{"id":334320,"date":"2017-02-26T06:00:07","date_gmt":"2017-02-25T22:00:07","guid":{"rendered":"http:\/\/techcrunch.com\/?p=1455900"},"modified":"2017-02-26T06:00:07","modified_gmt":"2017-02-25T22:00:07","slug":"conversational-ai-and-the-road-ahead","status":"publish","type":"post","link":"https:\/\/people.utm.my\/asmawisham\/conversational-ai-and-the-road-ahead\/","title":{"rendered":"Conversational AI and the road ahead"},"content":{"rendered":"<p><img decoding=\"async\" src=\"https:\/\/tctechcrunch2011.files.wordpress.com\/2017\/01\/invisible-bot.png?w=738\" class=\"\"\/><\/p>\n<div class=\"byline contributor-byline\" readability=\"8.3206106870229\">\n<div class=\"name\" readability=\"7\">\n<p><span class=\"name-container\">Katherine Bailey<\/span> <span>Crunch Network Contributor<\/span><\/p>\n<\/p><\/div>\n<p class=\"bio\"><a href=\"https:\/\/www.linkedin.com\/in\/katbailey\">Katherine Bailey<\/a> is principal data scientist at <a href=\"https:\/\/www.acquia.com\/\">Acquia<\/a>.<\/p>\n<p>More posts by this contributor:<\/p>\n<\/div>\n<p>In recent years, we\u2019ve seen an increasing number of so-called \u201cintelligent\u201d digital assistants being introduced on various devices. At the recent CES, both <a target=\"_blank\" href=\"https:\/\/techcrunch.com\/2017\/01\/03\/hyundai-links-up-with-google-assistant-for-car-voice-commands\/\">Hyundai<\/a> and <a target=\"_blank\" href=\"https:\/\/techcrunch.com\/2017\/01\/06\/heres-what-its-like-to-drive-with-toyotas-yui-ai-in-car-assistant\/\">Toyota<\/a> announced new in-car assistants. Although the technology behind these applications keeps getting better, there\u2019s still a tendency for people to be disappointed by their capabilities \u2014 the expectation of \u201cintelligence\u201d is not being met.<\/p>\n<div readability=\"13.16067146283\">Despite great strides in natural language processing (NLP) by data-driven approaches, natural language <em>understanding<\/em> remains elusive. The <a target=\"_blank\" href=\"https:\/\/en.wikipedia.org\/wiki\/Winograd_Schema_Challenge\">Winograd Schema Challenge<\/a> is a recently proposed improvement on the Turing Test for assessing whether a machine can be judged \u201cintelligent.\u201d It\u2019s named after Terry Winograd, who produced the first example of the type of pronoun disambiguation problem used in the challenge:<\/div>\n<p>\u201c<em>The city councilmen refused the demonstrators a permit because they feared violence.<\/em>\u201d<\/p>\n<p>What does the word \u201cthey\u201d refer to here \u2014 \u200athe councilmen or the demonstrators? What if instead of \u201cfeared\u201d we wrote \u201cadvocated?\u201d This changes what we understand by the word \u201cthey.\u201d Why? It is clear to us that councilmen are more likely to fear violence, whereas demonstrators are more likely to advocate it. This information, which is vital for disambiguating the pronoun \u201cthey,\u201d is not in the text itself, which makes these problems extremely difficult for AI programs.<\/p>\n<p>The first ever Winograd Schema Challenge was held last July, and the winning algorithm achieved a score on the challenge that was \u201ca bit better than random.\u201d<\/p>\n<h2>Representations versus understanding<\/h2>\n<p>There\u2019s a technique for representing the words of a language that\u2019s proving incredibly useful in many NLP tasks, such as sentiment analysis and machine translation. The representations are known as word embeddings, and they are mathematical representations of words that are trained from millions of examples of word usage in order to capture meaning. This is done by capturing relationships between words. To use a classic example, a good set of representations would capture the relationship \u201cking is to man as queen is to woman\u201d by ensuring that a particular mathematical relationship holds between the respective vectors (specifically, king \u2013 man + woman = queen).<\/p>\n<p>Such vectorized representations are at the heart of Google\u2019s new translation system, although they are representations of entire sentences, not just words. The new system \u201creduces translation errors by more than 55-85 percent on several major language pairs\u201d and can perform zero-shot translation: translation between language pairs for which no training data exists.<\/p>\n<p>Given all this, it may seem surprising to hear Oren Etzioni, a leading AI researcher with a particular focus on NLP, quip: <em>When AI can\u2019t determine what \u201cit\u201d refers to in a sentence, it\u2019s hard to believe that it will take over the world.<\/em><\/p>\n<p>So, AI can perform adequate translations between language pairs it was never trained on but it can\u2019t determine what \u201cit\u201d refers to? How can this be?<\/p>\n<h2>Meaning is only captured indirectly<\/h2>\n<p>When hearing about how vectorized representations of words and sentences work, it can be tempting to think they really are capturing meaning in the sense that there is some understanding happening. But this would be a mistake. The representations are derived from examples of language use. Our use of language is driven by meaning. Therefore, the derived representations naturally reflect that meaning. But the AI systems learning such representations have no direct access to actual meaning.<\/p>\n<div><img loading=\"lazy\" decoding=\"async\" class=\"alignnone wp-image-1455903 size-full\" src=\"https:\/\/tctechcrunch2011.files.wordpress.com\/2017\/02\/meaning_usage.png?w=469&#038;h=283\" alt=\"meaning_usage\" width=\"469\" height=\"283\" srcset=\"https:\/\/tctechcrunch2011.files.wordpress.com\/2017\/02\/meaning_usage.png 469w, https:\/\/tctechcrunch2011.files.wordpress.com\/2017\/02\/meaning_usage.png?w=150&#038;h=91 150w, https:\/\/tctechcrunch2011.files.wordpress.com\/2017\/02\/meaning_usage.png?w=300&#038;h=181 300w\" sizes=\"(max-width: 469px) 100vw, 469px\"\/><\/div>\n<p>For the purposes of many NLP tasks, lack of access to actual meaning is not a serious problem.<\/p>\n<p>Not understanding what \u201cit\u201d refers to in a sentence is not going to have an enormous effect on translation accuracy\u200a \u2014 \u200ait might mean \u201cil\u201d is used instead of \u201celle\u201d when translating into French, but that\u2019s probably not a big deal.<\/p>\n<p>However, problems arise when trying to create a conversational AI:<\/p>\n<div class=\"wp-caption alignnone c4\" dir=\"ltr\" readability=\"7.3846153846154\"><img loading=\"lazy\" decoding=\"async\" class=\"wp-image-1455905 size-full\" src=\"https:\/\/tctechcrunch2011.files.wordpress.com\/2017\/02\/pasted-image-0-5.png?w=316&#038;h=177\" alt=\"pasted-image-0-5\" width=\"316\" height=\"177\" srcset=\"https:\/\/tctechcrunch2011.files.wordpress.com\/2017\/02\/pasted-image-0-5.png 316w, https:\/\/tctechcrunch2011.files.wordpress.com\/2017\/02\/pasted-image-0-5.png?w=150&#038;h=84 150w, https:\/\/tctechcrunch2011.files.wordpress.com\/2017\/02\/pasted-image-0-5.png?w=300&#038;h=168 300w\" sizes=\"(max-width: 316px) 100vw, 316px\"\/><\/p>\n<p class=\"wp-caption-text\">Screenshot from the sample bot you can create with IBM\u2019s conversation service following this <a target=\"_blank\" href=\"https:\/\/www.ibm.com\/watson\/developercloud\/doc\/conversation\/tutorial.html\">tutorial<\/a>.<\/p>\n<\/div>\n<p>Understanding the referents of pronouns is a pretty important skill for holding conversations. As stated above, the training data used to train AIs that perform NLP tasks does not include the necessary information for disambiguating these words. That information comes from knowledge about the world. Whether it\u2019s necessary to actually act as an embodied entity in the world or simply have vast amounts of \u201ccommon sense knowledge\u201d programmed in,\u00a0to glean the necessary information is still an open question. Perhaps it\u2019s something in-between.<\/p>\n<h2>Working within constraints<\/h2>\n<div readability=\"16.326264274062\">The problem of natural language understanding in AI is being worked on by some very smart people. At NIPS 2016, the biggest AI conference of the year, researchers at OpenAI presented <a target=\"_blank\" href=\"https:\/\/arxiv.org\/pdf\/1610.03585v1.pdf\">A Paradigm for Situated and Goal-Driven Language Learning<\/a>. Researchers at Stanford are looking into interactive language learning, an approach that recognizes the importance of interacting with the world in order to learn meaning. Interestingly, their example system pays homage to Terry Winograd\u2019s SHRDLU system, an early conversational system that restricted itself to statements and questions about a world made up of blocks.<\/div>\n<div class=\"wp-caption alignnone c5\" dir=\"ltr\" readability=\"8.4517766497462\"><img loading=\"lazy\" decoding=\"async\" class=\"wp-image-1455904 size-full\" src=\"https:\/\/tctechcrunch2011.files.wordpress.com\/2017\/02\/pasted-image-0-6.png?w=300&#038;h=241\" alt=\"pasted-image-0-6\" width=\"300\" height=\"241\" srcset=\"https:\/\/tctechcrunch2011.files.wordpress.com\/2017\/02\/pasted-image-0-6.png 300w, https:\/\/tctechcrunch2011.files.wordpress.com\/2017\/02\/pasted-image-0-6.png?w=150&#038;h=121 150w\" sizes=\"(max-width: 300px) 100vw, 300px\"\/><\/p>\n<p class=\"wp-caption-text\">Terry Winograd\u2019s early Natural Language Understanding program SHRDLU restricted itself to statements about a world made up of blocks. By Ksloniewski (Own work) <a target=\"_blank\" href=\"https:\/\/creativecommons.org\/licenses\/by-sa\/4.0\/\">CC BY-SA 4.0<\/a>, via Wikimedia Commons<\/p>\n<\/div>\n<div readability=\"8.309243697479\"> For anybody wanting to build a conversational AI today, such restrictions are still absolutely necessary. Both Amazon\u2019s Lex and IBM\u2019s conversation service work by allowing the developer to specify the constraints within which their app should work. They define a set of intentions that the app can carry out, and map to those intentions the set of possible ways a user might request them.<\/div>\n<p>But there are ways of enhancing such conversational AI experiences even without solving natural language understanding (which may take decades, or longer). The image above showing a bot not understanding \u201cnow turn them back on\u201d when the immediately prior request was \u201cturn off the windshield wipers\u201d demonstrates how disappointing it is when a totally unambiguous pronoun cannot be understood. That is definitely solvable with today\u2019s technology.<\/p>\n<div readability=\"12.247104247104\">Understanding what\u2019s possible and what isn\u2019t with today\u2019s AI and machine learning capabilities is key for anyone looking to use such technology to build or enhance applications. If you\u2019re not skeptical enough about claims being made about current capabilities, you could waste enormous amounts of time and money trying to do something that can\u2019t (yet) be done. On the other hand, if you\u2019re too skeptical, you risk missing opportunities to deploy today\u2019s AI tech in <a target=\"_blank\" href=\"https:\/\/research.googleblog.com\/2016\/11\/zero-shot-translation-with-googles.html\">incredibly<\/a> <a target=\"_blank\" href=\"https:\/\/www.ncbi.nlm.nih.gov\/pubmed\/15360876\">useful<\/a> and <a target=\"_blank\" href=\"http:\/\/www.forbes.com\/forbes\/welcome\/?toURL=http:\/\/www.forbes.com\/sites\/laurashin\/2015\/12\/09\/kensho-the-financial-answer-machine\/2\/&#038;refURL=&#038;referrer=#197f179160d3\">profitable<\/a>\u00a0<a target=\"_blank\" href=\"https:\/\/medium.com\/talla-inc\/talla-for-q-a-is-available-on-slack-today-333b11be375e#.h6z45ghqm\">ways<\/a>.<img loading=\"lazy\" decoding=\"async\" class=\"CToWUd\" src=\"https:\/\/ci5.googleusercontent.com\/proxy\/wS3TpLZK0Yv_Q9_OIw4mqvy_-ss-FJJdH90SY9oBIdvExjgBkdV_Amkf29hoE2q42-bgSmiB1t0QThObcw6HqiXioIVDedbhAng4spy4jR4vQ4_Q-gEIPRXbUEXFIWpDuLAK-GtB1LojZ79LHQl4K845uGdgSBPV-26FSPDfYn-XTykYMDK4E7Yieb2YrVJwrXMmPBLvnDKE=s0-d-e1-ft#https:\/\/track.mixmax.com\/api\/track\/v2\/DB1TY37m7WnEuOFEU\/ISbvNmLkFWdxNHaj5WdhxGQk5WYyxmI\/i02bj5Caj5WdyNGajVGdAJXZiVWaoNnI\/?sc=false\" alt=\"\" width=\"0\" height=\"0\" align=\"left\"\/><\/div>\n<div class=\"yj6qo ajU ajR\"><img decoding=\"async\" class=\"ajT\" src=\"https:\/\/ssl.gstatic.com\/ui\/v1\/icons\/mail\/images\/cleardot.gif\"\/><\/div>\n<p><small>Featured Image: Bryce Durbin\/TechCrunch<\/small><\/p>\n","protected":false},"excerpt":{"rendered":"<p><img loading=\"lazy\" decoding=\"async\" width=\"680\" height=\"383\" src=\"https:\/\/tctechcrunch2011.files.wordpress.com\/2017\/01\/invisible-bot.png?w=680\" class=\"attachment-large size-large wp-post-image\" alt=\"invisible-bot\" srcset=\"https:\/\/tctechcrunch2011.files.wordpress.com\/2017\/01\/invisible-bot.png?w=680 680w, https:\/\/tctechcrunch2011.files.wordpress.com\/2017\/01\/invisible-bot.png?w=1360 1360w, https:\/\/tctechcrunch2011.files.wordpress.com\/2017\/01\/invisible-bot.png?w=150 150w, https:\/\/tctechcrunch2011.files.wordpress.com\/2017\/01\/invisible-bot.png?w=300 300w, https:\/\/tctechcrunch2011.files.wordpress.com\/2017\/01\/invisible-bot.png?w=768 768w\" sizes=\"(max-width: 680px) 100vw, 680px\" \/>&nbsp;In recent years, we&#8217;ve seen an increasing number of so-called &#8220;intelligent&#8221; digital assistants being introduced on various devices. Although the technology behind these applications keeps getting better, there&#8217;s still a tendency for people to be disappointed by their capabilities &#8212; the expectation of &#8220;intelligence&#8221; is not being met. <a href=\"https:\/\/techcrunch.com\/2017\/02\/25\/conversational-ai-and-the-road-ahead\/?ncid=rss\">Read More<\/a><\/p>\n<div class=\"feedflare\">\n<a href=\"http:\/\/feeds.feedburner.com\/~ff\/Techcrunch?a=Z8uV7XoSi6w:l1gz3ObuBvM:2mJPEYqXBVI\"><img decoding=\"async\" src=\"http:\/\/feeds.feedburner.com\/~ff\/Techcrunch?d=2mJPEYqXBVI\" border=\"0\"><\/img><\/a> <a href=\"http:\/\/feeds.feedburner.com\/~ff\/Techcrunch?a=Z8uV7XoSi6w:l1gz3ObuBvM:7Q72WNTAKBA\"><img decoding=\"async\" src=\"http:\/\/feeds.feedburner.com\/~ff\/Techcrunch?d=7Q72WNTAKBA\" border=\"0\"><\/img><\/a> <a href=\"http:\/\/feeds.feedburner.com\/~ff\/Techcrunch?a=Z8uV7XoSi6w:l1gz3ObuBvM:yIl2AUoC8zA\"><img decoding=\"async\" src=\"http:\/\/feeds.feedburner.com\/~ff\/Techcrunch?d=yIl2AUoC8zA\" border=\"0\"><\/img><\/a> <a href=\"http:\/\/feeds.feedburner.com\/~ff\/Techcrunch?a=Z8uV7XoSi6w:l1gz3ObuBvM:-BTjWOF_DHI\"><img decoding=\"async\" src=\"http:\/\/feeds.feedburner.com\/~ff\/Techcrunch?i=Z8uV7XoSi6w:l1gz3ObuBvM:-BTjWOF_DHI\" border=\"0\"><\/img><\/a> <a href=\"http:\/\/feeds.feedburner.com\/~ff\/Techcrunch?a=Z8uV7XoSi6w:l1gz3ObuBvM:D7DqB2pKExk\"><img decoding=\"async\" src=\"http:\/\/feeds.feedburner.com\/~ff\/Techcrunch?i=Z8uV7XoSi6w:l1gz3ObuBvM:D7DqB2pKExk\" border=\"0\"><\/img><\/a> <a href=\"http:\/\/feeds.feedburner.com\/~ff\/Techcrunch?a=Z8uV7XoSi6w:l1gz3ObuBvM:qj6IDK7rITs\"><img decoding=\"async\" src=\"http:\/\/feeds.feedburner.com\/~ff\/Techcrunch?d=qj6IDK7rITs\" border=\"0\"><\/img><\/a>\n<\/div>\n<p><img loading=\"lazy\" decoding=\"async\" src=\"http:\/\/feeds.feedburner.com\/~r\/Techcrunch\/~4\/Z8uV7XoSi6w\" height=\"1\" width=\"1\" alt=\"\"\/> <\/p>\n","protected":false},"author":5817,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_et_pb_use_builder":"","_et_pb_old_content":"","_et_gb_content_width":"","footnotes":""},"categories":[25],"tags":[59,66,26],"class_list":["post-334320","post","type-post","status-publish","format-standard","hentry","category-technology","tag-media","tag-techcrunch","tag-technology"],"_links":{"self":[{"href":"https:\/\/people.utm.my\/asmawisham\/wp-json\/wp\/v2\/posts\/334320","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/people.utm.my\/asmawisham\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/people.utm.my\/asmawisham\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/people.utm.my\/asmawisham\/wp-json\/wp\/v2\/users\/5817"}],"replies":[{"embeddable":true,"href":"https:\/\/people.utm.my\/asmawisham\/wp-json\/wp\/v2\/comments?post=334320"}],"version-history":[{"count":0,"href":"https:\/\/people.utm.my\/asmawisham\/wp-json\/wp\/v2\/posts\/334320\/revisions"}],"wp:attachment":[{"href":"https:\/\/people.utm.my\/asmawisham\/wp-json\/wp\/v2\/media?parent=334320"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/people.utm.my\/asmawisham\/wp-json\/wp\/v2\/categories?post=334320"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/people.utm.my\/asmawisham\/wp-json\/wp\/v2\/tags?post=334320"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}