{"id":2075,"date":"2024-08-19T10:00:14","date_gmt":"2024-08-19T09:00:14","guid":{"rendered":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/?p=2075"},"modified":"2024-08-07T11:25:14","modified_gmt":"2024-08-07T10:25:14","slug":"2075-2","status":"publish","type":"post","link":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/2075-2\/","title":{"rendered":"AI voices should sound weird"},"content":{"rendered":"

There has been a lot of discussion recently \u2013 both among academics and in the popular press \u2013\u00a0 about what kinds of limits should be placed on content generated by AI. Although questions about how to implement them may be technically, legally, or politically tricky, there are very clear reasons for thinking some such limits are warranted. No one sensible wants to see more racist tirades like the one Microsoft\u2019s chatbot Tay went on within hours of its introduction, and we will all be worse off if bad actors can rely on AI to learn how to do things like manufacture drugs or explosives.<\/p>\n

My goal here will be to argue that in addition to limits on content, we should place restrictions on the forms that speech generated by AI can take. I will focus on the way such speech sounds \u2013 literally, what the voices it produces sound like \u2013 but similar considerations apply to signed and written language, too.<\/p>\n

Some of the terrain here is very straightforward. Consider a recent case that attracted international media attention. The actor Scarlett Johansson claims that despite her refusing them permission twice, OpenAI used her voice to implement their ChatGPT virtual assistant. The company deny the allegation, although on the day the assistant was released, OpenAI\u2019s CEO Sam Altman tweeted the word \u201cher\u201d, which many observers have interpreted as a not-very -subtle reference to the 2013 film Her, <\/em>starring Scarlett Johansson and telling the story of a man who falls in love with an AI whose most embodied feature is a bright and cheerful voice.<\/p>\n

To use a person\u2019s voice in this way, not only without their permission but indeed against their expressed wishes, is obviously unacceptable. But we might wonder \u2013 would things have been different if the actor had given her consent? What if the resemblance really had been a coincidence?<\/p>\n

I think the answer to these questions is `no\u2019. I don\u2019t think AI voices should sound like any <\/em>human voice, much less like that of a particular individual. At least until philosophers, linguists, and psychologists have had a chance to properly work through the sorts of issues I\u2019ll discuss here, AI voices ought to remain stuck in the \u2018uncanny valley\u2019 \u2013 close enough to human to make their speech intelligible, but far enough that they produce feelings of alienness in listeners instead of empathy.<\/p>\n

What might a voice from the uncanny valley sound like? The CUNY philosopher Daniel Harris recently pointed me towards the illustrative example of the character Data from the television show Star Trek: The Next Generation<\/em>. Although the example isn\u2019t perfect, the show\u2019s writers have taken several steps to make the fact that Data is an android immediately audible. For one thing, except for the occasional slip from the human actor, Data avoids contractions. Instead of \u201cI don\u2019t\u201d he says \u201cI do not\u201d, and instead of \u201cgoin\u2019\u201d he says \u201cgoing\u201d. For another \u2013 again, to the extent possible for an actor \u2013 Data\u2019s speech lacks many of the subtle sonic cues that mark human speakers\u2019 emotional inflections. To appreciate some of these, think of the way you might produce the greeting \u201cHow are you?\u201d in a way that would convey warmth, excitement, or frustration.<\/p>\n

The simplest reason to hobble AI voices with regard to their naturalness is that it would make the fact that they are AI voices readily apparent to human listeners. Transparency in this sense would serve several purposes. First, along the same lines as the warning labels that many have suggested should accompany images and content produced by AI, it would put people in a position from which they could make properly informed judgments about how much stock to put in what they hear. More like a watermark layered over an image than like a warning label or badge in the corner, the markers of artificiality would be present throughout an AI system\u2019s speech, which might increase their efficacy. Second, this kind of consistent audible signal would address a worry about the role AI voices might otherwise play in undermining epistemic networks. If I don\u2019t know which voices are human and which AI, and I don\u2019t trust AI, I might respond by systematically lowering my trust in whatever I hear. Recent work on fragmentation and polarization suggests that this would be a bad outcome.<\/p>\n

There\u2019s a second reason for restricting AI voices to an unnatural range that is related to but distinct from the first. By making AI voices flat affect and clearly robotic, we might be able to reduce the risk of a powerful set of psychological and emotional tools being used to sidestep people\u2019s rational engagement with the messages AI speech presents.<\/p>\n

Many years\u2019 worth of research in linguistics and in psychology has demonstrated that our impressions about the credibility of the people we talk to and the things they say, as well as our emotional responses to them, depend a lot on the way they sound. For example, people who speak with certain accents are judged to be more reliable sources of information than others, and there are patterns of pronunciation that systematically produce the impression that a speaker is friendly, the kind of person who has your best interests at heart.<\/p>\n

It\u2019s a totally normal feature of human life that we use these facts to navigate the social world. In a job interview, I might speak one way in order to create a certain impression, and at a party with dear friends, another. When I\u2019m angry I sound one way, and when I\u2019m asking for a favor, another still. Allowing artificial systems to reproduce the full range of human variation in these dimensions, however, may cause more harm than good. While there are clear reasons to do what we can to make a pilot more likely to listen when the computer says `collision warning!\u2019, a world in which advertisers can speak to you in precisely the way their data suggests will be more likely to get you to make decisions that go against your interests seems like a bad one. So much the worse if the same goes for political messaging or legal advice.<\/p>\n

Photo: David Underland on Unsplash<\/a><\/p>\n

 <\/p>\n

 <\/p>\n

 <\/p>\n

 <\/p>\n

 <\/p>\n

 <\/p>\n

 <\/p>\n

 <\/p>\n

 <\/p>\n","protected":false},"excerpt":{"rendered":"There has been a lot of discussion recently \u2013 both among academics and in the popular press \u2013\u00a0 about what kinds of limits should be placed on content generated by […]","protected":false},"author":1319,"featured_media":2077,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_feature_clip_id":0,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_publicize_message":"Ethan Nowak \"AI voices should sound weird\" on Open for Debate","jetpack_publicize_feature_enabled":true,"jetpack_social_post_already_shared":true,"jetpack_social_options":{"image_generator_settings":{"template":"highway","default_image_id":0,"font":"","enabled":false},"version":2},"jetpack_post_was_ever_published":false},"categories":[19],"tags":[379,462],"custom_author":[460],"class_list":["post-2075","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-open-for-debate","tag-ai","tag-voice","custom_author-nowak-ethan"],"jetpack_publicize_connections":[],"jetpack_sharing_enabled":true,"jetpack_shortlink":"https:\/\/wp.me\/p8KPlM-xt","jetpack-related-posts":[{"id":2161,"url":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/philosophical-foundations-for-chatbot-regulation\/","url_meta":{"origin":2075,"position":0},"title":"Philosophical Foundations for Chatbot Regulation","author":"Alessandra Tanesini","date":"December 9, 2024","format":false,"excerpt":"In an age where we increasingly converse with artifacts - whose interfaces include AI, a pressing question emerges: Who, or what, are we really talking with? As AI-based chatbots, often termed \u2018conversational Ais\u2019, become more sophisticated, the line between human and machine communication blurs, raising profound questions about the nature\u2026","rel":"","context":"In "Open for Debate"","block_context":{"text":"Open for Debate","link":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/category\/open-for-debate\/"},"img":{"alt_text":"","src":"https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/12\/A-realistic-black-and-white-photograph-of-a-personstanding-in-front-of-a-full-length-mirror-speaking-with-a-robot-reflection.jpg?resize=350%2C200&ssl=1","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/12\/A-realistic-black-and-white-photograph-of-a-personstanding-in-front-of-a-full-length-mirror-speaking-with-a-robot-reflection.jpg?resize=350%2C200&ssl=1 1x, https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/12\/A-realistic-black-and-white-photograph-of-a-personstanding-in-front-of-a-full-length-mirror-speaking-with-a-robot-reflection.jpg?resize=525%2C300&ssl=1 1.5x, https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/12\/A-realistic-black-and-white-photograph-of-a-personstanding-in-front-of-a-full-length-mirror-speaking-with-a-robot-reflection.jpg?resize=700%2C400&ssl=1 2x"},"classes":[]},{"id":2040,"url":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/truth-fiction-and-llms\/","url_meta":{"origin":2075,"position":1},"title":"Truth, fiction and LLMs","author":"Alessandra Tanesini","date":"June 10, 2024","format":false,"excerpt":"The trope of the working novelist isolated for months at a time, hunched over a desk with yellow stained and repetitively strained fingers clacking away on a keyboard is no more. Today\u2019s author needs only a few spare hours, a title, a vague plot, an internet connection and a subscription\u2026","rel":"","context":"In "Open for Debate"","block_context":{"text":"Open for Debate","link":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/category\/open-for-debate\/"},"img":{"alt_text":"","src":"https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/06\/janko-ferlic-sfL_QOnmy00-unsplash-scaled.jpg?resize=350%2C200&ssl=1","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/06\/janko-ferlic-sfL_QOnmy00-unsplash-scaled.jpg?resize=350%2C200&ssl=1 1x, https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/06\/janko-ferlic-sfL_QOnmy00-unsplash-scaled.jpg?resize=525%2C300&ssl=1 1.5x, https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/06\/janko-ferlic-sfL_QOnmy00-unsplash-scaled.jpg?resize=700%2C400&ssl=1 2x, https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/06\/janko-ferlic-sfL_QOnmy00-unsplash-scaled.jpg?resize=1050%2C600&ssl=1 3x, https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/06\/janko-ferlic-sfL_QOnmy00-unsplash-scaled.jpg?resize=1400%2C800&ssl=1 4x"},"classes":[]},{"id":1922,"url":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/the-dark-side-of-digital-influence\/","url_meta":{"origin":2075,"position":2},"title":"The Dark Side of Digital Influence","author":"Alessandra Tanesini","date":"March 18, 2024","format":false,"excerpt":"In the previous post, I outlined three reasons for paying closer attention to social influence in the digital landscape: the proliferation of social influence, the informational empowerment of social influence, and the fact that AI increasingly mediates social influence. In this post, I will argue that this not only perpetuates\u2026","rel":"","context":"In "Open for Debate"","block_context":{"text":"Open for Debate","link":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/category\/open-for-debate\/"},"img":{"alt_text":"","src":"https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/03\/man-6533390_1280.png?resize=350%2C200&ssl=1","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/03\/man-6533390_1280.png?resize=350%2C200&ssl=1 1x, https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/03\/man-6533390_1280.png?resize=525%2C300&ssl=1 1.5x, https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/03\/man-6533390_1280.png?resize=700%2C400&ssl=1 2x, https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/03\/man-6533390_1280.png?resize=1050%2C600&ssl=1 3x"},"classes":[]},{"id":1647,"url":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/leave-it-to-the-machines-re-evaluating-the-kasparov-reply\/","url_meta":{"origin":2075,"position":3},"title":"Leave it to the Machines? Re-evaluating the Kasparov reply","author":"Alessandra Tanesini","date":"March 6, 2023","format":false,"excerpt":"In 1997, Garry Kasparov became the first world chess champion to lose a match to a computer, IBM\u2019s Deep Blue. Kasparov initially thought the IBM team cheated after the computer played what GM Yasser Seirawan described as a \u2018human like move\u2019 that \u2018sent Garry into a tizzy\u2019. As Kasparov later\u2026","rel":"","context":"In "Open for Debate"","block_context":{"text":"Open for Debate","link":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/category\/open-for-debate\/"},"img":{"alt_text":"A man walks away from a chess match","src":"https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2023\/03\/Kasparov.png?resize=350%2C200&ssl=1","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2023\/03\/Kasparov.png?resize=350%2C200&ssl=1 1x, https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2023\/03\/Kasparov.png?resize=525%2C300&ssl=1 1.5x"},"classes":[]},{"id":1920,"url":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/unpacking-manipulation-for-the-digital-age\/","url_meta":{"origin":2075,"position":4},"title":"Unpacking Manipulation for the Digital Age","author":"Alessandra Tanesini","date":"March 18, 2024","format":false,"excerpt":"Public debate is shaped partly by human social influence, and we routinely distinguish different types of social influence, such as persuasion, coercion, and manipulation. While persuasion and coercion are reasonably well understood, philosophers have only recently begun to study the nature and ethics of manipulation. More attention to manipulation is\u2026","rel":"","context":"In "Open for Debate"","block_context":{"text":"Open for Debate","link":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/category\/open-for-debate\/"},"img":{"alt_text":"","src":"https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/03\/man-220975_1280.jpg?resize=350%2C200&ssl=1","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/03\/man-220975_1280.jpg?resize=350%2C200&ssl=1 1x, https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/03\/man-220975_1280.jpg?resize=525%2C300&ssl=1 1.5x, https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/03\/man-220975_1280.jpg?resize=700%2C400&ssl=1 2x, https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/03\/man-220975_1280.jpg?resize=1050%2C600&ssl=1 3x"},"classes":[]},{"id":2144,"url":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/ai-regulation-and-the-right-to-meaningful-explanation-pt-1-why-not\/","url_meta":{"origin":2075,"position":5},"title":"AI regulation and the right to meaningful explanation. Pt 1. Why (not)?","author":"Alessandra Tanesini","date":"November 11, 2024","format":false,"excerpt":"Ask anyone except a gun rights activist, and they will agree that the rise of new technologies requires the implementation of new regulations. Not too long ago, a law against human cloning would have sounded ridiculous. Since we are now actually able to clone human embryos,[1] we should expect legislators\u2026","rel":"","context":"In "Open for Debate"","block_context":{"text":"Open for Debate","link":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/category\/open-for-debate\/"},"img":{"alt_text":"","src":"https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/11\/Vaassem-Image-post-1.jpg?resize=350%2C200&ssl=1","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/11\/Vaassem-Image-post-1.jpg?resize=350%2C200&ssl=1 1x, https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/11\/Vaassem-Image-post-1.jpg?resize=525%2C300&ssl=1 1.5x, https:\/\/i0.wp.com\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/11\/Vaassem-Image-post-1.jpg?resize=700%2C400&ssl=1 2x"},"classes":[]}],"meta_box":[],"jetpack_featured_media_url":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/wp-content\/uploads\/sites\/533\/2024\/08\/david-underland-doEEsAb890c-unsplash-scaled.jpg","_links":{"self":[{"href":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/wp-json\/wp\/v2\/posts\/2075","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/wp-json\/wp\/v2\/users\/1319"}],"replies":[{"embeddable":true,"href":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/wp-json\/wp\/v2\/comments?post=2075"}],"version-history":[{"count":4,"href":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/wp-json\/wp\/v2\/posts\/2075\/revisions"}],"predecessor-version":[{"id":2088,"href":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/wp-json\/wp\/v2\/posts\/2075\/revisions\/2088"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/wp-json\/wp\/v2\/media\/2077"}],"wp:attachment":[{"href":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/wp-json\/wp\/v2\/media?parent=2075"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/wp-json\/wp\/v2\/categories?post=2075"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/wp-json\/wp\/v2\/tags?post=2075"},{"taxonomy":"custom_author","embeddable":true,"href":"https:\/\/blogs.cardiff.ac.uk\/openfordebate\/wp-json\/wp\/v2\/custom_author?post=2075"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}