The impact of generative artificial intelligence on socioeconomic inequalities and policy making.

Generative artificial intelligence (AI) has the potential to both exacerbate and ameliorate existing socioeconomic inequalities. In this article, we provide a state-of-the-art interdisciplinary overview of the potential impacts of generative AI on (mis)information and three information-intensive domains: work, education, and healthcare. Our goal is to highlight how generative AI could worsen existing inequalities while illuminating how AI may help mitigate pervasive social problems. In the information domain, generative AI can democratize content creation and access but may dramatically expand the production and proliferation of misinformation. In the workplace, it can boost productivity and create new jobs, but the benefits will likely be distributed unevenly. In education, it offers personalized learning, but may widen the digital divide. In healthcare, it might improve diagnostics and accessibility, but could deepen pre-existing inequalities. In each section, we cover a specific topic, evaluate existing research, identify critical gaps, and recommend research directions, including explicit trade-offs that complicate the derivation of a priori hypotheses. We conclude with a section highlighting the role of policymaking to maximize generative AI's potential to reduce inequalities while mitigating its harmful effects. We discuss strengths and weaknesses of existing policy frameworks in the European Union, the United States, and the United Kingdom, observing that each fails to fully confront the socioeconomic challenges we have identified. We propose several concrete policies that could promote shared prosperity through the advancement of generative AI. This article emphasizes the need for interdisciplinary collaborations to understand and address the complex challenges of generative AI.

Psychological AI: Designing Algorithms Informed by Human Psychology.

Psychological artificial intelligence (AI) applies insights from psychology to design computer algorithms. Its core domain is decision-making under uncertainty, that is, ill-defined situations that can change in unexpected ways rather than well-defined, stable problems, such as chess and Go. Psychological theories about heuristic processes under uncertainty can provide possible insights. I provide two illustrations. The first shows how recency-the human tendency to rely on the most recent information and ignore base rates-can be built into a simple algorithm that predicts the flu substantially better than did Google Flu Trends's big-data algorithms. The second uses a result from memory research-the paradoxical effect that making numbers less precise increases recall-in the design of algorithms that predict recidivism. These case studies provide an existence proof that psychological AI can help design efficient and transparent algorithms.

How do narratives relate to heuristics?

Narratives and heuristics are both tools for dealing with uncertainty, intractability, and incommensurability, that is, for all real-world situations outside the domain of Bayesian decision theory. But how do narratives and heuristics relate? I suggest two links: Heuristics select narratives to explain events, and "big" narratives select the heuristics that people live by, to execute their values and moral principles.

Heterogeneity of rules in Bayesian reasoning: A toolbox analysis.

How do people infer the Bayesian posterior probability from stated base rate, hit rate, and false alarm rate? This question is not only of theoretical relevance but also of practical relevance in medical and legal settings. We test two competing theoretical views: single-process theories versus toolbox theories. Single-process theories assume that a single process explains people's inferences and have indeed been observed to fit people's inferences well. Examples are Bayes's rule, the representativeness heuristic, and a weighing-and-adding model. Their assumed process homogeneity implies unimodal response distributions. Toolbox theories, in contrast, assume process heterogeneity, implying multimodal response distributions. After analyzing response distributions in studies with laypeople and professionals, we find little support for the single-process theories tested. Using simulations, we find that a single process, the weighing-and-adding model, nevertheless can best fit the aggregate data and, surprisingly, also achieve the best out-of-sample prediction even though it fails to predict any single respondent's inferences. To identify the potential toolbox of rules, we test how well candidate rules predict a set of over 10,000 inferences (culled from the literature) from 4,188 participants and 106 different Bayesian tasks. A toolbox of five non-Bayesian rules plus Bayes's rule captures 64% of inferences. Finally, we validate the Five-Plus toolbox in three experiments that measure response times, self-reports, and strategy use. The most important conclusion from these analyses is that the fitting of single-process theories to aggregate data risks misidentifying the cognitive process. Antidotes to that risk are careful analyses of process and rule heterogeneity across people.

Modeling fast-and-frugal heuristics.

Heuristics are simple rules that experts and laypeople rely on to make decisions under uncertainty as opposed to situations with calculable risk. The research program on fast-and-frugal heuristics studies formal models of heuristics and is motivated by Herbert Simon's seminal work on bounded rationality and satisficing. In this article, we first introduce the major theoretical principles (e.g., ecological rationality) and research approaches (e.g., competitive testing) that have been adopted in this research program, and then illustrate these principles and approaches with two heuristics: take-the-best and fast-and-frugal trees. We describe conditions under which simple heuristics predict as accurately as or better than more complex models, despite requiring less effort. We close by pointing out several issues that need to be further studied and better understood in the research on fast-and-frugal heuristics.

Simple heuristics to run a research group.

Collaboration between researchers has become increasingly common, enabling a level of discovery and innovation that is difficult if not impossible to achieve by a single person. But how can one establish and maintain an environment that fosters successful collaboration within a research group? In this case study, I use my own experience when directing the ABC Research Group at the Max Planck Institute for Human Development in Berlin. I first describe the heuristic principles for setting up a research group, including (i) common topic and multiple disciplines, (ii) open culture, (iii) spatial proximity, and (iv) temporal proximity. Then I describe heuristics for maintaining the open culture, such as setting collective goals, including contrarians, distributing responsibility, making bets, the cake rule, and side-by-side writing. These heuristics form an "adaptive toolbox" that shapes the intellectual and social climate. They create a culture of friendly but rigorous discussion, embedded in a family-like climate of trust where everyone is willing to expose their ignorance and learn from the other members. Feeling accepted and trusted encourages taking the necessary risks to achieve progress in science.

We need to think more about how we conduct research.

Research practice is too often shaped by routines rather than reflection. The routine of sampling subjects, but not stimuli, is a case in point, leading to unwarranted generalizations. It likely originated out of administrative rather than scientific concerns. The routine of sampling subjects and testing their averages for significance is reinforced by delusions about its meaningfulness, including the replicability delusion.

Embodied Heuristics.

Intelligence evolved to cope with situations of uncertainty generated by nature, predators, and the behavior of conspecifics. To this end, humans and other animals acquired special abilities, including heuristics that allow for swift action in face of scarce information. In this article, I introduce the concept of embodied heuristics, that is, innate or learned rules of thumb that exploit evolved sensory and motor abilities in order to facilitate superior decisions. I provide a case study of the gaze heuristic, which solves coordination problems from intercepting prey to catching a fly ball. Various species have adapted this heuristic to their specific sensorimotor abilities, such as vision, echolocation, running, and flying. Humans have enlisted it for solving tasks beyond its original purpose, a process akin to exaptation. The gaze heuristic also made its way into rocket technology. I propose a systematic study of embodied heuristics as a research framework for situated cognition and embodied bounded rationality.

People underestimate the errors made by algorithms for credit scoring and recidivism prediction but accept even fewer errors.

This study provides the first representative analysis of error estimations and willingness to accept errors in a Western country (Germany) with regards to algorithmic decision-making systems (ADM). We examine people's expectations about the accuracy of algorithms that predict credit default, recidivism of an offender, suitability of a job applicant, and health behavior. Also, we ask whether expectations about algorithm errors vary between these domains and how they differ from expectations about errors made by human experts. In a nationwide representative study (N = 3086) we find that most respondents underestimated the actual errors made by algorithms and are willing to accept even fewer errors than estimated. Error estimates and error acceptance did not differ consistently for predictions made by algorithms or human experts, but people's living conditions (e.g. unemployment, household income) affected domain-specific acceptance (job suitability, credit defaulting) of misses and false alarms. We conclude that people have unwarranted expectations about the performance of ADM systems and evaluate errors in terms of potential personal consequences. Given the general public's low willingness to accept errors, we further conclude that acceptance of ADM appears to be conditional to strict accuracy requirements.

Using variation between countries to estimate demand for Cochrane reviews when access is free: a cost-benefit analysis.

OBJECTIVES: Cochrane reviews are currently of limited use as many healthcare professionals and patients have no access to them. Most member states of the Organisation for Economic Co-operation and Development (OECD) choose not to pay for nationwide access to the reviews, possibly uncertain whether there is enough demand to warrant the costs of a national subscription. This study estimates the demand for review downloads and summary views under free access across all OECD countries. DESIGN: The study employs a retrospective design in analysing observational data of web traffic to Cochrane websites in 2014. Specifically, we model for each country downloads of Cochrane reviews and views of online summaries as a function of free access status and alternative sources of variation across countries. The model is then used to estimate demand if a country with restricted access were to purchase free access. We use these estimates to perform a cost-benefit analysis. RESULTS: For one group of eight OECD countries, the additional downloads under free access are estimated to cost between US$4 and more than US$20 each. Three countries are expected to save money under free access, as existing institutional subscriptions would no longer be needed. For the largest group of 17 member states, free access is estimated to cost US$0.05-US$2 per additional review download. On average, the increase in review downloads does not appear to be associated with a decrease in the number of summary views. Instead, translations of plain-language summaries into national languages can serve as an additional strategy for dissemination. CONCLUSIONS: We estimate that free access would cost less than US$2 per additional download for 20 of the 28 OECD countries without national subscriptions, including Canada, Germany and Israel. These countries may be encouraged by our findings to provide free access to their citizens.

Mismatch of corneal specialists' expectations and keratoconus knowledge in general ophthalmologists - a prospective observational study in Switzerland.

BACKGROUND: To assess whether Swiss general ophthalmologists have the minimal keratoconus knowledge that corneal specialists would expect them to have. METHODS: Corneal specialists defined "minimal keratoconus knowledge" (MKK) with respect to definition, risk factors, symptoms and possible treatment options of keratoconus. A telephone interview survey was conducted among one hundred ophthalmologists (mean age 51.9 years (SD 9.5), 60 % male) from the German-speaking part of Switzerland. For each participant, years of work experience, number of keratoconus patients seen per year and access to a topography device were obtained. We calculated the proportion of MKK and examined in multivariate analyses whether ophthalmologists with access to topography and with greater work experience performed better than other groups. RESULTS: No single ophthalmologist had MKK. The mean MKK was 52.0 %, and the range was 28.6-81.0 %. Per 10 years of working in private practice, the MKK decreased by 8.1 % points (95 % CI: -14.2, -2.00; p = 0.01). Only 24 % of participants correctly recalled the definition of keratoconus, 9 % all risk factors, 5 % all symptoms and 20 % all treatment modalities. The MKK values were not associated with the number of keratoconus patients seen per year and the availability of topography to diagnose keratoconus. CONCLUSIONS: There is a substantial mismatch between corneal specialist' expectations and general ophthalmologists' knowledge about keratoconus. The low recall of symptoms and risk factors may explain why ophthalmologists diagnose relatively few cases of keratoconus, resulting in inefficient care delivery and delayed intervention.

Acceptance of criteria for health and driver scoring in the general public in Germany.

Numerous health insurers offer bonus programmes that score customers' health behaviour, and car insurers offer telematics tariffs that score driving behaviour. In many countries, however, only a minority of customers participate in these programmes. In a population-representative survey of private households in Germany (N = 2,215), we study the acceptance of the criteria (features) on which the scoring programmes are based: the features for driver scoring (speed, texting while driving, time of driving, area of driving, accelerating and braking behaviour, respectively) and for health scoring (walking distance per day, sleeping hours per night, alcohol consumption, weight, participation in recommended cancer screenings, smoking status). In a second step, we model participants' acceptance of both programmes with regard to the underlying feature acceptance. We find that insurers in Germany rarely use the features which the participants consider to be the most relevant and justifiable, that is, smoking status for health scoring and smartphone use for driver scoring. Heuristic models (fast-and-frugal trees) show that programme acceptance depends on the acceptance of a few features. These models can help to understand customers' preferences and to design scoring programmes that are based on scientific evidence regarding behaviours and factors associated with good health and safe driving and are thus more likely to be accepted.

Do children have Bayesian intuitions?

Can children solve Bayesian problems, given that these pose great difficulties even for most adults? We present an ecological framework in which Bayesian intuitions emerge from a match between children's numerical competencies and external representations of numerosity. Bayesian intuition is defined here as the ability to determine the exact Bayesian posterior probability by minds untutored in probability theory or in Bayes' rule. As we show, Bayesian intuitions do not require processing of probabilities or Arabic numbers, but basically the ability to count tokens in icon arrays and to understand what to count. A series of experiments demonstrates for the first time that icon arrays elicited Bayesian intuitions in children as young as second-graders for 22% to 32% of all problems; fourth-graders achieved 50% to 60%. Most surprisingly, icon arrays elicited Bayesian intuitions in children with dyscalculia, a specific learning disorder that has been attributed to genetic causes. These children could solve an impressive 50% of Bayesian problems, a level similar to that of children without dyscalculia. By seventh grade, children solved about two thirds of Bayesian problems with natural frequencies alone, without the additional help of icon arrays. We identify four non-Bayesian rules. On the basis of these results, we propose a common solution for the phylogenetic, the ontogenetic, and the 1970s puzzles in the Bayesian literature and argue for a revision of how to teach statistical thinking. In accordance with recent work on infants' numerical abilities, these findings indicate that children have more numerical ability than previously assumed. (PsycInfo Database Record (c) 2021 APA, all rights reserved).

[Assessing the quality of digital health services: How can informed decisions be promoted?] / Einschätzung der Qualität digitaler Gesundheitsangebote: Wie können informierte Entscheidungen gefördert werden?

An important prerequisite for the success of the digitisation of the healthcare system are risk-literate users. Risk literacy means the ability to weigh potential benefits and harms of digital technologies and information, to use digital services critically, and to understand statistical evidence. How do people find reliable and comprehensible health information on the Internet? How can they better assess the quality of algorithmic decision systems? This narrative contribution describes two approaches that show how the competence to make informed decisions can be promoted.Evidence-based and reliable health information exists on the Internet but must be distinguished from a large amount of unreliable information. Various institutions in the German-speaking world have therefore provided guidance to help laypersons make informed decisions. The Harding Center for Risk Literacy in Potsdam, for example, has developed a decision tree ("fast-and-frugal tree"). When dealing with algorithms, natural frequency trees (NFTs) can help to assess the quality and fairness of an algorithmic decision system.Independent of reliable and comprehensible digital health services, further tools for laypersons to assess information and algorithms should be developed and provided. These tools can also be included in institutional training programmes for the promotion of digital literacy. This would be an important step towards the success of digitisation in prevention and health promotion.

How to Explain Behavior?

Unlike behaviorism, cognitive psychology relies on mental concepts to explain behavior. Yet mental processes are not directly observable and multiple explanations are possible, which poses a challenge for finding a useful framework. In this article, I distinguish three new frameworks for explanations that emerged after the cognitive revolution. The first is called tools-to-theories: Psychologists' new tools for data analysis, such as computers and statistics, are turned into theories of mind. The second proposes as-if theories: Expected utility theory and Bayesian statistics are turned into theories of mind, describing an optimal solution of a problem but not its psychological process. The third studies the adaptive toolbox (formal models of heuristics) that describes mental processes in situations of uncertainty where an optimal solution is unknown. Depending on which framework researchers choose, they will model behavior in either situations of risk or of uncertainty, and construct models of cognitive processes or not. The frameworks also determine what questions are asked and what kind of data are generated. What all three frameworks have in common, however, is a clear preference for formal models rather than explanations by general dichotomies or mere verbal concepts. The frameworks have considerable potential to inform each other and to generate points of integration.

Effect of Tabular and Icon Fact Box Formats on Comprehension of Benefits and Harms of Prostate Cancer Screening: A Randomized Trial.

BACKGROUND: Fact boxes employ evidence-based guidelines on risk communication to present benefits and harms of health interventions in a balanced and transparent format. However, little is known about their short- and long-term efficacy and whether designing fact boxes to present multiple outcomes with icon arrays would increase their efficacy. METHOD: In study 1, 120 men (30-75 y) completed a lab study. Participants were randomly assigned to 1 of 3 fact box formats on prostate cancer screening: a tabular fact box with numbers, a fact box with numbers and icon array, and a fact box with numbers, separate icon arrays, and text to describe each benefit and harm. Comprehension of information (while materials were present) and short-term knowledge recall were assessed. Study 2 recruited an online sample of 244 German men (40-75 y). Participants were randomly assigned to 1 of the 3 fact box formats or widely distributed health information, and knowledge was assessed at baseline, shortly after presentation, and at 6-mo follow-up, along with comprehension while materials were present. RESULTS: In both studies, comprehension and knowledge-recall scores were similar when comparing tabular and icon fact boxes. In the 6-mo follow-up, this positive effect on knowledge recall disappeared. Fact boxes increased knowledge relative to baseline but did not affect decision intentions or perceptions of having complete information to make decisions. CONCLUSIONS: This study shows that fact boxes with and without icon arrays are equally effective at improving comprehension and knowledge recall over the short-term and are simple formats that can improve on current health information. Specifically, if fact boxes are used at the time or immediately before a decision is made, they promote informed decisions about prostate cancer screening.

US gynecologists' estimates and beliefs regarding ovarian cancer screening's effectiveness 5 years after release of the PLCO evidence.

Efficient patient care requires the conscientious use of current best evidence. Such evidence on ovarian cancer screening showed that the screening has no survival benefit but considerable harms; currently no medical organization recommends it. In a cross-sectional online survey study with 401 US outpatient gynecologists we investigated whether they follow the recommendation of their medical organizations in daily practice and report estimates of ovarian cancer screening's effectiveness that approximate current best evidence (within a ± 10 percent margin of error), and if not, whether a fact box intervention summarizing current best evidence improves judgments. Depending on question, 44.6% to 96.8% reported estimates and beliefs regarding screening's effectiveness that diverged from evidence, and 57.6% reported regularly recommending the screening. Gynecologists who recommend screening overestimated the benefit and underestimated the harms more frequently. After seeing the fact box, 51.6% revised initial estimates and beliefs, and the proportion of responses approximating best evidence increased on all measures (e.g., mortality reduction: 32.9% [95% CI, 26.5 to 39.7] before intervention, 77.3% [71.0 to 82.8] after intervention). Overall, results highlight the need for intensified training programs on the interpretation of medical evidence. The provision of fact box summaries in medical journals may additionally improve the practice of evidence-based medicine.

Assessing minimal medical statistical literacy using the Quick Risk Test: a prospective observational study in Germany.

OBJECTIVES: To assess minimal medical statistical literacy in medical students and senior educators using the 10-item Quick Risk Test; to assess whether deficits in statistical literacy are stable or can be reduced by training. DESIGN: Prospective observational study on the students, observational study on the university lecturers. SETTING: Charité University Medicine medical curriculum for students and a continuing medical education (CME) course at a German University for senior educators. PARTICIPANTS: 169 students taking part in compulsory final-year curricular training in medical statistical literacy (63% female, median age 25 years). Sixteen professors of medicine and other senior educators attending a CME course on medical statistical literacy (44% female, age range=30-65 years). INTERVENTIONS: Students completed a 90 min training session in medical statistical literacy. No intervention for the senior educators. OUTCOME MEASURES: Primary outcome measure was the number of correct answers out of four multiple-choice alternatives per item on the Quick Risk Test. RESULTS: Final-year students answered on average half (median=50%) of the questions correctly while senior educators answered three-quarters correctly (median=75%). For comparison, chance performance is 25%. A 90 min training session for students increased the median percentage correct from 50% to 90%. 82% of participants improved their performance. CONCLUSIONS: Medical students and educators do not master all basic concepts in medical statistics. This can be quickly assessed with the Quick Risk Test. The fact that a 90 min training session on medical statistical literacy improves students' understanding from 50% to 90% indicates that the problem is not a hard-wired inability to understand statistical concepts. This gap in physicians' education has long-lasting effects; even senior medical educators could answer only 75% of the questions correctly on average. Hence, medical students and professionals should receive enhanced training in how to interpret risk-related medical statistics.

Less can be more: How to make operations more flexible and robust with fewer resources.

We review empirical evidence from practice and general theoretical conditions, under which simple rules of thumb can help to make operations flexible and robust. An operation is flexible when it responds adaptively to adverse events such as natural disasters; an operation is robust when it is less affected by adverse events in the first place. We illustrate the relationship between flexibility and robustness in the context of supply chain risk. In addition to increasing flexibility and robustness, simple rules simultaneously reduce the need for resources such as time, money, information, and computation. We illustrate the simple-rules approach with an easy-to-use graphical aid for diagnosing and managing supply chain risk. More generally, we recommend a four-step process for determining the amount of resources that decision makers should invest in so as to increase flexibility and robustness.