limited sample sizes. OBJECTIVES: This study sought to construct a genomic risk score for CAD and to estimate its potential as a screening tool for primary prevention. METHODS: Using a meta-analytic approach to combine large-scale, genome-wide, and targeted genetic association data, we developed a new genomic risk score for CAD (metaGRS) consisting of 1.7 million genetic variants. We externally tested metaGRS, both by itself and in combination with available data on conventional risk factors, in 22,242 CAD cases and 460,387 noncases from the UK Biobank. RESULTS: The hazard ratio (HR) for CAD was 1.71 (95% confidence interval [CI]: 1.68 to 1.73) per SD increase in metaGRS, an association larger than any other externally tested genetic risk score previously published. The metaGRS stratified individuals into significantly different life course trajectories of CAD risk, with those in the top 20% of metaGRS distribution having an HR of 4.17 (95% CI: 3.97 to 4.38) compared with those in the bottom 20%. The corresponding HR was 2.83 (95% CI: 2.61 to 3.07) among individuals on lipid-lowering or antihypertensive medications. The metaGRS had a higher C-index (C = 0.623; 95% CI: 0.615 to 0.631) for incident CAD than any of 6 conventional factors (smoking, diabetes, hypertension, body mass index, self-reported high cholesterol, and family history). For men in the top 20% of metaGRS with >2 conventional factors, 10% cumulative risk of CAD was reached by 48 years of age. CONCLUSIONS: The genomic score developed and evaluated here substantially advances the concept of using genomic information to stratify individuals with different trajectories of CAD risk and highlights the potential for genomic screening in early life to complement conventional risk prediction.