{"499521":{"#nid":"499521","#data":{"type":"news","title":"Using Stories to Teach Human Values to Artificial Agents","body":[{"value":"\u003Cp\u003E\u003Cstrong\u003EATLANTA \u003C\/strong\u003E\u2014\u003Cstrong\u003E Feb. 12, 2016 \u003C\/strong\u003E\u2014 The rapid pace of artificial intelligence (AI) has raised fears about whether robots could act unethically or soon choose to harm humans. Some are calling for bans on robotics research; others are calling for more research to understand how AI might be constrained. But how can robots learn ethical behavior if there is no \u201cuser manual\u201d for being human?\u003C\/p\u003E\u003Cp\u003EResearchers \u003Ca href=\u0022https:\/\/research.cc.gatech.edu\/inc\/mark-riedl\u0022\u003E\u003Cstrong\u003EMark Riedl \u003C\/strong\u003E\u003C\/a\u003Eand \u003Cstrong\u003E\u003Ca href=\u0022http:\/\/www.brenteharrison.com\/\u0022\u003EBrent Harrison\u003C\/a\u003E\u003C\/strong\u003E from the School of Interactive Computing at the Georgia Institute of Technology believe the answer lies in \u201cQuixote\u201d \u2013 to be unveiled at the \u003Ca href=\u0022http:\/\/www.aaai.org\/Conferences\/AAAI\/aaai16.php\u0022\u003EAAAI-16 Conference\u003C\/a\u003E in Phoenix, Ariz. (Feb. 12 \u2013 17). Quixote teaches \u201cvalue alignment\u201d to robots by training them to read stories, learn acceptable sequences of events and understand successful ways to behave in human societies.\u003C\/p\u003E\u003Cp\u003E\u201cThe collected stories of different cultures teach children how to behave in socially acceptable ways with examples of proper and improper behavior in fables, novels and other literature,\u201d says Riedl, associate professor and director of the Entertainment Intelligence Lab. \u201cWe believe story comprehension in robots can eliminate psychotic-appearing behavior and reinforce choices that won\u2019t harm humans and still achieve the intended purpose.\u201d\u003C\/p\u003E\u003Cp\u003EQuixote is a technique for aligning an AI\u2019s goals with human values by placing rewards on socially appropriate behavior. It builds upon Riedl\u2019s prior research \u2013 the \u003Ca href=\u0022http:\/\/www.news.gatech.edu\/2015\/09\/01\/georgia-tech-uses-artificial-intelligence-crowdsource-interactive-fiction\u0022\u003EScheherazade system\u003C\/a\u003E \u2013 which demonstrated how artificial intelligence can gather a correct sequence of actions by crowdsourcing story plots from the Internet.\u003C\/p\u003E\u003Cp\u003EScheherazade learns what is a normal or \u201ccorrect\u201d plot graph. It then passes that data structure along to Quixote, which converts it into a \u201creward signal\u201d that reinforces certain behaviors and punishes other behaviors during trial-and-error learning. In essence, Quixote learns that it will be rewarded whenever it acts like the protagonist in a story instead of randomly or like the antagonist.\u003C\/p\u003E\u003Cp\u003EFor example, if a robot is tasked with picking up a prescription for a human as quickly as possible, the robot could a) rob the pharmacy, take the medicine, and run; b) interact politely with the pharmacists, or c) wait in line. Without value alignment and positive reinforcement, the robot would learn that robbing is the fastest and cheapest way to accomplish its task. With value alignment from Quixote, the robot would be rewarded for waiting patiently in line and paying for the prescription.\u003C\/p\u003E\u003Cp\u003ERiedl and Harrison demonstrate in their research how a value-aligned reward signal can be produced to uncover all possible steps in a given scenario, map them into a plot trajectory tree, which is then used by the robotic agent to make \u201cplot choices\u201d (akin to what humans might remember as a Choose-Your-Own-Adventure novel) and receive rewards or punishments based on its choice.\u003C\/p\u003E\u003Cp\u003EThe Quixote technique is best for robots that have a limited purpose but need to interact with humans to achieve it, and it is a primitive first step toward general moral reasoning in AI, Riedl says.\u003C\/p\u003E\u003Cp\u003E\u201cWe believe that AI has to be enculturated to adopt the values of a particular society, and in doing so, it will strive to avoid unacceptable behavior,\u201d he adds. \u201cGiving robots the ability to read and understand our stories may be the most expedient means in the absence of a human user manual.\u201d\u003C\/p\u003E\u003Cp\u003E\u003Cstrong\u003E\u003Ca href=\u0022http:\/\/www.cc.gatech.edu\/%7Eriedl\/pubs\/aaai-ethics16.pdf\u0022\u003EDownload\u003C\/a\u003E the complete research paper.\u003C\/strong\u003E\u003C\/p\u003E\u003Cp\u003E\u003Cem\u003EThis project undertaken was or is sponsored by the U.S. Defense Advanced Research Projects Agency (DARPA) under grant #D11AP00270 and the Office of Naval Research (ONR) under grant #N00014-14-1-0003. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the view of DARPA or the ONR.\u003Cbr \/\u003E\u003C\/em\u003E\u003C\/p\u003E","summary":null,"format":"limited_html"}],"field_subtitle":"","field_summary":[{"value":"\u003Cp\u003EThe rapid pace of artificial intelligence (AI) has raised fears about whether robots could act unethically to harm humans. But how can robots learn ethical behavior if there is no \u201cuser manual\u201d for being human? Researchers \u003Ca href=\u0022https:\/\/research.cc.gatech.edu\/inc\/mark-riedl\u0022\u003E\u003Cstrong\u003EMark Riedl \u003C\/strong\u003E\u003C\/a\u003Eand \u003Cstrong\u003E\u003Ca href=\u0022http:\/\/www.brenteharrison.com\/\u0022\u003EBrent Harrison\u003C\/a\u003E\u003C\/strong\u003E from the School of Interactive Computing believe the answer lies in \u201cQuixote\u201d \u2013 unveiled at the \u003Ca href=\u0022http:\/\/www.aaai.org\/Conferences\/AAAI\/aaai16.php\u0022\u003EAAAI-16 Conference\u003C\/a\u003E in Phoenix, Ariz. (Feb. 12 \u2013 17).\u003C\/p\u003E","format":"limited_html"}],"field_summary_sentence":[{"value":"Researchers from Interactive Computing unveil \u201cQuixote\u201d to teach AI positive behavior."}],"uid":"27490","created_gmt":"2016-02-12 10:31:53","changed_gmt":"2016-10-08 03:20:42","author":"Tara La Bouff","boilerplate_text":"","field_publication":"","field_article_url":"","dateline":{"date":"2016-02-12T00:00:00-05:00","iso_date":"2016-02-12T00:00:00-05:00","tz":"America\/New_York"},"extras":[],"hg_media":{"499531":{"id":"499531","type":"image","title":"Mark Riedl portrait","body":null,"created":"1455332400","gmt_created":"2016-02-13 03:00:00","changed":"1475895258","gmt_changed":"2016-10-08 02:54:18","alt":"Mark Riedl portrait","file":{"fid":"205887","name":"riedl_protrait_web.jpg","image_path":"\/sites\/default\/files\/images\/riedl_protrait_web.jpg","image_full_path":"http:\/\/www.tlwarc.hg.gatech.edu\/\/sites\/default\/files\/images\/riedl_protrait_web.jpg","mime":"image\/jpeg","size":268140,"path_740":"http:\/\/www.tlwarc.hg.gatech.edu\/sites\/default\/files\/styles\/740xx_scale\/public\/images\/riedl_protrait_web.jpg?itok=IVr-JhZg"}},"499551":{"id":"499551","type":"image","title":"Quixote flow chart","body":null,"created":"1455332400","gmt_created":"2016-02-13 03:00:00","changed":"1475895258","gmt_changed":"2016-10-08 02:54:18","alt":"Quixote flow chart","file":{"fid":"205888","name":"quixote_-_.jpg","image_path":"\/sites\/default\/files\/images\/quixote_-_.jpg","image_full_path":"http:\/\/www.tlwarc.hg.gatech.edu\/\/sites\/default\/files\/images\/quixote_-_.jpg","mime":"image\/jpeg","size":389839,"path_740":"http:\/\/www.tlwarc.hg.gatech.edu\/sites\/default\/files\/styles\/740xx_scale\/public\/images\/quixote_-_.jpg?itok=7atIB48V"}}},"media_ids":["499531","499551"],"groups":[{"id":"47223","name":"College of Computing"}],"categories":[{"id":"152","name":"Robotics"}],"keywords":[{"id":"2556","name":"artificial intelligence"},{"id":"169135","name":"Brent Harrison"},{"id":"66281","name":"Mark Riedl"},{"id":"166848","name":"School of Interactive Computing"}],"core_research_areas":[{"id":"39521","name":"Robotics"}],"news_room_topics":[{"id":"71881","name":"Science and Technology"},{"id":"71901","name":"Society and Culture"}],"event_categories":[],"invited_audience":[],"affiliations":[],"classification":[],"areas_of_expertise":[],"news_and_recent_appearances":[],"phone":[],"contact":[{"value":"\u003Cp\u003E\u003Ca href=\u0022mailto:tlabouff@cc.gatech.edu\u0022\u003E\u003Cstrong\u003ETara La \u003C\/strong\u003E\u003Cstrong\u003EBouff\u003C\/strong\u003E\u003C\/a\u003E\u003Cbr \/\u003E Communications Manager\u003Cbr \/\u003E 404-894-7253 (Office)\u003Cbr \/\u003E 404-769-5408 (Mobile)\u003C\/p\u003E","format":"limited_html"}],"email":["tlabouff@cc.gatech.edu"],"slides":[],"orientation":[],"userdata":""}}}