{"id":161,"date":"2019-05-02T16:46:42","date_gmt":"2019-05-02T16:46:42","guid":{"rendered":"https:\/\/cusjc.ca\/mrp\/dataandthedisplaced\/?page_id=161"},"modified":"2020-01-16T00:32:14","modified_gmt":"2020-01-16T00:32:14","slug":"governing-ethical-ai","status":"publish","type":"page","link":"https:\/\/cusjc.ca\/mrp\/dataandthedisplaced\/governing-ethical-ai\/","title":{"rendered":"Governing ethical AI"},"content":{"rendered":"

[et_pb_section fb_built=”1″ fullwidth=”on” _builder_version=”3.22.5″ background_image=”https:\/\/cusjc.ca\/mrp\/dataandthedisplaced\/wp-content\/uploads\/sites\/20\/2019\/06\/molnar-report-2.jpg” parallax=”on” min_height=”779px” custom_padding=”222px||2px|||”][et_pb_fullwidth_header title=”Governing ethical AI” subhead=”The use of artificial intelligence and digital identity might streamline the refugee experience, but what if these technologies harm more than they help?” content_max_width=”63%” _builder_version=”3.22.5″ title_font=”Jura|600|||||||” title_text_color=”#ffffff” title_font_size=”66px” title_line_height=”1.3em” title_text_shadow_style=”preset2″ title_text_shadow_blur_strength=”0.05em” subhead_font=”Georgia||||||||” subhead_text_align=”left” subhead_font_size=”22px” subhead_line_height=”1.4em” subhead_text_shadow_style=”preset2″ subhead_text_shadow_blur_strength=”0.04em” background_color=”rgba(0,40,150,0)” min_height=”461px” custom_padding=”0px|||||” text_shadow_style=”preset2″][\/et_pb_fullwidth_header][\/et_pb_section][et_pb_section fb_built=”1″ _builder_version=”3.22.5″ min_height=”7832px”][et_pb_row custom_padding=”|||||” custom_margin=”|||||” _builder_version=”3.22.5″ collapsed=”off”][et_pb_column type=”4_4″ _builder_version=”3.0.47″][et_pb_text _builder_version=”3.22.5″ width=”80%” width_last_edited=”on|desktop” module_alignment=”center”]<\/p>\n

I<\/span>mmigration lawyer Petra Molnar insists she\u2019s not a fearmonger. <\/span><\/p>\n

In September 2018, Molnar and human rights researcher Lex Gill issued a damning 88-page report<\/a> warning about the use of artificial intelligence in Canada\u2019s immigration and refugee system. The report was published through the University of Toronto\u2019s human rights program and the Citizen Lab and bore a chilling title: Bots at the Gate.\u00a0<\/span><\/p>\n

\u201cThe ramifications of using automated decision-making in the immigration and refugee space are far reaching,\u201d the report noted. \u00a0<\/span><\/p>\n

\u201cThe nuanced and complex nature of many refugee and immigration claims may be lost on these technologies, leading to serious breaches of internationally and domestically protected human rights, in the form of bias, discrimination, privacy breaches, due process and procedural fairness issues, among others.\u201d <\/span><\/p>\n

[\/et_pb_text][et_pb_image src=”https:\/\/cusjc.ca\/mrp\/dataandthedisplaced\/wp-content\/uploads\/sites\/20\/2019\/05\/molnar-pose.jpg” _builder_version=”3.22.5″ width=”80%” module_alignment=”center”][\/et_pb_image][et_pb_text _builder_version=”3.22.5″ text_font=”||||||||” text_line_height=”1.4em” width=”80%” module_alignment=”center” custom_margin=”-20px|||||”]<\/p>\n

Petra Molnar in February 2019, visiting Ottawa for a symposium on ethics and AI. Molnar said the idea behind Bots at the Gate was sparked a year earlier, when she first read about new technologies being used in the immigration and refugee sector. [Photo \u00a9 Raisa Patel]<\/span><\/em><\/p>\n

[\/et_pb_text][et_pb_text _builder_version=”3.22.5″ width=”80%” module_alignment=”center”]<\/p>\n

But Molnar maintains that she\u2019s not trying to instill terror over the use of AI in these systems. Her goal is to create nationwide dialogue about the responsible use of automation in government, a trend that is no longer speculative. <\/span><\/p>\n

\u201cIt\u2019s important to have a balanced discussion,\u201d Molnar said, \u201cbecause fear mongering isn’t really going to get us anywhere.\u201d<\/span><\/p>\n

[\/et_pb_text][et_pb_text _builder_version=”3.22.5″ width=”80%” module_alignment=”center” custom_margin=”|||||” custom_padding=”||0px|||”]<\/p>\n

Bots on the ground<\/strong><\/span><\/h3>\n

<\/strong><\/span><\/p>\n

Artificial intelligence in this context refers to a form of technology that sorts through data to mimic the human decision-making process. In Canada, using these tools for automated decision-making isn\u2019t something the federal government is thinking of using: it\u2019s already doing so. <\/span><\/p>\n

For the past several years, Ottawa has been quietly developing AI solutions to streamline a number of its administrative processes in the immigration and refugee sector. As early as 2014, the government has used predictive analytics to sort through immigrant and visitor applications. Then, in an early 2018 <\/span>request for information<\/span><\/a> about AI tools for several federal departments, it was revealed Immigration, Refugees and Citizenship Canada (IRCC) was seeking an \u201cartificial intelligence solution\u201d for a number of processes. <\/span><\/p>\n

In the request, IRCC \u2013 in partnership with the Justice Department \u2013 sought a \u201cbroad-based solution that can analyze large volumes of immigration litigation data to assist in the development of policy positions, program decisions and program guidance.\u201d The notice mentioned using historical data, identifying factors to predict successful litigation outcomes and putting this information in the hands of administrative decision-makers.<\/span><\/p>\n

\u201cIn other words,\u201d Molnar\u2019s report explained, \u201cthe proposed technology would be involved in screening cases for strategic legal purposes, as well as potentially for assisting administrative decision-makers in reaching conclusions in the longer-term.\u201d<\/span><\/p>\n

Last year, IRCC also revealed it has been using an AI tool since early 2018 to triage a large volume of temporary resident visa applications pouring in from India and China, a project slated to cost the department more than $850,000. <\/span><\/p>\n

[\/et_pb_text][et_pb_text _builder_version=”3.22.5″ width=”80%” module_alignment=”center” custom_margin=”|||||” custom_padding=”|||||”]<\/p>\n

Garbage in, garbage out<\/strong><\/span><\/h3>\n

<\/strong><\/span><\/p>\n

In federal departments, automated technology can provide significant time and cost-saving benefits. IRCC is often beleaguered by application backlog and sometimes boasts processing times that span multiple years. As Molnar notes, however, the significance of immigration and refugee decisions means applications must be handled carefully. <\/span><\/p>\n

But instead of providing a solution, artificial intelligence can introduce a serious problem: bias. <\/span><\/p>\n

[\/et_pb_text][\/et_pb_column][\/et_pb_row][et_pb_row custom_padding=”0px|||||” _builder_version=”3.22.5″ module_alignment=”center” collapsed=”off”][et_pb_column type=”1_2″ _builder_version=”3.0.47″][et_pb_text _builder_version=”3.22.5″ width=”80%” module_alignment=”right” custom_padding=”|||8px||”]<\/p>\n

In an algorithmic context, bias refers to a simple premise many experts classify as \u201cgarbage in, garbage out\u201d. Essentially, if the quality of data entering an algorithm is poor, it will generate poor results. In 2015, Amazon discovered an AI system it was using to select possible job candidates was <\/span>unintentionally sexist<\/span><\/a>. That\u2019s because it was trained on 10 years of resumes from successful applicants \u2013 who mostly turned out to be men. In another example, researchers from MIT and Stanford University <\/span>found<\/span><\/a> that facial recognition software was much more successful at identifying white men than it was black women. A data sample that predominantly consisted of white faces was the culprit.<\/span><\/p>\n

[\/et_pb_text][\/et_pb_column][et_pb_column type=”1_2″ _builder_version=”3.0.47″][et_pb_text quote_border_color=”#e04b43″ _builder_version=”3.22.5″ text_font=”||||||||” quote_font=”||||||||” quote_text_align=”left” quote_text_color=”#0c71c3″ header_font=”||||||||” transform_translate=”5px|27px” width=”80%” module_alignment=”left” custom_margin=”94px|||||” custom_padding=”|||||”]<\/p>\n

\n

\u201cI struggle with this presupposition that technology is somehow less biased than humans, because it’s created by human beings.”<\/span><\/h2>\n

\u2013 Petra Molnar<\/em><\/span><\/h2>\n<\/blockquote>\n

[\/et_pb_text][\/et_pb_column][\/et_pb_row][et_pb_row custom_padding=”0px||0px|||” _builder_version=”3.22.5″ module_alignment=”center” collapsed=”off”][et_pb_column type=”4_4″ _builder_version=”3.0.47″][et_pb_text _builder_version=”3.22.5″ width=”80%” max_width=”1080px” module_alignment=”center”]<\/p>\n

But humans are biased too, which includes administrative decision-makers who make life-altering calls about who gains entry into Canada. Documented cases of <\/span>bias<\/span><\/a> or potentially <\/span>culturally insensitive<\/span><\/a> language can be found in notes from immigration officers, which means this is the sort of data automated systems will rely on to sift through immigration and refugee applications. <\/span><\/p>\n

\u201cI struggle with this presupposition that technology is somehow less biased than humans, because it’s created by human beings,\u201d Molnar said of the belief that AI might be more fair than human decision-makers. <\/span><\/p>\n

[\/et_pb_text][et_pb_video src=”https:\/\/youtu.be\/NRDaZ5KAwnc” _builder_version=”3.22.5″ width=”80%” module_alignment=”center”][\/et_pb_video][et_pb_text _builder_version=”3.22.5″ width=”80%” module_alignment=”center” custom_margin=”|||||”]<\/p>\n

It would be reasonable to assume that Canada has a strict set of regulations to govern the use of such technology, particularly when an automated system\u2019s thought process is not as transparent as a human one. The reality, however, is not so clear.<\/span><\/p>\n

[\/et_pb_text][et_pb_text _builder_version=”3.22.5″ width=”80%” module_alignment=”center” custom_margin=”|||||” custom_padding=”|||||”]<\/p>\n

Who’s in charge?<\/strong><\/span><\/h3>\n

<\/strong><\/span><\/p>\n

In March 2019, then-Treasury Board president Jane Philpott announced the publication of Canada\u2019s <\/span>Directive on Automated Decision-Making<\/span><\/a>, an extensive policy document outlining how government should properly develop and deploy decision-making AI. The release of the directive marked the country\u2019s first step toward monitoring the use of AI in federal departments. But it was published mere hours before Philpott\u2019s resignation from cabinet, so the directive quickly disappeared from public consciousness. \u00a0<\/span><\/p>\n

While it may be a relatively unknown piece of policy, it\u2019s a critical one. The directive is a sweeping guide intended to reduce the risk of automated systems. It outlines rules for establishing transparency, like making sure it is clear how systems work and providing explanations for AI-based decisions; testing for unintended biases algorithms may generate on their own; ensuring humans can intervene in a system\u2019s decision-making process when necessary and listing consequences for a failure to comply. <\/span><\/p>\n

 <\/p>\n

[\/et_pb_text][et_pb_code _builder_version=”3.22.5″ width=”80%” module_alignment=”center” custom_margin=”|||||”]<\/p>\n

<\/div>\n