{"id":544,"date":"2025-01-09T22:58:24","date_gmt":"2025-01-10T03:58:24","guid":{"rendered":"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/chapter\/evaluating-ai-generated-content\/"},"modified":"2025-09-07T12:53:21","modified_gmt":"2025-09-07T16:53:21","slug":"evaluating-ai-generated-content","status":"publish","type":"chapter","link":"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/chapter\/evaluating-ai-generated-content\/","title":{"raw":"Evaluating AI-generated content","rendered":"Evaluating AI-generated content"},"content":{"raw":"[caption id=\"attachment_405\" align=\"aligncenter\" width=\"1200\"]<a href=\"-revised\/chapter\/182\/ai-am-over-it_nadia-piet-aixdesign_archival-images-of-ai_2565x1854\/\" rel=\"attachment wp-att-405\"><img class=\"size-full wp-image-405\" src=\"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-content\/uploads\/sites\/2495\/2025\/01\/ai-am-over-it_nadia-piet-aixdesign_archival-images-of-ai_2565x1854.png\" alt=\"A red-toned illustration shows a man's head surrounded by swirling AI icons, with small, mischievous witch-like figures flying around him. The man's expression appears disoriented and fatigued, symbolizing the mental overload caused by the overwhelming flood of AI tools and news. The witches represent the chaotic, cackling nature of rapid AI developments, adding to the sense of dizziness and confusion.\" width=\"1200\" height=\"867\" \/><\/a> <a href=\"https:\/\/nadiapiet.com\">Nadia Piet<\/a> + <a href=\"https:\/\/aixdesign.co\/posts\/archival-images-of-ai\">AIxDESIGN &amp; Archival Images of AI<\/a> \/ <a href=\"https:\/\/www.betterimagesofai.org\">Better Images of AI<\/a> \/ AI Am Over It \/ <a href=\"https:\/\/creativecommons.org\/licenses\/by\/4.0\/\">CC-BY 4.0<\/a>[\/caption]\r\n\r\nYou should evaluate the quality and reliability of AI-generated content before relying on it, just as you would information from any source. Information provided by generative AI tools may be:\r\n<ul>\r\n \t<li>incorrect<\/li>\r\n \t<li>out of date<\/li>\r\n \t<li>biased or offensive<\/li>\r\n \t<li>lacking common sense<\/li>\r\n \t<li>lacking originality.<\/li>\r\n<\/ul>\r\nAI tools tend to produce 'middle-of-the-road' answers, based on a consensus of the most common information in the AI's training data. You should continue to think critically as you use the tools for your learning. Ask yourself:\r\n<ul>\r\n \t<li>is the response you've been given too conservative?<\/li>\r\n \t<li>is there an alternative viewpoint that has been missed?<\/li>\r\n \t<li>what are your views \u2014 do you disagree with the information?<\/li>\r\n<\/ul>\r\n<h2>Methods for evaluating information<\/h2>\r\nThere are many methods for evaluating information. The <strong>TRAAP<\/strong> test is useful to consider when evaluating information generally and also emphasises some of the challenges with the evaluation of AI-generated content.\r\n<div class=\"textbox\">\r\n\r\n<img class=\"alignnone wp-image-529\" src=\"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-content\/uploads\/sites\/2495\/2023\/03\/alert-circle.png\" alt=\"important icon\" width=\"30\" height=\"30\" \/> <a id=\"traap\"><\/a>Applying the TRAAP test\r\n<ul>\r\n \t<li>Timeliness<\/li>\r\n \t<li>Relevance<\/li>\r\n \t<li>Authority<\/li>\r\n \t<li>Accuracy<\/li>\r\n \t<li>Purpose<\/li>\r\n<\/ul>\r\n<\/div>\r\n<h2>Challenges<\/h2>\r\nDifferent generative AI tools have different limitations. It is not always clear how <strong>current\u00a0<\/strong>information contained in LLMs is.\u00a0 <a href=\"https:\/\/openai.com\/index\/gpt-4-research\/\">OpenAI's documentation<\/a> states that GPT-4 \"generally lacks knowledge of events that have occurred after the vast majority of its data cuts off (September 2021)\". It is, however, capable of searching the web to find more recent information.\r\n\r\nLLMs may not always present you with the sources for answers, or rely on unsuitable sources. This can make it difficult to judge the <strong>relevance<\/strong>, <strong>authority<\/strong>, <strong>accuracy<\/strong> and <strong>purpose<\/strong> of the information.\r\n<div class=\"textbox shaded\">\r\n\r\nI can help with a wide range of topics, but there are some limitations. For example, I don\u2019t have access to:\r\n<ul>\r\n \t<li><strong>Personal data<\/strong>\u00a0unless shared with me during our conversation.<\/li>\r\n \t<li><strong>Real-time data<\/strong>\u00a0like live sports scores or stock prices.<\/li>\r\n \t<li><strong>Confidential or proprietary information<\/strong>.<\/li>\r\n \t<li><strong>Certain copyrighted content<\/strong>\u00a0in full, such as books, articles, or songs.<\/li>\r\n<\/ul>\r\n<em>Source: Answer provided by Microsoft Copilot on 26 November 2024.<\/em>\r\n\r\n<\/div>\r\n<h2><a id=\"tips\"><\/a><a id=\"tips\"><\/a>Tips for confirming the information provided by AI tools<\/h2>\r\n<ol>\r\n \t<li>Ask the tool to provide you with <strong>sources<\/strong>. You can ask for a specific type of source (peer reviewed journal articles, news articles or academic sources). You can provide other constraints such as a time limit, e.g. 'Can you provide academic sources from the last 5 years?'. Writing your prompt in academic or formal language will increase the chance of getting those types of sources. Note that there's no guarantee that the AI tool will give you what you ask for but these techniques can increase the chance of better outcomes.<\/li>\r\n \t<li>Locate the sources provided and <strong>confirm the information is<\/strong>\u00a0<strong>real<\/strong>. Generative AI tools will <a href=\"https:\/\/en.wikipedia.org\/wiki\/Hallucination_(artificial_intelligence)\" target=\"_blank\" rel=\"noopener\">present false information as fact<\/a> and make up references.<\/li>\r\n \t<li>Once you confirm the sources, consider their <strong>quality<\/strong> and whether they are <strong>appropriate<\/strong> for your task.<\/li>\r\n \t<li>Look for <strong>other reputable sources<\/strong> that also confirm the information.<\/li>\r\n<\/ol>\r\n<div class=\"textbox shaded\">\r\n<blockquote>\"Treat the AI like a slightly unreliable friend. Have a chat, ask some questions. Don\u2019t trust the answers though.\"\r\n\r\n<a href=\"\/\" rel=\"attachment wp-att-34\"><img class=\"alignnone wp-image-34\" src=\"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-content\/uploads\/sites\/2495\/2025\/01\/book-open-bookmark.png\" alt=\"read icon\" width=\"30\" height=\"30\" \/><\/a> <a href=\"https:\/\/www.cilip.org.uk\/news\/661388\/Can-AI-do-your-reading-for-you--should-it.htm\" target=\"_blank\" rel=\"noopener\">Can AI do your reading for you and should it?<\/a><\/blockquote>\r\n<\/div>\r\n<h2>Human in the loop<\/h2>\r\nEvaluating the outputs of AI tools is sometimes referred to as \"human-in-the-loop\" work. Many of the AI models are based on predictive modelling and contextual understanding of the prompts they\u2019re given. These models make mistakes!\r\n<div class=\"textbox shaded\">\r\n\r\n<a href=\"\/\" rel=\"attachment wp-att-34\"><img class=\"alignnone wp-image-34\" src=\"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-content\/uploads\/sites\/2495\/2025\/01\/book-open-bookmark.png\" alt=\"read icon\" width=\"30\" height=\"30\" \/><\/a> Users of a new Google AI feature were told to <a href=\"https:\/\/www.bbc.com\/news\/articles\/cd11gzejgz4o\">eat rocks and add glue to pizza.<\/a>\r\n\r\n<\/div>\r\nConstant feedback by the human-in-the-loop can improve your specific output and also the AI tools and models \"and enhance the accuracy, reliability, and adaptability of ML systems, harnessing the unique capabilities of both humans and machines\" (Source: <a href=\"https:\/\/cloud.google.com\/discover\/human-in-the-loop#what-is-human-in-the-loop-hitl-in-ai-ml\">What is Human-in-the-Loop in AI &amp; ML?<\/a>).\r\n<div class=\"textbox textbox--exercises\"><header class=\"textbox__header\">\r\n<h2 class=\"textbox__title\">Activity<\/h2>\r\n<\/header>\r\n<div class=\"textbox__content\">\r\n\r\nChoose a topic you know well \u2014 this could be a hobby, sport, musical instrument, game, or any area where you have confidence in your knowledge.\r\n<ol>\r\n \t<li>Ask an AI tool (such as ChatGPT, Bing Copilot, or Google Gemini) a question about this topic. For example, \u201cHow do you tune a guitar?\u201d or \u201cWhat are the offside rules in soccer?\u201d<\/li>\r\n \t<li>Carefully read the AI-generated response and evaluate it for accuracy, clarity, and completeness.<\/li>\r\n \t<li>Identify any inaccuracies, misleading explanations, or gaps in the information provided.<\/li>\r\n<\/ol>\r\nConsider\r\n<ul>\r\n \t<li>What did the AI tool get right?<\/li>\r\n \t<li>What did it miss or get wrong?<\/li>\r\n \t<li>Did knowing the topic well help you spot issues?<\/li>\r\n \t<li>Does this exercise shape your view on using AI tools for learning?<\/li>\r\n<\/ul>\r\n<h3>Example<\/h3>\r\n<strong>A student asked ChatGPT:<\/strong> \u201cWhat are the strings on a standard 6-string guitar tuned to?\u201d\r\n\r\n<strong>AI Response:<\/strong> \u201cThe strings on a standard 6-string guitar are tuned to E-B-G-D-A-E, from the lowest (thickest) string to the highest (thinnest).\u201d\r\n\r\n<strong>What\u2019s incorrect: <\/strong>This is a reversal. The correct tuning from the lowest (thickest) string to the highest (thinnest) is E-A-D-G-B-E. The AI listed the strings in reverse order, which could confuse a beginner.\r\n\r\n<strong>Reflection: <\/strong>A student familiar with the guitar would immediately recognise the error, but someone new might unknowingly accept the incorrect answer.\r\n\r\n<\/div>\r\n<\/div>\r\n&nbsp;","rendered":"<figure id=\"attachment_405\" aria-describedby=\"caption-attachment-405\" style=\"width: 1200px\" class=\"wp-caption aligncenter\"><a href=\"-revised\/chapter\/182\/ai-am-over-it_nadia-piet-aixdesign_archival-images-of-ai_2565x1854\/\" rel=\"attachment wp-att-405\"><img loading=\"lazy\" decoding=\"async\" class=\"size-full wp-image-405\" src=\"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-content\/uploads\/sites\/2495\/2025\/01\/ai-am-over-it_nadia-piet-aixdesign_archival-images-of-ai_2565x1854.png\" alt=\"A red-toned illustration shows a man's head surrounded by swirling AI icons, with small, mischievous witch-like figures flying around him. The man's expression appears disoriented and fatigued, symbolizing the mental overload caused by the overwhelming flood of AI tools and news. The witches represent the chaotic, cackling nature of rapid AI developments, adding to the sense of dizziness and confusion.\" width=\"1200\" height=\"867\" \/><\/a><figcaption id=\"caption-attachment-405\" class=\"wp-caption-text\"><a href=\"https:\/\/nadiapiet.com\">Nadia Piet<\/a> + <a href=\"https:\/\/aixdesign.co\/posts\/archival-images-of-ai\">AIxDESIGN &amp; Archival Images of AI<\/a> \/ <a href=\"https:\/\/www.betterimagesofai.org\">Better Images of AI<\/a> \/ AI Am Over It \/ <a href=\"https:\/\/creativecommons.org\/licenses\/by\/4.0\/\">CC-BY 4.0<\/a><\/figcaption><\/figure>\n<p>You should evaluate the quality and reliability of AI-generated content before relying on it, just as you would information from any source. Information provided by generative AI tools may be:<\/p>\n<ul>\n<li>incorrect<\/li>\n<li>out of date<\/li>\n<li>biased or offensive<\/li>\n<li>lacking common sense<\/li>\n<li>lacking originality.<\/li>\n<\/ul>\n<p>AI tools tend to produce &#8216;middle-of-the-road&#8217; answers, based on a consensus of the most common information in the AI&#8217;s training data. You should continue to think critically as you use the tools for your learning. Ask yourself:<\/p>\n<ul>\n<li>is the response you&#8217;ve been given too conservative?<\/li>\n<li>is there an alternative viewpoint that has been missed?<\/li>\n<li>what are your views \u2014 do you disagree with the information?<\/li>\n<\/ul>\n<h2>Methods for evaluating information<\/h2>\n<p>There are many methods for evaluating information. The <strong>TRAAP<\/strong> test is useful to consider when evaluating information generally and also emphasises some of the challenges with the evaluation of AI-generated content.<\/p>\n<div class=\"textbox\">\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone wp-image-529\" src=\"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-content\/uploads\/sites\/2495\/2023\/03\/alert-circle.png\" alt=\"important icon\" width=\"30\" height=\"30\" srcset=\"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-content\/uploads\/sites\/2495\/2023\/03\/alert-circle.png 96w, https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-content\/uploads\/sites\/2495\/2023\/03\/alert-circle-65x65.png 65w\" sizes=\"auto, (max-width: 30px) 100vw, 30px\" \/> <a id=\"traap\"><\/a>Applying the TRAAP test<\/p>\n<ul>\n<li>Timeliness<\/li>\n<li>Relevance<\/li>\n<li>Authority<\/li>\n<li>Accuracy<\/li>\n<li>Purpose<\/li>\n<\/ul>\n<\/div>\n<h2>Challenges<\/h2>\n<p>Different generative AI tools have different limitations. It is not always clear how <strong>current\u00a0<\/strong>information contained in LLMs is.\u00a0 <a href=\"https:\/\/openai.com\/index\/gpt-4-research\/\">OpenAI&#8217;s documentation<\/a> states that GPT-4 &#8220;generally lacks knowledge of events that have occurred after the vast majority of its data cuts off (September 2021)&#8221;. It is, however, capable of searching the web to find more recent information.<\/p>\n<p>LLMs may not always present you with the sources for answers, or rely on unsuitable sources. This can make it difficult to judge the <strong>relevance<\/strong>, <strong>authority<\/strong>, <strong>accuracy<\/strong> and <strong>purpose<\/strong> of the information.<\/p>\n<div class=\"textbox shaded\">\n<p>I can help with a wide range of topics, but there are some limitations. For example, I don\u2019t have access to:<\/p>\n<ul>\n<li><strong>Personal data<\/strong>\u00a0unless shared with me during our conversation.<\/li>\n<li><strong>Real-time data<\/strong>\u00a0like live sports scores or stock prices.<\/li>\n<li><strong>Confidential or proprietary information<\/strong>.<\/li>\n<li><strong>Certain copyrighted content<\/strong>\u00a0in full, such as books, articles, or songs.<\/li>\n<\/ul>\n<p><em>Source: Answer provided by Microsoft Copilot on 26 November 2024.<\/em><\/p>\n<\/div>\n<h2><a id=\"tips\"><\/a><a id=\"tips\"><\/a>Tips for confirming the information provided by AI tools<\/h2>\n<ol>\n<li>Ask the tool to provide you with <strong>sources<\/strong>. You can ask for a specific type of source (peer reviewed journal articles, news articles or academic sources). You can provide other constraints such as a time limit, e.g. &#8216;Can you provide academic sources from the last 5 years?&#8217;. Writing your prompt in academic or formal language will increase the chance of getting those types of sources. Note that there&#8217;s no guarantee that the AI tool will give you what you ask for but these techniques can increase the chance of better outcomes.<\/li>\n<li>Locate the sources provided and <strong>confirm the information is<\/strong>\u00a0<strong>real<\/strong>. Generative AI tools will <a href=\"https:\/\/en.wikipedia.org\/wiki\/Hallucination_(artificial_intelligence)\" target=\"_blank\" rel=\"noopener\">present false information as fact<\/a> and make up references.<\/li>\n<li>Once you confirm the sources, consider their <strong>quality<\/strong> and whether they are <strong>appropriate<\/strong> for your task.<\/li>\n<li>Look for <strong>other reputable sources<\/strong> that also confirm the information.<\/li>\n<\/ol>\n<div class=\"textbox shaded\">\n<blockquote><p>&#8220;Treat the AI like a slightly unreliable friend. Have a chat, ask some questions. Don\u2019t trust the answers though.&#8221;<\/p>\n<p><a href=\"\/\" rel=\"attachment wp-att-34\"><img loading=\"lazy\" decoding=\"async\" class=\"alignnone wp-image-34\" src=\"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-content\/uploads\/sites\/2495\/2025\/01\/book-open-bookmark.png\" alt=\"read icon\" width=\"30\" height=\"30\" \/><\/a> <a href=\"https:\/\/www.cilip.org.uk\/news\/661388\/Can-AI-do-your-reading-for-you--should-it.htm\" target=\"_blank\" rel=\"noopener\">Can AI do your reading for you and should it?<\/a><\/p><\/blockquote>\n<\/div>\n<h2>Human in the loop<\/h2>\n<p>Evaluating the outputs of AI tools is sometimes referred to as &#8220;human-in-the-loop&#8221; work. Many of the AI models are based on predictive modelling and contextual understanding of the prompts they\u2019re given. These models make mistakes!<\/p>\n<div class=\"textbox shaded\">\n<p><a href=\"\/\" rel=\"attachment wp-att-34\"><img loading=\"lazy\" decoding=\"async\" class=\"alignnone wp-image-34\" src=\"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-content\/uploads\/sites\/2495\/2025\/01\/book-open-bookmark.png\" alt=\"read icon\" width=\"30\" height=\"30\" \/><\/a> Users of a new Google AI feature were told to <a href=\"https:\/\/www.bbc.com\/news\/articles\/cd11gzejgz4o\">eat rocks and add glue to pizza.<\/a><\/p>\n<\/div>\n<p>Constant feedback by the human-in-the-loop can improve your specific output and also the AI tools and models &#8220;and enhance the accuracy, reliability, and adaptability of ML systems, harnessing the unique capabilities of both humans and machines&#8221; (Source: <a href=\"https:\/\/cloud.google.com\/discover\/human-in-the-loop#what-is-human-in-the-loop-hitl-in-ai-ml\">What is Human-in-the-Loop in AI &amp; ML?<\/a>).<\/p>\n<div class=\"textbox textbox--exercises\">\n<header class=\"textbox__header\">\n<h2 class=\"textbox__title\">Activity<\/h2>\n<\/header>\n<div class=\"textbox__content\">\n<p>Choose a topic you know well \u2014 this could be a hobby, sport, musical instrument, game, or any area where you have confidence in your knowledge.<\/p>\n<ol>\n<li>Ask an AI tool (such as ChatGPT, Bing Copilot, or Google Gemini) a question about this topic. For example, \u201cHow do you tune a guitar?\u201d or \u201cWhat are the offside rules in soccer?\u201d<\/li>\n<li>Carefully read the AI-generated response and evaluate it for accuracy, clarity, and completeness.<\/li>\n<li>Identify any inaccuracies, misleading explanations, or gaps in the information provided.<\/li>\n<\/ol>\n<p>Consider<\/p>\n<ul>\n<li>What did the AI tool get right?<\/li>\n<li>What did it miss or get wrong?<\/li>\n<li>Did knowing the topic well help you spot issues?<\/li>\n<li>Does this exercise shape your view on using AI tools for learning?<\/li>\n<\/ul>\n<h3>Example<\/h3>\n<p><strong>A student asked ChatGPT:<\/strong> \u201cWhat are the strings on a standard 6-string guitar tuned to?\u201d<\/p>\n<p><strong>AI Response:<\/strong> \u201cThe strings on a standard 6-string guitar are tuned to E-B-G-D-A-E, from the lowest (thickest) string to the highest (thinnest).\u201d<\/p>\n<p><strong>What\u2019s incorrect: <\/strong>This is a reversal. The correct tuning from the lowest (thickest) string to the highest (thinnest) is E-A-D-G-B-E. The AI listed the strings in reverse order, which could confuse a beginner.<\/p>\n<p><strong>Reflection: <\/strong>A student familiar with the guitar would immediately recognise the error, but someone new might unknowingly accept the incorrect answer.<\/p>\n<\/div>\n<\/div>\n<p>&nbsp;<\/p>\n<div class=\"media-attributions clear\" prefix:cc=\"http:\/\/creativecommons.org\/ns#\" prefix:dc=\"http:\/\/purl.org\/dc\/terms\/\"><h2>Media Attributions<\/h2><ul><li >alert-circle       <\/li><\/ul><\/div>","protected":false},"author":2509,"menu_order":8,"template":"","meta":{"pb_show_title":"on","pb_short_title":"","pb_subtitle":"","pb_authors":["uq-library-2"],"pb_section_license":"cc-by-nc"},"chapter-type":[],"contributor":[71],"license":[56],"class_list":["post-544","chapter","type-chapter","status-publish","hentry","contributor-uq-library-2","license-cc-by-nc"],"part":92,"_links":{"self":[{"href":"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-json\/pressbooks\/v2\/chapters\/544","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-json\/pressbooks\/v2\/chapters"}],"about":[{"href":"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-json\/wp\/v2\/types\/chapter"}],"author":[{"embeddable":true,"href":"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-json\/wp\/v2\/users\/2509"}],"version-history":[{"count":2,"href":"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-json\/pressbooks\/v2\/chapters\/544\/revisions"}],"predecessor-version":[{"id":1037,"href":"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-json\/pressbooks\/v2\/chapters\/544\/revisions\/1037"}],"part":[{"href":"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-json\/pressbooks\/v2\/parts\/92"}],"metadata":[{"href":"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-json\/pressbooks\/v2\/chapters\/544\/metadata\/"}],"wp:attachment":[{"href":"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-json\/wp\/v2\/media?parent=544"}],"wp:term":[{"taxonomy":"chapter-type","embeddable":true,"href":"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-json\/pressbooks\/v2\/chapter-type?post=544"},{"taxonomy":"contributor","embeddable":true,"href":"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-json\/wp\/v2\/contributor?post=544"},{"taxonomy":"license","embeddable":true,"href":"https:\/\/pressbooks.bccampus.ca\/introductiontoresearch\/wp-json\/wp\/v2\/license?post=544"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}