On the risk of confusing interpretability with explicability

Herzog, Christian

doi:10.1007/s43681-021-00121-9

Cited by 15 publications

(5 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To be sure, explainability may often be important in some public-facing practices (e.g., medicine and health care). Still, in such circumstances, while ethics is internal to the practice, it can, in principle, be disconnected from epistemology (Herzog 2022). 7 The need for transparency is a controversial point in the literature, as we have just seen with Durán and Formanek, but is also clear from other contributions such as Lenhard and Winsberg (2010) or Humphreys (2009).…”

Section: From the Reliability Of The Outcome To The Reliability Of Th...mentioning

confidence: 99%

Connecting ethics and epistemology of AI

2023

View full text Add to dashboard Cite

The need for fair and just AI is often related to the possibility of understanding AI itself, in other words, of turning an opaque box into a glass box, as inspectable as possible. Transparency and explainability, however, pertain to the technical domain and to philosophy of science, thus leaving the ethics and epistemology of AI largely disconnected. To remedy this, we propose an integrated approach premised on the idea that a glass-box epistemology should explicitly consider how to incorporate values and other normative considerations, such as intersectoral vulnerabilities, at critical stages of the whole process from design and implementation to use and assessment. To connect ethics and epistemology of AI, we perform a double shift of focus. First, we move from trusting the output of an AI system to trusting the process that leads to the outcome. Second, we move from expert assessment to more inclusive assessment strategies, aiming to facilitate expert and non-expert assessment. Together, these two moves yield a framework usable for experts and non-experts when they inquire into relevant epistemological and ethical aspects of AI systems. We dub our framework ‘epistemology-cum-ethics’ to signal the equal importance of both aspects. We develop it from the vantage point of the designers: how to create the conditions to internalize values into the whole process of design, implementation, use, and assessment of an AI system, in which values (epistemic and non-epistemic) are explicitly considered at each stage and inspectable by every salient actor involved at any moment.

show abstract

Section: From the Reliability Of The Outcome To The Reliability Of Th...mentioning

confidence: 99%

Connecting ethics and epistemology of AI

2023

View full text Add to dashboard Cite

show abstract

“…Another more complex concern, also related to explainability, is the principle of explicability, a concept that combines intelligibility and accountability as the basis of an AI interpretable model [63]. The latter concept points to the importance of transparency, in the sense that all the procedures and details used to build, training and test the AI model during its development and use should be present.…”

Section: Explainable Algorithmsmentioning

confidence: 99%

Trust in algorithms. An experimental approach

Marmolejo-Ramos

2023

Preprint

View full text Add to dashboard Cite

Algorithms have become ubiquitous in our day-to-day activities. Their presence ranges from low-stakes decisions such as what music we should listen to, to high-stakes decisions regarding promotion decisions. Despite the frequency and range with which we interact with algorithms, academic research, as well as public outcries, suggest that humans often express distrust of algorithms and their decisions. Previous research suggests that there may be ways to mitigate this distrust through educational efforts that help explain how algorithms operate. With a sample of 1,841 participants across 19 countries, the present study investigates these notion by examining variations in algorithmic trust across low and high-stakes decisions, and the role of explainability and statistical literacy in shaping reactions. Broadly, results suggest a relationship between how much one understands about algorithms and their propensity to trust. Implicationsof these findings are discussed.

show abstract

“…In order to use the best classifier for the following tests, and include a model in further implementations, we select Random Forest. This selection corresponds to the capability to detect bots (true negative values in the confusion matrices), and the possibility of incorporate more explicability of our models [81]. Additionally, Random Forest only learns from a small subset of features and the random feature selection makes the trees more independent of one another than other classifiers, which frequently yields better predictive performance (due to better variance-bias trade-offs).…”

Section: ) Internal Validationmentioning

confidence: 99%

CrediBot: Applying Bot Detection for Credibility Analysis on Twitter

Aguilera,

Quinteros,

Dongo

et al. 2023

IEEE Access

View full text Add to dashboard Cite

Nowadays, people and organizations use social networks for allowing and facilitating the transfer of information among groups that share similar interests. Due to the wide repertoire of users that these social platforms have and the amount of information generated within them, the presence of bots has become a relevant issue, both to facilitate the sharing of true information or to disseminate false information (fake news). In the second case, bots could manipulate political opinions, be perpetrators of identity or information theft, among other possible dangers that can cause when interacting on the platform. Thus, the identification of bots in social networks can become a useful practice to evaluate credibility or to detect fake news. In this work, we extend a previously proposed credibility model for Twitter, by incorporating bot detection. The original model calculates the credibility of tweets based on three measures: text, account/user, and social impact, using different filters to analyse text (SPAM, bad words,and good spelling) and account attributes (e.g., creation date, followers, following) to calculate account/user and social credibility scores. The extended model considers in the user credibility, the bot verification. Additionally, the extended credibility model is implemented in T-CREo, a framework for real time credibility analysis. For bot detection, some machine learning algorithms for supervised learning, such as AdaBoost, Bagging, Decision Tree, Logistic Regression, and Random Forest are trained and evaluated. Results show that the best algorithm is the Random Forest for its capacity of generalization with an accuracy and F1-score values over 97% both in English and Spanish. The evaluation of the bot detection functionality in the credibility analysis shows a performance of precision=1.0, recall=0.8462, F1-score=0.9167, and accuracy=0.92 for both English and Spanish models in our validation tests.

show abstract

On the risk of confusing interpretability with explicability

Cited by 15 publications

References 24 publications

Connecting ethics and epistemology of AI

Connecting ethics and epistemology of AI

Trust in algorithms. An experimental approach

CrediBot: Applying Bot Detection for Credibility Analysis on Twitter

Contact Info

Product

Resources

About