{
    "componentChunkName": "component---src-templates-generic-content-index-en-js",
    "path": "/ai-policy",
    "result": {"data":{"topNav":{"menus":[{"uuid":"","name":"About","navMenuLinks":[],"link":{"url":null,"page":{"path":"/about"}}},{"uuid":"work","name":"Our Work","navMenuLinks":[{"text":"Civic Engagement","page":{"path":"/civic-engagement"},"isSubMenu":false},{"text":"Public Opinion","page":{"path":"/public-opinion"},"isSubMenu":false},{"text":"Decision Science","page":{"path":"/decision-science"},"isSubMenu":false}],"link":null},{"uuid":"news","name":"News","navMenuLinks":[],"link":{"url":null,"page":{"path":"/news"}}},{"uuid":"","name":"Careers","navMenuLinks":[],"link":{"url":null,"page":{"path":"/careers"}}},{"uuid":"","name":"Data Explorer","navMenuLinks":[{"text":"Data Explorer","page":null,"isSubMenu":false}],"link":{"url":"https://data.voxpoplabs.com","page":null}},{"uuid":null,"name":"Français","navMenuLinks":[],"link":{"url":"/lang/fr","page":null}},{"uuid":"topNav_joinPanel","name":"Get Involved","navMenuLinks":[],"link":{"url":"/signup","page":null}},{"uuid":"topNav_contactUs","name":"Contact us","navMenuLinks":[],"link":{"url":null,"page":{"path":"/contact"}}}]},"genericContentPage":{"id":"9ae784e8-c83f-552e-a7e1-eb79601cc97d","uuid":"aiPolicy","path":"/ai-policy","locale":"en","seo":{"metaTitle":"Vox Pop Labs - AI Policy","metaDescription":"Empowering informed decisions. We build data-driven digital products that help navigate complex choices.","metaImage":{"url":"https://s3.ca-central-1.amazonaws.com/site.voxpoplabs.com/vpl_logo_56477a84de.png"},"twitterHandle":null},"sections":[{"uuid":"genericContent","title":"Our Approach to Artificial Intelligence","description":{"data":{"description":"Vox Pop Labs works at the intersection of digital technology, civic engagement, and democratic participation. AI is changing the conditions under which we do this work. It is also changing how citizens encounter information, formulate opinions, and participate in public life. We cannot adopt AI uncritically. This document sets out how we are currently thinking about AI and the principles we are applying as we go forward with our work. We are writing this knowing that the technology, and our understanding of it, will almost invariably change. \n\n**What we mean by AI:** In this document, “AI” refers primarily to generative AI — large language models and related systems that can analyze, generate, and synthesize content. Where the specific model, technology, or method matters for context, now, or in the future, we will name it explicitly. This document does not cover the statistical and machine learning methods that have long been part of standard data science practice in survey research, though many of the same principles apply.\n\n## **Tl;dr**\n\nWe are a small organization using AI tools to extend the reach of work that matters for democracy. We are aware that the companies building these tools have incentives to grow quickly and that civic information is a domain where those incentives can do real damage. Our approach is iterative: test, document, interrogate the results, ensure compliance with our standards and first principles,, and reserve the right to change course as we learn.\n\n# **Awareness of AI’s impact**\n\nAI brings a set of ethical, societal, and environmental considerations. As a certified [B Corp](https://www.bcorporation.net/en-us/find-a-b-corp/company/vox-pop-labs/), we factor our use of AI into the broader question of how we operate in the world. The list below is not exhaustive, but it reflects what is front of mind for us right now.\n\n* **Environmental and community costs.** AI models consume substantially more electricity than traditional computing, and the data centres they depend on draw heavily on local water and power. People deserve a voice in decisions about the infrastructure built in their communities for this purpose. We have the research capabilities to surface those voices for policymakers, and we should use them.\n\n* **Social and economic dynamics.** AI is reshaping industries, including ours — survey research, civic media, and political communication are all changing in real time. We take seriously what this means for labour markets and for the dignity of work. The largest models are built by a small number of well-resourced companies, often relying on underpaid data labellers. Labour practices are an active consideration in choosing which AI vendors we work with.\n\n* **Intellectual property.** Large AI models are trained on vast amounts of data — including much of the open web — frequently without permission. We are not legal experts, but we take the ethics of building on unlicensed or appropriated work seriously, and we follow the responses from the companies behind these models on the question of crediting the work they stand on.\n\n* **Misaligned incentives.** Two decades of social media have shown what happens when civic life is shaped by businesses optimizing for engagement and growth. We will not let the same incentives shape AI’s role in democratic infrastructure. Success in our work is not measured by time spent or queries served. We use AI to build for prosocial outcomes: comprehension, deliberation, perspective-taking, informed participation, and a stronger sense of agency.\n\n* **Information integrity and democratic risk.** Our work touches elections, public opinion, and political discourse, so we hold ourselves to a higher standard on accuracy and influence. Generative models can produce confident falsehoods, mirror partisan framings, and be used to manufacture astroturf. We will not use AI in ways that could distort the signals — voter intent, public sentiment, deliberative reasoning — that our work exists to measure and amplify.\n\n* **Privacy and respondent trust.** Our products gather sensitive information about people’s political views. Protecting that data is non-negotiable, and we are especially cautious about sending respondent data to third-party AI services whose data-handling practices we have not vetted.\n\nAI is already changing how citizens encounter information and form views. That makes engagement urgent. Alongside the work of pushing on the items above, we also want to help build AI tools that serve the public and support democratic institutions, rather than waiting to see what others do.\n\n# **Where we engage as critics and builders**\n\nWe have been using AI tools across our work for some time. The areas below are where we have something concrete to share right now.\n\n**Building civic tools, faster.** We are a small team in a space where the dominant players pay individual researchers more than our organization’s entire operating budget. AI-assisted design and engineering let our researchers, designers, and developers prototype and ship at a pace that was not previously available to us. This is not about replacing craft. It is about giving a small team enough leverage to keep building public-interest tools.\n\n***In practice:** Our product team uses AI-assisted coding for prototyping new features in Vote Compass and related tools, so that designers and researchers can test ideas in working interfaces before we commit engineering time to a full build. We are also prototyping features that use AI to process hundreds of thousands of municipal documents, surfacing issues for inclusion in Vote Compass at a scale that would not be feasible for human researchers alone.*\n\n**Making sense of public opinion at scale.** Our work generates large volumes of qualitative data — open-ended responses, comments, deliberative inputs — that have historically been costly to analyze well. AI can help cluster, summarize, and surface patterns in this material, which makes it more practical to take seriously what people say in their own words rather than collapsing everything into closed-ended scales. Used carefully, this can make our research more representative of the voices it claims to capture, not less.\n\n***In practice:** Our research team is testing AI-assisted analysis of open-ended survey responses, with human researchers validating clusters, checking for representational gaps, and making the interpretive calls.*\n\n**Studying political bias in commercial AI tools.** Early research suggests that when people use commercial AI chatbots to help decide how to vote, the tools tend to push toward more extreme positions. We are running our own experiments to understand how and why this happens, and whether the effect can be mitigated. The aim is to contribute to the academic literature in this area and to apply what we learn to our own products.\n\n***In practice:** In our work on AI-supported municipal Vote Compasses, we designed and ran an agentic deliberation: we engaged several different AI tools (each perceived by the public to come from a different position on the political spectrum) and had them reason with each other to surface bias and to see how they interact when they were asked to reach a joint conclusion.*\n\n**Working out what “human in the loop” actually means.** Organizations across many fields are working out what it means to do a job that interfaces with AI, and what accountability looks like in that arrangement. Our position is that any efficiency gains from AI should benefit the work itself, not the bottom line. Our staff decide for themselves where they are comfortable using AI and where they are not, so that they can take responsibility for the outputs. That also means we do not push AI into tasks where our people find value in doing the work themselves.\n\n***In practice:** Many organizations are moving quickly toward “silicon samples” — synthetic research data where a simulated respondent generated by a language model stands in for a human (for example, prompting an LLM to answer survey questions as if it were a person). We are well-positioned to benefit from this approach, but we have deliberately engaged with it less than we could. Before going further, we want to ask whether this is what people actually want: technology representing them and their opinions without their being consulted. Our researchers currently do not want to replace human respondents with AI, even where it would produce efficiencies or cost savings.*\n\n# **Our guiding principles**\n\nAs we keep going, the principles below describe how we want to use AI in line with our values and our mission.\n\n**We work transparently.** Our work depends on public trust, which means we hold ourselves to a transparency standard higher than the norm. When AI is involved in something users see, read, or interact with through our products, we will say so. We will share what we are doing openly, including what worked, what failed, and what we changed our minds about. We will not let the cultural noise around AI, in either direction, prevent honest conversation about how we are using it. Our team remains accountable for every output that carries our name.\n\n**We keep humans in charge.** At Vox Pop Labs, humans define the methods, interpret the results, and make every final call. AI does not make research, editorial, or governance decisions. We recognize that AI’s benefit — for example, its ability to process large volumes of information — sometimes makes its outputs hard to review. If a mistake gets through in something we release publicly, we will be transparent about it with our partners and the wider community, and we will revisit our processes to make sure we have not put ourselves in a position where humans cannot own the work.\n\n**We navigate AI’s biases, including political ones.** Today’s models are not neutral. They are trained predominantly on English-language, Western, online text, and they carry the political, cultural, and ideological tilts of that corpus. For an organization working across multiple countries, languages, and political traditions, this is a real limitation. We cross-check AI outputs against original sources. We test for political slant in any model output that touches issue framing. We involve people with different perspectives in reviewing what AI produces on contested topics. And we treat any AI-generated characterization of a party, candidate, or policy position as a draft to be verified.\n\n**We protect respondent data.** We read the terms and conditions of the AI tools we use. We do not feed identifiable respondent data into third-party models without first reviewing how that data will be stored, used, or used for training. When AI is used to analyze sensitive material, we prefer arrangements that keep the data within environments we control or have specifically vetted. Privacy is not a checkbox; it is a precondition for the trust our work depends on.\n\n**We slow down on purpose.** In a tech culture that prizes scale and speed, we think there is value in slowing down. Before adopting an AI tool for a given task, we ask:\n\n* What is the specific task at hand?\n\n* What role are we assigning to AI, and why this tool rather than another approach?\n\n* Who is the intended audience, and what are the intended outcomes?\n\n* Who might be affected — respondents, users, partners, the public — and how?\n\n* What happens if the AI is wrong, and how would we know?\n\nAsking these questions first helps us pick the right tool for the job and limits the downstream harm of getting it wrong. We are not interested in adopting technology for its own sake.\n\n**We expect our approach to keep changing.** There is no single comprehensive list of do’s and don’ts — AI is changing too quickly, and context matters. Our engineering team may use AI extensively for code, while our research team uses it sparingly and with heavy oversight for any analysis that touches respondent data, and we may avoid it entirely for work where our own voice and judgment are essential. A fixed rulebook would be obsolete within a quarter. Instead, we commit to ongoing reflection, team-by-team norms, and shared responsibility as our practices evolve with the technology.\n\n**We design for pluralism and the public good.** Across our work, we want to elevate diverse voices, give people meaningful agency in articulating their views, and build understanding across difference. We will use AI to further this work, and we will refuse to use it in ways that undermine it. We will not use AI to automate the human relationships at the core of democratic life — between citizens and institutions, between researchers and the communities they study, between deliberators encountering one another’s reasoning. We will not use AI to make consequential decisions about people. And we will not use it in ways that mislead users about what they are interacting with.\n\n**We invest in our team’s capacity to do this well.** Using AI responsibly takes more than principles. It takes literacy, practice, and resources. We commit to:\n\n* Regular team-specific sessions to review how we are actually using AI in our work\n\n* Building AI literacy across the organization, including for non-technical staff\n\n* Setting aside time and budget for experimentation, evaluation, and unlearning\n\n* Sharing what we learn, including the mistakes, with peers and the public\n\nThis is a work in progress, not a final statement of our relationship with AI. The tools will keep changing, and so will our use of them. Our values will not. We intend to give this technology the considered, critical attention it deserves — in service of a healthier public sphere and a more informed democratic life."}},"link":null,"media":[],"partners":[],"timeline":[],"statistics":[],"miscText":[]}]},"bottomNav":{"menus":[{"uuid":"","name":"Civic engagement","navMenuLinks":[{"text":"Vote Compass","url":null,"page":{"path":"/vote-compass"}},{"text":"The Signal","url":null,"page":{"path":"/the-signal"}},{"text":"Polimeter","url":null,"page":{"path":"/polimeter"}}],"link":null},{"uuid":null,"name":"Public opinion","navMenuLinks":[{"text":"OurCare","url":null,"page":{"path":"/our-care"}},{"text":"Federal-Provincial Relations","url":null,"page":{"path":"/canada-federal-provincial-relations-study"}},{"text":"COVID-19 Monitor","url":null,"page":{"path":"/covid19-monitor"}},{"text":"News Effectiveness Study","url":null,"page":{"path":"/news-effectiveness-study"}},{"text":"Australia Talks","url":null,"page":{"path":"/australia-talks"}}],"link":null},{"uuid":"decisionTools","name":"Decision science","navMenuLinks":[{"text":"Vote Compass","url":null,"page":{"path":"/vote-compass"}},{"text":"DegreeHub","url":null,"page":{"path":"/degreehub"}},{"text":"Career Myth Buster","url":null,"page":{"path":"/career-myth-buster"}}],"link":{"url":null,"page":null}}],"sections":[{"uuid":"bottomNav_panel","title":"Join the Thought Leader Community","description":{"data":{"description":"Become a member of our online respondent panel, made up of people from all over the world and all walks of life. You’ll occasionally receive requests to participate in surveys about politics and public affairs. By participating, you’ll help us reflect the public interest to those in power."}},"link":{"text":"Get started","url":"","page":{"path":"/signup"}},"media":[],"partners":[],"timeline":[],"statistics":[],"miscText":[]},{"uuid":"bottomNav_logos","title":"Bottom Nav Logs - Vox Pop","description":null,"link":null,"media":[{"uuid":"bottomNav_logos-vpl","name":"Vox Pop Logo","image":{"url":"https://s3.ca-central-1.amazonaws.com/site.voxpoplabs.com/vpl_logo_56477a84de.png"},"imageAlt":"Vox Pop Labs Logo"}],"partners":[],"timeline":[],"statistics":[],"miscText":[]},{"uuid":"bottomNav_logos","title":"Bottom Nav Logos - BCorp","description":null,"link":{"text":null,"url":"https://www.bcorporation.net/en-us/find-a-b-corp/company/vox-pop-labs","page":null},"media":[{"uuid":"bottomNav_logos-bcorp","name":"B Corp Logo","image":{"url":"https://s3.ca-central-1.amazonaws.com/site.voxpoplabs.com/bcorp_7410a87cd2.png"},"imageAlt":"B Corp Logo"}],"partners":[],"timeline":[],"statistics":[],"miscText":[]},{"uuid":"bottomNav_footerDisclaimer","title":"© $YEAR Vox Pop Labs. All rights reserved.","description":null,"link":null,"media":[],"partners":[],"timeline":[],"statistics":[],"miscText":[]},{"uuid":"bottomNav_footerLink","title":"Footer link - About","description":null,"link":{"text":"About","url":null,"page":{"path":"/about"}},"media":[],"partners":[],"timeline":[],"statistics":[],"miscText":[]},{"uuid":"bottomNav_footerLink","title":"Footer link - News","description":null,"link":{"text":"News","url":null,"page":{"path":"/news"}},"media":[],"partners":[],"timeline":[],"statistics":[],"miscText":[]},{"uuid":"bottomNav_footerLink","title":"Footer Link - Contact","description":null,"link":{"text":"Contact","url":null,"page":{"path":"/contact"}},"media":[],"partners":[],"timeline":[],"statistics":[],"miscText":[]},{"uuid":"bottomNav_footerLink","title":"Footer Link - Careers","description":null,"link":{"text":"Careers","url":null,"page":{"path":"/careers"}},"media":[],"partners":[],"timeline":[],"statistics":[],"miscText":[]},{"uuid":"bottomNav_footerLink","title":"Footer Link - Privacy Policy","description":null,"link":{"text":"Privacy Policy","url":null,"page":{"path":"/privacy-policy"}},"media":[],"partners":[],"timeline":[],"statistics":[],"miscText":[]},{"uuid":"bottomNav_footerLink","title":"Footer Link - AI Policy","description":null,"link":{"text":"AI Policy","url":null,"page":{"path":"/ai-policy"}},"media":[],"partners":[],"timeline":[],"statistics":[],"miscText":[]}]},"socialMedia":{"nodes":[{"uuid":"facebook","url":"https://www.facebook.com/voxpoplabs","icon":{"url":"https://s3.ca-central-1.amazonaws.com/site.voxpoplabs.com/facebook_29effa11f3.png"}},{"uuid":"linkedin","url":"https://www.linkedin.com/company/vox-pop-labs","icon":{"url":"https://s3.ca-central-1.amazonaws.com/site.voxpoplabs.com/linked_in_f1fd9dcbb8.png"}},{"uuid":"x","url":"https://x.com/voxpoplabs","icon":{"url":"https://s3.ca-central-1.amazonaws.com/site.voxpoplabs.com/logo_black_03f130acf2.png"}},{"uuid":"instagram","url":"https://www.instagram.com/votecompass/","icon":{"url":"https://s3.ca-central-1.amazonaws.com/site.voxpoplabs.com/instagram_a2aa23ed89.png"}},{"uuid":"bluesky","url":"https://voxpoplabs.bsky.social","icon":{"url":"https://s3.ca-central-1.amazonaws.com/site.voxpoplabs.com/bluesky_840aff3bfa.png"}}]}},"pageContext":{"id":"9ae784e8-c83f-552e-a7e1-eb79601cc97d"}},
    "staticQueryHashes": ["1645764798","1923763410"]}