Section outline

  • LESSON 4 - ARTIFICIAL INTELLIGENCE FOUNDATIONS FOR BROADENING CULTURAL HORIZONS

    What Kinds of Mistakes Do AI Models Make?


    In this lesson, we analyze two types of errors language models can make: classification errors and hallucinations. We also anticipate another source of error: social biases, also known as stereotypes. A classification error occurs when the model assigns an incorrect category to a text or image. Hallucinations, on the other hand, involve presenting false or generated information as if it were true, even citing non-existent sources that seem plausible at first glance. These situations can reinforce automation bias, as we tend to trust responses that sound technical or well-founded. We review complex examples, such as incorrect medical interpretations of X-rays or errors in dosage calculations, even when the model uses official sources via Retrieval-Augmented Generation (RAG) systems.

    As a case study on classification errors, we analyze the difficulty of distinguishing human-written text from AI-generated text, and how even detection tools can fail. Through concrete examples, we examine how models can mix up information, commit inaccuracies, or make mathematical miscalculations. This lesson invites us to present a critical mindset by recognizing the ways AI makes mistakes and why those errors are often difficult to detect.

    At the end of the lesson, you will be asked to answer a few questions. 


    👩‍🏫 Instructor: Luciana Benotti

    • What Kinds of Mistakes Do AI Models Make? URL

      Prefer to watch in a language other than Spanish? Just turn on CC, go to Settings (⚙️) > Subtitles > Auto-translate, and select your language!

    • Below you will find the lesson slides, designed as a reference to revisit the ideas, questions, and tensions raised in the video. (Please note: The slides are in Spanish).

      📚 Key Concepts

      Hallucination (in AI): A phenomenon in which a language model generates information that is factually false, non-existent, or unsupported by real data, but presents it with an appearance of absolute certainty and syntactic coherence.

      Stereotype: An image or idea commonly accepted by a group or society as immutable.

      Reference: Royal Spanish Academy (RAE). Estereotipo. In Dictionary of the Spanish Language. Retrieved from https://dle.rae.es/estereotipo

      In AI, it refers to the reproduction or amplification of overgeneralized beliefs (also known as social biases) about specific groups of people, which can be incorporated into language models through training data or methods.

      Example:       A language model that consistently associates certain professions with specific genders, such as "nurse" with women and "engineer" with men.

      Out-group stereotype: Perceptions, often simplified or distorted, that members of a group hold about another group to which they do not belong.

      In-group stereotype: The beliefs, images, and descriptions that members of a social group hold about themselves.

      Automation bias: The human propensity to excessively or uncritically trust suggestions from an automated system, even ignoring one's own logic or contradictory information.

      Word embedding: A technique for representing words as dense numerical vectors in a multidimensional space. This representation captures semantic and syntactic relationships between words, allowing machines to process language more effectively.

      Example: In a word embedding model, semantically similar words like "king" and "queen" would be close together in the vector space, whereas "king" and "apple" would be further apart.

      A creation by Fundación Vía Libre in collaboration with FAMAF – UNC.

    • Assignment: Typical Phrases in EDIA

      Below you will find two videos to help you complete this activity. First, we explain how to use the EDIA tool, how to log in, and how to enter your school's information.

      Access the EDIA tool to start the activity: https://edia.ngrok.app/

      Part 1: How to log in to EDIA
      In this video, we show you how to log in. Save this information for upcoming activities, as we will continue using EDIA.



      Part 2: Typical Phrases Activity: Let's Get Started!
      Below we explain how to complete the activity. We have also included supplementary written materials with the same instructions.