Artificial Intelligence & Human Values Lecture Notes PDF

https://vis.ethz.ch/events/828 Artificial Intelligence & Human Values Lesson 9: Misinformation and Manipulation Fall 2024, ETH Zürich Dr. Gabriel Dorthe  Animating questions When and how did misinformation and manipulation become an issue? How...

https://vis.ethz.ch/events/828 Artificial Intelligence & Human Values Lesson 9: Misinformation and Manipulation Fall 2024, ETH Zürich Dr. Gabriel Dorthe  Animating questions When and how did misinformation and manipulation become an issue? How is this concern linked to the history of information and communication technologies? And what's specific to our present moment with AI? How are different actors (governments, tech leaders, publics) responding to misinformation and manipulation today? What else could be done? Bill Bramhill, New York Daily News, August 18, 2021  Manipulation of Causes information https://medium.com/better-humans/cognitive-bias-cheat-sheet-55a472476b18  November 19, 2018 www.theatlantic.com/science/archive/2018/11/psychologys-replication-crisis-real/576223 www.nytimes.com/interactive/2024/10/29/us/politics/trump-truth-social-conspiracy-theories.html www.washingtonpost.com/politics/interactive/2021/timeline-trump-claims-as-president www.washingtonpost.com/graphics/politics/trump-claims-database/?itid=lk_interstitial_manual_9 www.nytimes.com/interactive/2024/10/25/opinion/what-trump-says.html  Usual framings of the problem of trust It’s the science – but only because “abused” – If the facts are not being produced right, it is because of corruption, capture, or manufactured uncertainty It’s the people – If facts are (produced) right, then the problem (de cit) is in the (ignorant) recipient, the (inattentive) audience, or the (mis)communication It’s the institutions and relationships – If the breakdown is in trust (uptake, reception), then the problem is in the relations between science and society – and the state fi Les Moëres. Photo © Jelle Vermeersch 2013 2014 2019 English translation of the rst chapter, “Les Moëres” fi Les Moëres. Photo © Jelle Vermeersch «The Carnival» Marije Meerman (real.), Money & Speed - Inside the Black Box, VPRO, 2011  “There is no air traf c control for the market” Gregg Berman (SEC) Marije Meerman (real.), Money & Speed - Inside the Black Box, VPRO, 2011 fi  Usual framings of the problem of trust It’s the science – but only because “abused” – If the facts are not being produced right, it is because of corruption, capture, or manufactured uncertainty It’s the people – If facts are (produced) right, then the problem (de cit) is in the (ignorant) recipient, the (inattentive) audience, or the (mis)communication It’s the institutions and relationships – If the breakdown is in trust (uptake, reception), then the problem is in the relations between science and society – and the state fi Manipulation is not a bug but a de ning feature of information technologies. We are all affected by and affecting this issue We lack proper equipment to take this issue seriously fi  FAKE NEWS AND THE PUBLIC HOW THE PRESS COMBATS RUMOR. THE MARKET RIGGER, AND THE PROPAGANDIST Harper’s Magazine, October 1st, 1925 BY EDWARD McKERNON Superintendent oj the EM/ern Division oj The Associated Press T HE rush of the day seemed to have rushing wireless inquiries to the com- ended in the New York officeof The manders. The wire men were connect- Associated Press on the afternoon ing the news room with North Sydney, of June 8, 1921. The wires hummed mo- Nova Scotia, by direct cable in order to notonously with Wall Street's closing pick up any wireless messages intercepted prices, baseball scores, the price of pork, there. Ten minutes passed. They to-morrow's weather. In about twenty minutes the final editions in the East would go to press. The night shift was worked fast. But rumor worked faster. It had shipping offices by the ears. Telephone calls choked the switchboard, “misinformation is still circulated in the guise of news, and the nearly due. The Day Editor, that auto- "What about the Rochambeau?" "We crat of the news room, looked at his hear the Rochambeau struck an iceberg." watch and yawned. "Have you any word of a collision at Suddenly out of the air flashed an sea?" To all went the same answer: S 0 S followed by "Struck an iceberg off Newfoundland. Leaking." There was no signature, the name of the vessel " We do not know of any disaster." The Day Editor took out his watch. He knew some papers were going to dishonest reporter operating on in distress having been lost somewhere as her cry for help was picked up by the British liner Orduna and relayed to the press with the rumor. If it were true, it was news of the first magnitude. The A. P. might be licked. But it was the a big scale is a greater menace to wireless station at the Charlestown Navy same editor who had released the first Yard. Now this message in itself might have caused no great excitement except for the coincidence that during the pre- news of the Titanic horror and now the specter of that tragedy of the sea rose before his eyes. The peace of mind of society than ever before” ceding night the French line steamer many was in his keeping. "We'll wait," Rochambeau had wirelessed the offices of the line in New York that she had "sighted ice off the North Atlantic he said. Twelve minutes of suspense- thirteen-fourteen-fifteen- A flash from the cable: "Rochambeau reports all (534). coast." well." The Day Editor put up his The Rochambeau was known to be in watch and took out his cigarettes. the general locality from which the Any unimaginative reader with a pas- S 0 S had come. Instantly the two mes- sion for details is referred to the files sages were associated. As quickly mem- of the New York morning papers of bers of the Associated Press staff were June 9, 1921, where among the day's communicating with the offices of the news may be found an inconspicuous item French line. Others were locating by to the effect that the British freighter sea charts and shipping reports every Seapool, lumbering along the Newfound- large vessel in the iceberg zone and land coast, bumped an iceberg and took “We’re not just ghting an epidemic; we’re ghting an infodemic. Fake news spreads faster and more easily than this virus, and is just as dangerous. […] Now more than ever is the time for us to let science and evidence lead policy. If we don’t, we are headed down a dark path that leads nowhere but division and disharmony.” February 15, 2020 (photo Abdulhamid Hoşbaş/Anadolu Agency) fi fi https://misinforeview.hks.harvard.edu/article/misinformation-reloaded-fears-about-the-impact-of-generative-ai-on-misinformation-are-overblown/  Thanks to Paulo Fonseca! HEGEMONIC ETHICAL CONCERNS ON AI AND DISINFORMATION (SIMON ET AL, 2023) Increased Quantity of Misinformation The accessibility and usability of generative AI enable the creation of misinformation on a large scale at minimal cost for individuals and organized actors. An overflow of incorrect or misleading information could overshadow factual content and/or create confusion. Increased Quality of Misinformation More plausible and harder to debunk misinformation. Higher-quality misinformation is more persuasive, potentially leading to widespread false information and a general mistrust in all news sources. Personalization of Misinformation Generative AI can tailor misinformation to individual tastes and preferences, enhancing its impact and acceptance Personalized misinformation becomes more relevant and appealing to individuals, increasing its effectiveness and spread Involuntary Generation of Plausible but False Information generative AI can generate plausible yet entirely inaccurate information unintentionally. Users may inadvertently generate and spread misinformation. www.bbc.com/news/av/stories-52118949  Manipulation of Causes Responses information Lots of techniques well identi ed and strategies (but not helpful) (don’t work) fi  Manipulation of Causes Responses Lots of techniques well identi ed and strategies (but not helpful) information (don’t work) fi  Scienti cally accurate atomic model of the external structure of the Severe Acute Respiratory Syndrome CoronaVirus 2 (SARS-CoV-2) Source Wikipedia fi Nature Medicine, 17 March 2020 www.theguardian.com/environment/2020/mar/18/tip-of-the-iceberg-is-our-destruction-of-nature-responsible-for-covid-19-aoe Photo Hector Retamal/AFP https://edition.cnn.com/2020/04/13/us/coronavirus-made-in-lab-poll-trnd/index.html Virologists work in the P4 lab of the Wuhan Institute of Virology in central China’s Hubei province, February 23, 2017 © Getty Images (source: www.gisreportsonline.com/r/coronavirus-pandemic-biolab) 26.02.2023 21.06.2023 https://about.fb.com/news/2020/04/covid-19-misinfo-update Remember the dominant framing of misinformation? Quantity and scale -> How is it possible to decide what is relevant in such an immense amount of information? Quality and plausibility -> Scienti c journals were not able to process the new knowledge produced through regular peer-review Personalisation and relevance -> How is relevance negotiated? Involuntary generation and spread of false information -> What does it mean and for whom in context of high uncertainty? fi  Key Takeaways Misinformation and manipulation are de ning features of information technologies. Information is never just “information” (what gives a form to something), but is something that takes form in the process of human (and human- machines, and human-animals) interactions. Dominant framing of misinformation sees the world as at and in need of being polished or disciplined. It doesn’t understand that meaning is made in various ways. fi fl  What have we learned so far? In-class group review exercise: Inventory of Key takeaways and examples of AIxHV