{"id":977,"date":"2026-09-01T10:36:00","date_gmt":"2026-09-01T08:36:00","guid":{"rendered":"https:\/\/fromdatatoimpact.com\/?p=977"},"modified":"2026-04-20T10:40:14","modified_gmt":"2026-04-20T08:40:14","slug":"leadership-in-discourse-2","status":"publish","type":"post","link":"https:\/\/fromdatatoimpact.com\/index.php\/2026\/09\/01\/leadership-in-discourse-2\/","title":{"rendered":"Leadership in Discourse"},"content":{"rendered":"\n

by Stefan Kluge<\/a><\/strong><\/p>\n\n\n\n

Part 2 of 2: When human expertise meets generative AI<\/h2>\n\n\n\n

For the second essay of my dissertation (see Part 1<\/a>), we wanted to bring the same underlying question about leadership and semantic alignment of two streams of discourse <\/strong>into the present: when a new discourse stream appears, does it replace the old one, or does it settle into a different role? This time the setting was not parliaments and print advertising, but Stack Overflow and generative AI. In both cases, we are interested in who gets to shape meaning and problem-solving when actors with very different speeds enter the same space. The human expert used to dominate digital knowledge platforms. Now algorithmic answers arrive instantly, fluently, and at scale. The obvious fear is substitution. The more interesting question is whether that is actually what users want.<\/p>\n\n\n\n

So we studied a very large Stack Overflow dataset: more than 534,000 questions, 1.29 million answers, and over 7.2 million voting decisions. For each question, we also generated a ChatGPT answer and compared human answers to that synthetic benchmark. The central distinction in the essay is between more intellective problems, where users assume there is one demonstrably correct solution, and less intellective problems, where several reasonable solutions can coexist and judgment matters more. That difference turns out to be crucial. The essay tests two competing ideas: that users will prefer AI-like answers when they seek something like ground truth, or that AI-like answers become more useful when they contribute an alternative perspective. Our evidence supports the second view.<\/p>\n\n\n

\n
\"\"<\/figure><\/div>\n\n\n

This figures helps to see that this is not just a vague \u201cAI changes everything\u201d story. Intellectiveness becomes one of the key variables separating where AI-like answers are welcomed and where they are not.<\/strong> The results point away from the idea that users broadly reward AI-like answers for more factual, intellective problems. Instead, AI-like answers become more attractive when the problem is less intellective, more judgment-based, and when some time has already passed. In other words, the synthetic stream often seems to gain traction not by winning a race to the first answer, but by arriving later with a different angle once the human discussion has matured.<\/p>\n\n\n\n

That leaves us with a conclusion that is more hopeful and more interesting than a simple replacement story. On digital Q&A platforms, generative AI does not look like a universal truth machine that sweeps human expertise aside. It looks more like a complementary form of synthetic authority: sometimes useful, sometimes not, and especially promising when users need perspective rather than a single \u201ccorrect\u201d answer<\/strong>. That is also why this second essay speaks back to the first. The first essay<\/a> shows what happens when one discourse stream begins to outpace another over decades. The second suggests that in the human-AI case, we may instead be watching a new division of labor emerge. The real question is not whether humans or AI win outright. It is what kind of equilibrium they create together.<\/p>\n","protected":false},"excerpt":{"rendered":"

by Stefan Kluge Part 2 of 2: When human expertise meets generative AI For the<\/p>\n","protected":false},"author":1,"featured_media":980,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[49,4],"tags":[90,94],"coauthors":[16],"class_list":["post-977","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-artificial-intelligence","category-bigdata","tag-genai","tag-information-systems"],"jetpack_featured_media_url":"https:\/\/fromdatatoimpact.com\/wp-content\/uploads\/2026\/04\/ChatGPT-Image-Apr-20-2026-10_29_49-AM.png","_links":{"self":[{"href":"https:\/\/fromdatatoimpact.com\/index.php\/wp-json\/wp\/v2\/posts\/977","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/fromdatatoimpact.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/fromdatatoimpact.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/fromdatatoimpact.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/fromdatatoimpact.com\/index.php\/wp-json\/wp\/v2\/comments?post=977"}],"version-history":[{"count":5,"href":"https:\/\/fromdatatoimpact.com\/index.php\/wp-json\/wp\/v2\/posts\/977\/revisions"}],"predecessor-version":[{"id":1004,"href":"https:\/\/fromdatatoimpact.com\/index.php\/wp-json\/wp\/v2\/posts\/977\/revisions\/1004"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/fromdatatoimpact.com\/index.php\/wp-json\/wp\/v2\/media\/980"}],"wp:attachment":[{"href":"https:\/\/fromdatatoimpact.com\/index.php\/wp-json\/wp\/v2\/media?parent=977"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/fromdatatoimpact.com\/index.php\/wp-json\/wp\/v2\/categories?post=977"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/fromdatatoimpact.com\/index.php\/wp-json\/wp\/v2\/tags?post=977"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/fromdatatoimpact.com\/index.php\/wp-json\/wp\/v2\/coauthors?post=977"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}