For the previous few years, there was a rising must query and critique the language used to explain synthetic intelligence methods like ChatGPT.
The expertise corporations appear to revert to magical, mystical language which distracts from the realities of what the expertise is, and what it could actually do. On the flip aspect, many within the tutorial group use metaphors like stochastic parrots and bullshit machines to critique the truth that, as I as soon as mentioned (on a sticker!), “chatbots don’t make sense, they make phrases.”
In a brand new paper from Jasper Roe, Mike Perkins and me, we discover how metaphors can be utilized as a automobile for Crucial AI Literacy (CAIL). We additionally outline CAIL, and discover 4 potential lesson actions.
Right here’s the total summary from the paper, Reflecting Actuality, Amplifying Bias? Utilizing Metaphors to Educate Crucial AI Literacy printed in Journal of Interactive Media in Schooling (JIME):
Summary
As instructional establishments grapple with questions on more and more complicated Synthetic Intelligence (AI) methods, discovering efficient strategies for explaining these applied sciences and their societal implications to college students stays a serious problem. This research proposes a methodological method utilising Conceptual Metaphor Concept (CMT) and UNESCO’s AI competency framework to develop actions to foster Crucial AI Literacy (CAIL). By a scientific evaluation of metaphors generally used to explain AI methods, we develop standards for choosing pedagogically acceptable metaphors and display their alignment with established AI literacy competencies, in addition to UNESCO’s AI competency framework.
Our technique identifies and suggests 4 key metaphors for instructing CAIL. This contains AI as a funhouse mirror, a map, an echo chamber, and a black field. Every of those metaphors seeks to handle particular traits of GenAI methods, from filter bubbles to algorithmic opacity. We current these metaphors alongside pedagogical actions designed to interact college students in experiential studying of those ideas. In doing so, we provide educators a structured method to instructing CAIL that touches on features of technical understanding and provokes questions on societal implications. This work contributes to the rising area of AI and training by demonstrating how rigorously chosen metaphors could make complicated technological ideas extra accessible whereas selling CAIL.
Within the paper, we outline Crucial AI Literacy as follows:
The power to critically analyse and have interaction with AI methods by understanding their technical foundations, societal implications, and embedded energy constructions, whereas recognising their limitations, potential biases, and broader social, environmental, and financial impacts.
We then discover some widespread metaphors alongside a number of of our personal invention, and focus on how they could be aligned to the UNESCO AI Curriculum Objectives.
Lastly, we take 4 of the chosen metaphors and develop lesson concepts together with actions and discussions. Right here is the total lesson exercise for the metaphor “AI as Map”:
Exercise 3: AI as a Map – Illustration, Energy and Bias
Studying Aims
- Analyse how AI’s illustration of data parallels historic maps, specializing in inclusivity, bias, and energy.
- Recognise limitations in AI’s illustration of data.
- Critique the metaphor of AI as a map to uncover insights into expertise’s impression on notion and inclusivity.
Addresses Curriculum Aim
CG2.1.1: Floor moral controversies by a vital examination of the use instances of AI instruments in training.
Introducing the Metaphor
The trainer can start by discussing historic maps, such because the Mercator projection, which frequently distorts measurement and centralises Western international locations, as a springboard to AI’s function in shaping views. College students may discover questions resembling what elements of actuality are emphasised or minimised in a map, and why? Who decides what’s “on the map”, and the way does this have an effect on our understanding of the world?
The trainer can introduce AI as a map, illustrating that AI “maps” information by information choice, categorisation, and emphasis, with related energy dynamics shaping what’s seen or hidden. Mirror on the assertion “the map will not be the territory”, exploring how the illustration of the world based mostly on the Giant Language Mannequin dataset will not be a real reflection.
Studying Exercise: Mapping AI’s Data Terrain
In small teams, college students will select an space of data (e.g. cultural heritage, well being info, and social traits) and map it from the angle of an AI software. They need to analyse how AI represents this space, noting:
- What info is instantly accessible on-line that may be “mapped” by information scraping?
- What views or voices are lacking or minimised from the info?
- What biases or assumptions are current in AI output? (by way of the metaphor, how is the “map” completely different to the “territory” or actuality?
Examine and Distinction “Maps”: Teams can then evaluate their findings by analyzing variations in illustration and potential biases.
Dialogue Questions
In analysing AI’s ‘map’ of data, college students observe which views are amplified and that are uncared for, noting that AI typically prioritises dominant narratives whereas overlooking marginalised voices. This selective illustration shapes our notion of data, as AI’s outputs mirror the biases and gaps inherent in its coaching information. The metaphor of ‘AI as a Map’ highlights AI’s impression on information and energy by revealing how sure viewpoints are centred whereas others are diminished, very like historic maps that emphasise the views of these in management. Nonetheless, this metaphor has limitations, as it might indicate a static view of data reasonably than AI’s dynamic interplay with evolving information. Understanding these dynamics encourages a extra accountable method to AI use, prompting customers to critically assess an AI’s outputs and recognise the place broader, extra inclusive views are wanted.
Learn the total paper
The paper is Open Entry and obtainable in full within the new particular version of JIME. You’ll discover the paper alongside a commentary from Emily M. Bender on the well-known ‘Stochastic Parrots’ metaphor, in addition to a group of wonderful articles exploring the conceptual and sensible implications of metaphors for the vital instructing of AI.
Learn the total paper right here.
Entry the whole particular version right here.
Wish to study extra about GenAI skilled improvement and advisory companies, or simply have questions or feedback? Get in contact:
