{"id":105656,"date":"2023-12-04T05:50:00","date_gmt":"2023-12-04T10:50:00","guid":{"rendered":"https:\/\/www.bu.edu\/law\/?post_type=bu-article&#038;p=105656"},"modified":"2023-12-04T08:57:10","modified_gmt":"2023-12-04T13:57:10","slug":"drawing-on-law-to-study-ai","status":"publish","type":"bu-article","link":"https:\/\/www.bu.edu\/law\/record\/articles\/2023\/drawing-on-law-to-study-ai\/","title":{"rendered":"Drawing on Law to Study AI"},"content":{"rendered":"\t<div class=\"wp-block-editorial-leadin record-block-editorial-leadin is-style-text-to-image has-media has-media-focus-center-middle has-secondary-theme\">\n\t\t<div class=\"container-lockup\">\n\n\t\t\t<div class=\"wp-block-leadin-media\">\n\t\t\t\t\n\t\t\t\t<img width=\"2500\" height=\"1875\" src=\"\/law\/files\/2023\/11\/Vasanth-Sarathy-RF23.jpg\" class=\"\" alt=\"Vasanth Sarathy (&#039;10) sits in a blue chair with a robot next to him\" loading=\"lazy\" srcset=\"https:\/\/www.bu.edu\/law\/files\/2023\/11\/Vasanth-Sarathy-RF23.jpg 2500w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/Vasanth-Sarathy-RF23-636x477.jpg 636w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/Vasanth-Sarathy-RF23-1024x768.jpg 1024w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/Vasanth-Sarathy-RF23-768x576.jpg 768w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/Vasanth-Sarathy-RF23-1536x1152.jpg 1536w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/Vasanth-Sarathy-RF23-2048x1536.jpg 2048w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/Vasanth-Sarathy-RF23-1101x826.jpg 1101w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/Vasanth-Sarathy-RF23-909x682.jpg 909w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/Vasanth-Sarathy-RF23-1376x1032.jpg 1376w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/Vasanth-Sarathy-RF23-1761x1321.jpg 1761w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/Vasanth-Sarathy-RF23-459x344.jpg 459w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/Vasanth-Sarathy-RF23-705x529.jpg 705w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/Vasanth-Sarathy-RF23-917x688.jpg 917w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/Vasanth-Sarathy-RF23-1411x1058.jpg 1411w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/Vasanth-Sarathy-RF23-1819x1364.jpg 1819w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/Vasanth-Sarathy-RF23-1333x1000.jpg 1333w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/Vasanth-Sarathy-RF23-1600x1200.jpg 1600w\" sizes=\"(max-width: 2500px) 100vw, 2500px\" \/>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\n\n\t\t\t\n\t\t\t\t<div class=\"container-words-outer\">\n\n\t\t\t\t\t\n\t\t\t\t\t<div class=\"container-words-inner\">\n\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"wp-prepress-tag\">Alumni<\/span>\n\t\t\t\t\t\t\t\t<h1 class=\"head\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\tDrawing on Law to Study AI\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/h1>\n\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\t\t\t\t<h4 class=\"deck\">Researcher Vasanth Sarathy (\u201910) is exploring how AI algorithms can be deployed to flag and counteract misinformation online, in part by identifying spurious arguments.<\/h4>\n\t\t\t\t\t\t\n\t\t\t\t\t<\/div>\n\n\t\t\t\t<\/div>\n\n\t\t\t\n\t\t<\/div>\n\n\t\t\t\t\t<p class=\"wp-block-editorial-leadin-caption wp-prepress-component-caption\">Photo by Ciara Crocker<\/p>\n\t\t\n\t<\/div>\n\n\t\n<div class=\"wp-prepress-component-metabar js-block-editorial-metabar record-prepress-layout-metabar\">\n\t<div class=\"wp-prepress-component-metabar-wrapper\">\n\t\t\t\t\t<div class=\"wp-prepress-component-metabar-date\">December 4, 2023<\/div>\n\t\t\n\t\t\t\t<div class=\"wp-prepress-component-metabar-credits\">\n\t\t\t\t\t\t\t\t<ul data-credit-type=\"By\">\n\t\t\t\t\t\t<li><a href=\"https:\/\/www.bu.edu\/law\/record\/authors\/june-d-bell\/\">June D. Bell<\/a><\/li>\n\t\t\t\t\t<\/ul>\n\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<div class=\"wp-prepress-component-metabar-share js-bu-prepress-share-tools\">\n\t\t\t<span class=\"icon-twitter\"><span>Twitter<\/span><\/span>\n\t\t\t<span class=\"icon-facebook\"><span>Facebook<\/span><\/span>\n\t\t\t<span class=\"icon-action\"><\/span>\n\t\t<\/div>\n\t<\/div>\n<\/div>\n\n\t\n\n\n<div class=\"wp-block-editorial-introparagraph record-block-editorial-introparagraph is-style-large has-paragraph-color-tertiary\"><div class=\"wp-block-editorial-introparagraph-content\"><p>AI isn\u2019t funny. At least, not yet.<\/p><\/div><\/div>\n\n\n\n<p>Although artificial intelligence is being tapped for ever more sophisticated tasks\u2014from drafting legal documents to debugging code\u2014it hasn\u2019t been able to figure out what exactly makes us humans chuckle.<\/p>\n\n\n\n<p>Is it even possible for AI to master wit? <a href=\"https:\/\/www.vsarathy.com\/\" target=\"_blank\" rel=\"noreferrer noopener\">Vasanth Sarathy<\/a> (\u201910)\u2014lawyer, artificial intelligence expert, cartoonist, and <a href=\"https:\/\/engineering.tufts.edu\/cs\/people\/faculty\/vasanth-sarathy\" target=\"_blank\" rel=\"noreferrer noopener\">Tufts University faculty member<\/a>\u2014can\u2019t say for sure. He\u2019s been tinkering with an AI tool to generate cartoon ideas, but progress, he admits, is slow. \u201cIt doesn\u2019t understand why something is funny,\u201d he says, \u201cwhich may be a very personal human thing, a consequence of life experiences. In which case, it\u2019s even harder for an AI system to replicate it.\u201d<\/p>\n\n\n\n<p>Even AI specialists like Sarathy, who holds a doctorate in computer science and cognitive science in addition to his JD, have been astounded by the sophistication and wildfire adaptation of generative technology. In just the past year, large language models such as OpenAI\u2019s GPT-4, which are trained on internet content to identify patterns and predict language, have been increasingly deployed to write essays, reply to patients\u2019 queries to their doctors, and even create artwork and music. Their popularity and rapid deployment\u2014in just two months after its launch in November 2022, ChatGPT had amassed more than 100 million users\u2014have heightened fears about the morals and ethics of this astonishingly powerful tool.<\/p>\n\n\n\n<p>Sarathy is more curious than anxious, perhaps because he sees AI\u2019s potential to solve problems it has been accused of creating. He is particularly interested in how AI can be leveraged to combat rampant disinformation. He\u2019s focusing on how AI\u2019s algorithms can be deployed to flag and counteract misinformation online, in part by identifying spurious arguments.<\/p>\n\n\n\n<p>Picking apart arguments is, not surprisingly, a skill at which lawyers excel, Sarathy says. The logical reasoning section of the LSAT is a perfect example: aspiring lawyers read excerpts and then pinpoint implicit assumptions and flaws in reasoning. This type of analytical thinking becomes second nature to litigators and appellate lawyers, but it\u2019s challenging for those who lack training or practice in thinking critically. This is the audience most likely to accept fake information and dubious arguments and then spread untruths.<\/p>\n\n\n\n<div class=\"wp-block-bu-pullquote record-block-bu-pullquote alignwide is-style-modern has-image-focus-center-middle has-tertiary-theme\"><div class=\"wp-block-bu-pullquote-inner\"><blockquote><div class=\"container-lockup\"><div class=\"container-icon-outer\"><div class=\"container-icon-inner\"><\/div><\/div><div class=\"container-text\"><hr\/><div class=\"quote-sizing\">Helping humans with critical thinking is something that machines can do, I think. We\u2019re sort of scratching the surface of that.<\/div><footer class=\"caption\">Vasanth Sarathy (&#8217;10)<\/footer><hr\/><\/div><\/div><\/blockquote><\/div><\/div>\n\n\n\n<p>\u201cWe don\u2019t have the capacity to quickly do critical thinking at the level we need to, to combat disinformation,\u201d Sarathy says. \u201cIf the language is fluent and the argument seems relatively good, we tend to believe it. That\u2019s really challenging because\u2026a little bit of critical thinking can go a long way\u2026. Helping humans with critical thinking is something that machines can do, I think. We\u2019re sort of scratching the surface of that.\u201d<\/p>\n\n\n\n<p>As a member of a multidisciplinary team that includes a social anthropologist, Sarathy has been exploring how to foster online communities that encourage healthy but respectful discourse while respecting cultural differences in speech and intent. \u201cYou have systems that can understand the language, but then you introduce social science theories and the extensive work that people have done in anthropology, studying different cultures\u2026and then you have the AI system produce responses that are more nuanced and more informed,\u201d he says. The project is in its earliest stages, undergoing extensive testing to see how effectively the AI system can generate speech that is relevant, useful, and accurate. Eventually, the group might partner with government agencies and social media companies.<\/p>\n\n\n\n<p>One of the biggest shortcomings of AI systems is that they lack a model of the world humans build over a lifetime. People spend years weaving a rich and vast contextual network of memories, knowledge, facts, education, and connections. When listening to or making an argument, we tap into a deep web of experience to formulate it, understand it, and evaluate its merits, Sarathy says. \u201cThere\u2019s rhetoric, there is understanding, forming mental models of the other person, and understanding what they know so that you don\u2019t just repeat what they [already] know. You\u2019re telling them something new but also building on what they know so that they believe you. There\u2019s a notion of trust and persuasiveness.\u201d<\/p>\n\n\n\n<p>This uniquely human ability to blend experience and context also helps pinpoint why AI can\u2019t quite nail humor. A single-panel cartoon\u2014the kind Sarathy draws and the <em>New Yorker<\/em> showcases\u2014appears to be a simple pairing of a sketch and a line of text. A clever cartoonist presents a familiar situation, such as a dinner party or a parent-teacher conference, with a caption that tweaks the typical scenario. That mismatch between expectation and \u201creality\u201d is the crux of humor, but the interplay between the art and the words matters, too.<\/p>\n\n\n\n<p>\u201cThat\u2019s where the human piece comes in,\u201d Sarathy says. \u201cIf your timing is off, if you wait too long, then [readers] are going to think of that situation and not find the joke funny. But if you do it too soon, they\u2019re not going to have enough time to form that first mental model of the situation. It\u2019s a beautiful dance to get it right. AI systems can replicate what\u2019s already out there in terms of captions and such, but they\u2019re not original enough yet. They can\u2019t come up with completely different ways of thinking or new ideas.\u201d Which means ChatGPT won\u2019t be winning the <em>New Yorker<\/em>\u2019s caption contest anytime soon.<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" width=\"1000\" height=\"1000\" src=\"\/law\/files\/2023\/11\/AI-Cartoon-Sarathy.jpg\" alt=\"A cartoon shows a long line of people lining up to enter paradise with an AI screen in front of the closed gates. The AI screen says: &quot;As an AI, I don't possess the authority to make ultimate decisions regarding your eternal salvation or damnation.&quot;\" class=\"wp-image-105654\" srcset=\"https:\/\/www.bu.edu\/law\/files\/2023\/11\/AI-Cartoon-Sarathy.jpg 1000w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/AI-Cartoon-Sarathy-636x636.jpg 636w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/AI-Cartoon-Sarathy-150x150.jpg 150w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/AI-Cartoon-Sarathy-768x768.jpg 768w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/AI-Cartoon-Sarathy-826x826.jpg 826w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/AI-Cartoon-Sarathy-682x682.jpg 682w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/AI-Cartoon-Sarathy-344x344.jpg 344w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/AI-Cartoon-Sarathy-529x529.jpg 529w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/AI-Cartoon-Sarathy-688x688.jpg 688w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/AI-Cartoon-Sarathy-500x500.jpg 500w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/AI-Cartoon-Sarathy-300x300.jpg 300w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/AI-Cartoon-Sarathy-600x600.jpg 600w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/AI-Cartoon-Sarathy-550x550.jpg 550w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/AI-Cartoon-Sarathy-710x710.jpg 710w, https:\/\/www.bu.edu\/law\/files\/2023\/11\/AI-Cartoon-Sarathy-100x100.jpg 100w\" sizes=\"(max-width: 1000px) 100vw, 1000px\" \/><figcaption>Cartoon by Vasanth Sarathy (&#8217;10)<\/figcaption><\/figure>\n\n\n\n<p>Sarathy draws on his experience as a cartoonist and lawyer in his AI research, a field that is by nature integrative. \u201cWe\u2019re working with humans, and so all the issues that we face with these AI systems are going to be inherently multidisciplinary,\u201d he says. \u201cWe\u2019re not just going to have computer scientists build AI tools and then put them out there. That\u2019s one of the things that I\u2019m excited about: I\u2019m able to bring my legal background and some of my social science background to this technical side of things and work on both those issues and bring people on these two sides together.\u201d<\/p>\n\n\n\n<p>Sarathy had no plans for a legal career when he studied electrical engineering on a full scholarship at the University of Arkansas or as he pursued a doctorate at MIT in the early 2000s. But while in graduate school, he learned that Boston-based law firm Ropes &amp; Gray needed engineers who could understand the complex technology behind clients\u2019 inventions. Intrigued, Sarathy made a career pivot. He registered to practice at the US Patent Office and took the patent prosecution exam, which does not require a law degree. Sarathy enrolled in BU Law five years later while working full time. \u201cI don\u2019t know how I did that,\u201d he admits, \u201cbut it was absolutely insane.\u201d<\/p>\n\n\n\n<p>While on the partner track, he represented Google, MIT, medical device makers, healthcare companies, and Apple, and he collaborated with litigators and advised on intellectual property, data security, and privacy issues. But Sarathy found himself pondering how, exactly, innovators work their magic, spinning an insight or pain point into a start-up or a patent-worthy invention. \u201cIt just became a thing I wanted to study: how the human mind works and how we humans are creative. How do we come up with new ideas? How do we invent things?\u201d Sarathy, then 35, headed to Tufts in 2015 to pursue a doctorate in computer science and cognitive science.<\/p>\n\n\n\n<p>He spent nearly three years at research firm <a href=\"https:\/\/www.sift.net\/\" target=\"_blank\" rel=\"noreferrer noopener\">SIFT<\/a> (Smart Information Flow Technologies), probing questions such as the <a href=\"https:\/\/dl.acm.org\/doi\/pdf\/10.1145\/3341166\" target=\"_blank\" rel=\"noreferrer noopener\">parameters of consent in human-robot interactions<\/a>. It\u2019s less far-fetched than it might seem. A robot waiter, for example, should be programmed to clear a diner\u2019s plate only when the diner has provided \u201cconsent cues,\u201d such as placing her utensils in the \u201cfinished\u201d position on her plate or sitting back from the table. As robots become more integrated into society, human social norms will provide implicit and explicit consent cues for interactions, Sarathy wrote in a <a href=\"https:\/\/dl.acm.org\/doi\/pdf\/10.1145\/3341166\" target=\"_blank\" rel=\"noreferrer noopener\">2019 research paper<\/a>.<\/p>\n\n\n\n<p>\u201cA lot of the consent work was based on my legal training,\u201d he says. \u201cI would not have been able to write that paper if it had not been for the fact that I went to law school and took classes in torts.\u201d He continues to tap his legal experience, crediting his years in law school and practicing law with sharpening his writing and thinking\u2014and giving him ample fodder for <em>Legally Drawn<\/em>, the (now-defunct) cartoon blog he launched as a BU Law student.<\/p>\n\n\n\n<p>Though the legal profession will increasingly outsource tasks such as document review to AI tools, Sarathy is confident that AI is a poor substitute for the experience, context, and nuanced analysis that trusted attorneys provide their clients. \u201cI\u2019m not saying that these systems are bad,\u201d he says. \u201cI use them in my work as well. I\u2019m just saying that there\u2019s not a risk that lawyers are going be out of their jobs anytime soon.\u201d<\/p>\n\n\n\n<p>Lawyers and ethicists will, however, inevitably tangle with AI\u2019s legal gray areas. Leveraging ChatGPT\u2019s open-source software, developers and start-ups have been building apps that let users craft college application essays and create images in the style of famous artists and celebrity designers. When the large language models produce new images after being trained on copyright-protected images, are those creations copyright violations?<\/p>\n\n\n\n<p class=\"is-style-end-of-article\">\u201cWe don\u2019t know where the law is going to end up on that,\u201d Sarathy says. \u201cIt\u2019s not clear what the law says about this situation. You\u2019re going to have to think about the legal consequences. What are the different ways that AI systems can be more\u2026transparent [and] aligned with human values?&#8230; The best-case scenario is that we humans get better in figuring out which AI tools are good and which ones are not, because they\u2019re all going to be there, and they\u2019re all being used.\u201d<\/p>\n","protected":false},"excerpt":{"rendered":"<p>AI isn\u2019t funny. At least, not yet. Although artificial intelligence is being tapped for ever more sophisticated tasks\u2014from drafting legal documents to debugging code\u2014it hasn\u2019t been able to figure out what exactly makes us humans chuckle.<\/p>\n","protected":false},"author":11260,"featured_media":105655,"comment_status":"closed","ping_status":"closed","template":"","meta":{"bu_prepress_billboard":"","_bu_prepress_primary_term":"Alumni","_bu_prepress_primary_term_manual":""},"tags":[1667],"bu-publication":[3742],"record-article-category":[3918,3744,3776,4135,3786,3775],"record-topic":[],"bu_edition":[],"media_type":[],"_links":{"self":[{"href":"https:\/\/www.bu.edu\/law\/wp-json\/wp\/v2\/bu-article\/105656"}],"collection":[{"href":"https:\/\/www.bu.edu\/law\/wp-json\/wp\/v2\/bu-article"}],"about":[{"href":"https:\/\/www.bu.edu\/law\/wp-json\/wp\/v2\/types\/bu-article"}],"author":[{"embeddable":true,"href":"https:\/\/www.bu.edu\/law\/wp-json\/wp\/v2\/users\/11260"}],"replies":[{"embeddable":true,"href":"https:\/\/www.bu.edu\/law\/wp-json\/wp\/v2\/comments?post=105656"}],"version-history":[{"count":6,"href":"https:\/\/www.bu.edu\/law\/wp-json\/wp\/v2\/bu-article\/105656\/revisions"}],"predecessor-version":[{"id":105663,"href":"https:\/\/www.bu.edu\/law\/wp-json\/wp\/v2\/bu-article\/105656\/revisions\/105663"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.bu.edu\/law\/wp-json\/wp\/v2\/media\/105655"}],"wp:attachment":[{"href":"https:\/\/www.bu.edu\/law\/wp-json\/wp\/v2\/media?parent=105656"}],"wp:term":[{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.bu.edu\/law\/wp-json\/wp\/v2\/tags?post=105656"},{"taxonomy":"bu-publication","embeddable":true,"href":"https:\/\/www.bu.edu\/law\/wp-json\/wp\/v2\/bu-publication?post=105656"},{"taxonomy":"record-article-category","embeddable":true,"href":"https:\/\/www.bu.edu\/law\/wp-json\/wp\/v2\/record-article-category?post=105656"},{"taxonomy":"record-topic","embeddable":true,"href":"https:\/\/www.bu.edu\/law\/wp-json\/wp\/v2\/record-topic?post=105656"},{"taxonomy":"bu_edition","embeddable":true,"href":"https:\/\/www.bu.edu\/law\/wp-json\/wp\/v2\/bu_edition?post=105656"},{"taxonomy":"media_type","embeddable":true,"href":"https:\/\/www.bu.edu\/law\/wp-json\/wp\/v2\/media_type?post=105656"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}