Presentacio sobre l'Oficina de Proteccio de Dades Personals de la UAB i la politica Open Science. Va formar part de la conferencia "Les politiques d'Open Data / Open Acces: Implicacions a la recerca" orientada a investigadors i gestors de projectes europeus que va tenir lloc el 20 de setembre de 2018 a la Universitat Autonoma de Barcelona

/pdf/general-data-protection-regulation-2z11cloyp7.pdf

General data protection regulation

With machine learning models being increasingly used to aid decision making even in high-stakes domains, there has been a growing interest in developing interpretable models. Although many supposedly interpretable models have been proposed, there have been relatively few experimental studies investigating whether these models achieve their intended effects, such as making people more closely follow a model's predictions when it is beneficial for them to do so or enabling them to detect when a model has made a mistake. We present a sequence of pre-registered experiments (N=3,800) in which we showed participants functionally identical models that varied only in two factors commonly thought to make machine learning models more or less interpretable: the number of features and the transparency of the model (i.e., whether the model internals are clear or black box). Predictably, participants who saw a clear model with few features could better simulate the model's predictions. However, we did not find that participants more closely followed its predictions. Furthermore, showing participants a clear model meant that they were less able to detect and correct for the model's sizable mistakes, seemingly due to information overload. These counterintuitive findings emphasize the importance of testing over intuition when developing interpretable models.

Manipulating and Measuring Model Interpretability

Problem Given the number of times in which an unknown event has happened and failed: Required the chance that the probability of its happening in a single trial lies somewhere between any two degrees of probability that can be named. SECTION 1 Definition 1. Several events are inconsistent, when if one of them happens, none of the rest can. 2. Two events are contrary when one, or other of them must; and both together cannot happen. 3. An event is said to fail, when it cannot happen; or, which comes to the same thing, when its contrary has happened. 4. An event is said to be determined when it has either happened or failed. 5. The probability of any event is the ratio between the value at which an expectation depending on the happening of the event ought to be computed, and the value of the thing expected upon it’s 2 happening.

An essay towards solving a problem in the doctrine of chances. [Facsimil]

Expert Political Judgment: How Good Is It? How Can We Know? By Philip E. Tetlock. Princeton: Princeton University Press, 2005. 352p. $45.00 cloth, $19.95 paper. This is a wonderful and important book. Philip Tetlock is a political psychologist who has a knack for innovative research projects (e.g., his earlier work on how people cope with trade-offs in politics). In this book, he addresses a question that would scare away more timid souls: How well do experts predict political and economic events?

Expert Political Judgment: How Good Is It? How Can We Know?

Deep learning bears promise for drug discovery, including advanced image analysis, prediction of molecular structure and function, and automated generation of innovative chemical entities with bespoke properties. Despite the growing number of successful prospective applications, the underlying mathematical models often remain elusive to interpretation by the human mind. There is a demand for ‘explainable’ deep learning methods to address the need for a new narrative of the machine language of the molecular sciences. This Review summarizes the most prominent algorithmic concepts of explainable artificial intelligence, and forecasts future opportunities, potential applications as well as several remaining challenges. We also hope it encourages additional efforts towards the development and acceptance of explainable artificial intelligence techniques. Drug discovery has recently profited greatly from the use of deep learning models. However, these models can be notoriously hard to interpret. In this Review, Jimenez-Luna and colleagues summarize recent approaches to use explainable artificial intelligence techniques in drug discovery.

/pdf/drug-discovery-with-explainable-artificial-intelligence-5ed06p2guc.pdf

Drug discovery with explainable artificial intelligence

A surge of interest in explainable AI (XAI) has led to a vast collection of algorithmic work on the topic. While many recognize the necessity to incorporate explainability features in AI systems, how to address real-world user needs for understanding AI remains an open question. By interviewing 20 UX and design practitioners working on various AI products, we seek to identify gaps between the current XAI algorithmic work and practices to create explainable AI products. To do so, we develop an algorithm-informed XAI question bank in which user needs for explainability are represented as prototypical questions users might ask about the AI, and use it as a study probe. Our work contributes insights into the design space of XAI, informs efforts to support design practices in this space, and identifies opportunities for future XAI work. We also provide an extended XAI question bank and discuss how it can be used for creating user-centered XAI.

Questioning the AI: Informing Design Practices for Explainable AI User Experiences

As more and more forms of AI become prevalent, it becomes increasingly important to understand how people develop mental models of these systems. In this work we study people's mental models of AI in a cooperative word guessing game. We run think-aloud studies in which people play the game with an AI agent; through thematic analysis we identify features of the mental models developed by participants. In a large-scale study we have participants play the game with the AI agent online and use a post-game survey to probe their mental model. We find that those who win more often have better estimates of the AI agent's abilities. We present three components for modeling AI systems, propose that understanding the underlying technology is insufficient for developing appropriate conceptual models (analysis of behavior is also necessary), and suggest future work for studying the revision of mental models over time.

https://dl.acm.org/doi/pdf/10.1145/3313831.3376316

Mental Models of AI Agents in a Cooperative Game Setting

Proper Bayesian reasoning is critical in a variety of domains that require practitioners to make predictions about the probability of events contingent upon earlier actions or events. However, much research on judgment has shown that people who are unfamiliar with Bayes’ Theorem often reason quite poorly with conditional probabilities due to various cognitive biases. Owing to previous successes of visualization techniques for debiasing judges and improving judgment performance, we created an interactive computer visualization designed to aid Bayes-naive people in solving conditional probability problems that would not require a training period to use, and would be flexible enough to accommodate many problem types. Results are suggestive that participants using our interactive visualization were able to substantially improve their Bayesian reasoning performance above that of previous debiasing methods. This finding has significant implications for expanding the toolbox of techniques that can be used to mor...

Interactive Visualizations to Improve Bayesian Reasoning

Researchers have long known that aggregate estimations built from the collected opinions of a large group of people often outperform the estimations of individual experts. This phenomenon is generally described as the "Wisdom of Crowds". This approach has shown promise with respect to the task of accurately forecasting future events. Previous research has demonstrated the value of utilizing meta-forecasts (forecasts about what others in the group will predict) when aggregating group predictions. In this paper, we describe an extension to meta-forecasting and demonstrate the value of modeling the familiarity among a population's members (its social network) and applying this model to forecast aggregation. A pair of studies demonstrates the value of taking this model into account, and the described technique produces aggregate forecasts for future events that are significantly better than the standard Wisdom of Crowds approach as well as previous meta-forecasting techniques.

Sarah Miller

Papers

Questioning the AI: Informing Design Practices for Explainable AI User Experiences

Questioning the AI: Informing Design Practices for Explainable AI User Experiences

Mental Models of AI Agents in a Cooperative Game Setting

Interactive Visualizations to Improve Bayesian Reasoning

Crowdsourcing the future: predictions made with a social network