<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by The Alternative AI on Medium]]></title>
        <description><![CDATA[Stories by The Alternative AI on Medium]]></description>
        <link>https://medium.com/@thealternativeai?source=rss-e40596fb795f------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*czP94GhSix2MvRBy616NSg.png</url>
            <title>Stories by The Alternative AI on Medium</title>
            <link>https://medium.com/@thealternativeai?source=rss-e40596fb795f------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 18:08:44 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@thealternativeai/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Terms & Conditions Buster]]></title>
            <link>https://medium.com/@thealternativeai/terms-conditions-buster-f64e474303eb?source=rss-e40596fb795f------2</link>
            <guid isPermaLink="false">https://medium.com/p/f64e474303eb</guid>
            <category><![CDATA[ai-and-ux]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-ethics]]></category>
            <dc:creator><![CDATA[The Alternative AI]]></dc:creator>
            <pubDate>Fri, 25 Oct 2024 06:01:43 GMT</pubDate>
            <atom:updated>2024-10-25T09:36:48.369Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zN1v4IidZF9McFJNn42jmw.png" /></figure><blockquote>Ever wondered what unfair clauses might lurk deep inside the terms and conditions of your favorite websites? Ever actually spent time looking for them? No? Then we have the perfect tool for you!</blockquote><h4>About Us</h4><p>We are The Alternative AI, a collective interested in doing ethics. We want to create tangible tools to help create responsible, purposeful AI. In 2021, after a series of interviews with experts and workshops, we made a catalogue — a collection of stories and fictive tools to help the audience <em>do</em> ethics while applying the ALTAI framework developed by the EU.</p><h3>Welcome to Story #2, the Terms &amp; Conditions Buster!</h3><h4>The context</h4><p><a href="https://blogs.ischool.berkeley.edu/w231/2021/07/09/do-we-actually-agree-to-these-terms-and-conditions/">A 2017 Deloitte survey of 2,000 consumers in the US </a>found that 91% of people consent to legal terms and services without reading them. For younger people, ages 18–34, this rate rises to 97% of users agreeing to conditions before reading. It is an easy assumption to expect those numbers to generalise quite well to the majority of the internet’s users.<br>This seemingly universal lack of interest in legal matters presents the less scrupulous companies and websites with a golden opportunity for circumventing legal requirements and other guidelines for ethical conduct and responsible use of personal data.</p><p>At a time when many countries and governing entities are looking into introducing stronger mechanisms for regulating AI-powered services, we believe that such a glaring loophole should also be addressed. One possibility could be to educate users on the importance of paying better attention to terms and conditions, especially when sensitive data comes into play. However, when factoring the habitual length and insipidity of said legal terms with the (understandable) distaste of most users for boring tasks, our possible resistance to change, or the lack of domain knowledge required to efficiently navigate through the maze of legal and technical jargon, such an approach seems rather impractical if not hopeless.</p><h4>The tool</h4><p>Our solution to this problem comes in the form of another browser extension to which one can defer the painstaking task of analysing the terms and conditions of any website and web service they encounter. Users simply need to point the forensic bot towards the legal terms page to have it scan in detail each paragraph and, thanks to underlying text analysis models, return in a simple digest of its content together with alerts regarding potentially suspicious elements.</p><h4>Technical Details</h4><p>When we first envisioned this tool a few years ago, the technical challenges were significant, requiring extensive efforts to gather the necessary data and expert annotations needed to train specialised models.<br>With the advent of Large Language Models (LLMs), building an automated tool to analyse terms and conditions has become more efficient and effective. These models excel at understanding and generating human-like text, making them ideal for tasks such as text classification, named entity recognition, summarization, and anomaly detection within documents.</p><p><strong>Capabilities of LLMs for Analysis</strong></p><p>LLMs bring several advancements that significantly streamline the creation of our tool:</p><ol><li><strong>Advanced language comprehension</strong>: Trained on extensive datasets, LLMs have a deep understanding of language structures and semantics, enabling them to interpret complex documents like terms and conditions.</li><li><strong>Reduced need for extensive annotation</strong>: LLMs can perform effectively with minimal fine-tuning, decreasing the reliance on large annotated datasets that were previously necessary.</li><li><strong>Efficient summarization and change detection</strong>: They can succinctly summarise lengthy documents and detect changes between different versions, helping users quickly grasp important updates.</li><li><strong>Anomaly detection</strong>: LLMs can identify unusual clauses or deviations from standard practices, alerting users to potential concerns within the terms.</li><li><strong>Flexibility and adaptability</strong>: These models can be easily adapted to new domains or updated legal standards, ensuring the tool remains current without extensive retraining.</li></ol><p><strong>Limitations</strong></p><p>Despite the advancements brought by LLMs, there are still limitations that impact the development of our tool:</p><ol><li><strong>Lack of specialized legal understanding</strong>: While LLMs are proficient in general language tasks, they may not fully grasp complex legal concepts or the nuanced language used in terms and conditions.<br><em>Mitigation</em>: Collaborate with legal experts to fine-tune the model on domain-specific datasets, enhancing its ability to interpret and analyse legal documents accurately.</li><li><strong>Potential for inaccuracies</strong>: LLMs can sometimes generate incorrect or misleading information, especially when dealing with intricate legal clauses or uncommon scenarios.<br><em>Mitigation</em>: Implement validation mechanisms where the model’s outputs are reviewed by legal professionals or cross-referenced with trusted legal resources to ensure accuracy.</li><li><strong>Dependence on quality data and annotations</strong>: Achieving high accuracy requires quality training data and annotations, particularly in specialised domains like law.<br><em>Mitigation</em>: Invest in creating a curated dataset with expert annotations to train and refine the model, ensuring it learns from accurate and relevant examples.</li><li><strong>Challenges with ambiguity and context</strong>: Legal documents often contain ambiguous language or require deep contextual understanding that LLMs might misinterpret.<br><em>Mitigation</em>: Incorporate context-aware algorithms and consider integrating additional models specialised in legal reasoning to complement the LLM’s capabilities.</li></ol><p>Despite their limitations, Large Language Models make it possible to address the analysis of terms and conditions with significantly less time and resources than previously required. Their advanced capabilities streamline the development process, reducing the need for extensive data collection and expert annotations that were once essential.</p><p><strong>Important remarks:</strong></p><ul><li>This tool does not require any personal data.</li><li>The assessment of terms and conditions is not performed in real-time or each time a user accesses a website. It is only necessary when one of our users visits a site not yet in our database, or when familiar websites update their T&amp;C. This means LLM usage would be higher at the beginning, as the tool needs to assess many websites during the initial phase, but it would quickly decrease to a low, steady level, ensuring minimal long-term energy consumption.</li></ul><p><strong>Here are the members of our collective who created the stories:<br></strong>Nadia Domide — Siddarth Pandit — Kevin Payet — Reza Arkan — Suhaib Aslam — Fred Wordie — Celine Dutier</p><p>A big thanks to the amazing community <strong>AI x Design</strong> who supported us.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f64e474303eb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Cookie Monster]]></title>
            <link>https://medium.com/@thealternativeai/the-cookie-monster-9ad8052096fd?source=rss-e40596fb795f------2</link>
            <guid isPermaLink="false">https://medium.com/p/9ad8052096fd</guid>
            <category><![CDATA[ethics-in-tech]]></category>
            <category><![CDATA[ux]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[The Alternative AI]]></dc:creator>
            <pubDate>Sun, 23 Apr 2023 14:07:36 GMT</pubDate>
            <atom:updated>2025-04-13T17:55:35.993Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PT5BEhzUe5KVIVsT2Kx9rg.png" /></figure><blockquote>Why should protecting one’s privacy be a hassle? Why should users “Accept All” cookies by default? The Cookies Monster deals with your cookie preferences and lets you focus on what actually matters.</blockquote><h4>About Us</h4><p>We are The Alternative AI, a collective interested in doing ethics. We want to create tangible tools to help create responsible, purposeful AI. In 2021, after a series of interviews with experts and workshops, we made a catalogue — a collection of stories and fictive tools to help the audience <em>do </em>ethics<em>. </em>You are currently reading our first story!</p><h3>Welcome to Story #1, the Cookie Monster!</h3><h4>The context</h4><p>In May 2021, all members of the EU adopted the “cookies directive” that requires websites used by its citizens to get consent from visitors to store or retrieve any information on their activity. Almost a year later and (almost) every website has implemented a mechanism to communicate explicitly on their cookies usage and ask for the appropriate permissions.</p><p>However, that newfound control over their personal data comes at the price of the smoothness of the experience. The disruptive cookies prompts have stopped more than one user in their flow, while many others choose to click on the fated “Accept All” to bypass that annoying obstacle quickly. Should privacy really be at odds with a satisfying user experience? We believe not.</p><h4><strong>The tool</strong></h4><p>The Cookies monster is a simple browser extension for modern browsers (Chrome, Firefox, Safari, …) that takes care of cookies each time you visit a website, remembers the choices for each of them, and lets you configure in advance your preferences (Yes to all, Yes to Analytics cookies no to the rest, Yes to all for somewebsite.org, etc.).</p><h4>Technical Details</h4><ul><li>A simple and mature stack: Javascript, HTML, CSS, Selenium</li><li>An easy standardisation: A majority of websites use the same types of banners that can automatically be filtered. For others, the simple regex could do the trick.</li></ul><h4><strong>Examples in the wild</strong></h4><ul><li><a href="https://ieeexplore.ieee.org/document/4272997">https://ieeexplore.ieee.org/document/4272997</a></li><li><a href="https://chrome.google.com/webstore/detail/i-dont-care-about-cookies/fihnjjcciajhdojfnbdddfaoknhalnja?hl=en">https://chrome.google.com/webstore/detail/i-dont-care-about-cookies/fihnjjcciajhdojfnbdddfaoknhalnja?hl=en</a></li></ul><p><strong>Here are the members of our collective who created those stories:<br></strong>Nadia Domide - Siddarth Pandit - Kevin Payet - Reza Arkan - Suhaib Aslam - Fred Wordie - Celine Dutier</p><p>A big thank you to the amazing community <strong>AI x Design</strong> and <strong>Nadia Piet</strong>, who supported us.</p><p><strong>In two weeks, we will share another story.<br>Please stay tuned!</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9ad8052096fd" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>