http://www.semanlink.net/tag/ai_and_trust Documents tagged with Ai and trust http://www.semanlink.net/doc/2025/06/michael_i_jordan_lecture_at_th The Current Dialog on LLMs is Missing Something Fundamental • A lack of focus on collectives • A lack of focus on uncertainty • A lack of focus on incentives What's missing is microeconomics. **Microeconomics generates a complementary kind of intelligence to that of prediction and optimization** Difficulty to cope with **uncertainty**. LLMs do not really know what "knowing" means. "ChatGPT are you sure about what you just wrote?". Overconfidence. Uncertainty: statistics, but also: what is it I know and you don't know, etc "Collectives" (made of humans and/or AIs) may help. Collectives provide context for decision-making under uncertainty Situation : Information flows between entities in a collective. Need to take into account asymetries of information (a special form of uncertainty != statical one, critical in economics, eg. when setting a price). We want to connect people and machines and things that give value to people and entities (not just connecting as Facebook does). More complex than just connecting entities and make predictions -> economics - what motivates people to connect? - and what prevents them from lying? (answer: nothing. I think there is some lie on FB.) An LLM cannot know everything about my current situation (so cannot provide a good answer to me). I am ready to interact, but I don't want to be in the current advertising model based on stats of my browsing history (I cannot trust such a model! It wants to sell to me. It doesn't work for me) I want more of an economics model, and I want to do that in the context of markets. 2025-06-05T13:50:19Z http://www.semanlink.net/doc/2023/01/2301_04709_causal_abstraction > A faithful and interpretable explanation of an AI model's behavior and internal structure is a **high-level explanation that is human-intelligible but also consistent with the known, but often opaque low-level causal details of the model**. We argue that the theory of **causal abstraction** provides the mathematical foundations for the desired kinds of model explanations > We take the fundamental question in explainable artificial intelligence (XAI) to be why a deep learning model makes the predictions it does. > XAI needs a theory for when a high-level causal explanation [that is, interpretable by humans] is harmonious with a low-level causal explanation. > A high-level (possibly symbolic) model is a faithful proxy for a low-lever (in our setting, usually neural) model when we can align high-level variables with sets of low-level variables that play the same causal role 2023-01-14T23:21:46Z http://www.semanlink.net/doc/2022/10/uncertainty_estimation_for_nlp_ 2022-10-18T15:05:39Z http://www.semanlink.net/doc/2022/10/tutorial_on_uncertainty_estimat 2022-10-18T15:02:39Z http://www.semanlink.net/doc/2022/09/2010_00711_a_survey_of_the_st 2022-09-08T09:30:14Z http://www.semanlink.net/doc/2022/09/uncertainty_estimation_for_natu Accurate estimates of uncertainty are important for many difficult or sensitive prediction tasks in natural language processing (NLP). Though large-scale pre-trained models have vastly improved the accuracy of applied machine learning models throughout the field, there still are many instances in which they fail. The ability to precisely quantify uncertainty while handling the challenging scenarios that modern models can face when deployed in the real world is critical for reliable, consequential-decision making. This tutorial is intended for both academic researchers and industry practitioners alike, and provides a comprehensive introduction to uncertainty estimation for NLP problems---from fundamentals in probability calibration, Bayesian inference, and confidence set (or interval) construction, to applied topics in modern out-of-distribution detection and selective inference. 2022-09-07T18:48:16Z http://www.semanlink.net/doc/2022/08/2208_11857_shortcut_learning_ 2022-08-27T10:39:46Z http://www.semanlink.net/doc/2022/08/anthropic_sur_twitter_we_exa 2022-08-25T18:31:06Z