<?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 sohayb el amraoui on Medium]]></title>
        <description><![CDATA[Stories by sohayb el amraoui on Medium]]></description>
        <link>https://medium.com/@elamraoui.sohayb?source=rss-d8caa06f38d5------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*zz_XOIaR5uQvIR2H7TIj4g.jpeg</url>
            <title>Stories by sohayb el amraoui on Medium</title>
            <link>https://medium.com/@elamraoui.sohayb?source=rss-d8caa06f38d5------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 04:38:15 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@elamraoui.sohayb/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[Conciliating AI & Privacy: Rules and Good Practices]]></title>
            <link>https://pub.towardsai.net/conciliating-ai-privacy-rules-and-good-practices-7859541c4977?source=rss-d8caa06f38d5------2</link>
            <guid isPermaLink="false">https://medium.com/p/7859541c4977</guid>
            <category><![CDATA[data]]></category>
            <category><![CDATA[ethics]]></category>
            <category><![CDATA[privacy]]></category>
            <category><![CDATA[gdpr]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[sohayb el amraoui]]></dc:creator>
            <pubDate>Sat, 19 Dec 2020 22:18:59 GMT</pubDate>
            <atom:updated>2020-12-19T22:19:25.293Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*UxYcjmCX24IGundnep-R1A.png" /><figcaption>Image by the author</figcaption></figure><h4><a href="https://towardsai.net/p/category/privacy-and-security">Privacy and Security</a></h4><h4>How to get the maximum value from data without violating customers privacy</h4><h4>1. AI and personal data: a sensitive debate</h4><p>Artificial Intelligence (AI), in its strict Computer Science definition, refers to any type of human intelligence presented by a computer, a robot, or another machine. Common use extends this definition to the ability of a computer or machine to mimic the capabilities of the human mind — learn from experiences and examples, recognize objects, understand and answer languages, take decisions — and the combination of these abilities with others to accomplish functions that a human being can, such as greeting a hotel client or driving a car.</p><p>The definitions of private data, for their part, vary according to the laws applicable in each situation, but we can describe it in general as a piece of information regarding a person or entity that can reasonably be excepted to be protected from the public view. <strong>The General Data Protection Regulation (GDPR)</strong> of the European Union, for instance, defines [1]:</p><blockquote>‘personal data’ refers to any information relating to an identified or identifiable natural person (‘data subject’)…directly or indirectly, …</blockquote><p>An overview of the commercial applications or the academic research in AI shows that <strong>as these models and algorithms continue to push the boundaries of state of the art, their appetite to consume any available data grows in ways that infringe on privacy interests.</strong></p><p>The debate on AI and the use of private data often highlight the limitations and shortcomings of these so-called intelligent applications. Implicit biases, massive collections, and the rapid evolution of the technologies used to make privacy laws quite complicated and often lagging behind.</p><p>This article will discuss <strong>the challenges of a new paradigm for the protection of privacy</strong>. We will then focus on <strong>the tools available to implement efficient AI systems</strong> while respecting the rules imposed by the General Data Protection Regulation.</p><blockquote><a href="https://datavalue-consulting.com/livre-blanc-migration-vers-le-big-data/"><strong>Download our white paper “Successful Big Data Migration”</strong></a></blockquote><h3>2. The 4 pillars to protect personal data in AI systems</h3><p>Ensuring privacy in AI systems and regulating the use of consumer data, therefore, requires a paradigm shift. This new vision approaches privacy in the context of AI and the associated risks in a more holistic way, with measures designed to regulate the processing of personal data and identify cases of breach.</p><p><strong>A number of organizations, companies, and several legislators offer different variations, but they all revolve around 4 pillars:</strong></p><ul><li><strong>Explainability:</strong></li></ul><p>Understanding how the data was used to arrive at a particular decision and which characteristics played an important role in the conclusion is no easy task. This requires:</p><ol><li>Identification of decisions made by AI</li><li>The analysis of specific decisions</li><li>Establishing means by which an individual can seek an explanation</li></ol><p><strong>Reverse engineering machine learning algorithms can be difficult, if not impossible, </strong>a difficulty which increases even more in the case of deep learning algorithms.</p><blockquote>The GDPR requires that, for any automated decision “where it produces legal effects concerning him or her or similarly significantly affects him or her” (credit, insurance coverage, etc.) the data subject can have recourse to a human being able to review the decision and explain its logic [2].</blockquote><p>Incorporating a human element into the decision-making loop adds an additional step in the development process in addition to a significant regulatory burden.</p><ul><li><strong>Transparency :</strong></li></ul><p><strong>Having a clear answer to the question “What does the company do with your data?” greatly facilitates the accountability of the company</strong> and dispels the concerns of users and/or partners.</p><p>The “privacy policy” is a concrete example. These traditionally long and unnecessary documents for most consumers can be replaced by statements:</p><ol><li>That provides a full description of nature and manner in which data is collected, used, and protected,</li><li>That identifies the important use cases of personal data by AI and the various decision-making algorithms implemented in a product.</li></ol><ul><li><strong>Risk Assessment :</strong></li></ul><p>This is another requirement of the GDPR for new technologies or for high-risk uses of data. In this context, it is about <strong>assessing and mitigating confidentiality risks in advance, </strong>including the potential biases in the design of an AI system and the data provided to that system, as well as the potential impact on users. For instance, Twitter created controversy last September because it was demonstrated by users that its photo cropping algorithm was “racially” biased [3].</p><ul><li><strong>Audits:</strong></li></ul><p>Whether internal (self-audits) or external (by third party organizations), <a href="https://datavalue-consulting.com/pilotage-transformation/consulting-it/"><strong>audits are still necessary for monitoring and compliance with requirements.</strong></a><strong> </strong>As audits are retrospective in nature, a good strategy is to combine their results for AI decisions with risk assessments, which are proactive. This can better inform the company’s position on AI and privacy, albeit — as is the case with explainability- auditing machine learning algorithms is difficult and still in development.</p><h3>3. AI and privacy: major technical trends</h3><p>Satisfying all the previously listed constraints while preserving the performance of the AI system may seem difficult, if not impossible, but there are several solutions to explore, in particular Federated Learning and Differential Privacy.</p><p><strong>Federated Learning is an innovative approach</strong> to train machine learning models in a decentralized manner. The dissociation between the training process and the need to store data internally allows companies to unlock more capacity while reducing possible expenses and risks.</p><p>This learning method goes beyond the use of local models that make predictions on the user’s devices, towards a real collaboration between the devices that are members of this architecture:</p><ul><li>The same machine learning model is first deployed on all devices,</li><li>These devices then complete the training phase with their local data and update the weights and hyper-parameters of the model locally,</li><li>Then they send these updates separately to the cloud or central server using encrypted communication,</li><li>These updates are then aggregated and averaged to obtain the average weights and hyper-parameters of the shared model.</li></ul><p>Most importantly, all training data remains on the user’s device, and no individual updates are stored in an identifiable way in the cloud.</p><figure><img alt="Communication in a federated learning architecture" src="https://cdn-images-1.medium.com/max/638/1*6r2nNTcq1wH9qy5lw0VSXw.png" /><figcaption>Communication in a federated learning architecture [4]</figcaption></figure><p>A flagship example of this type of application is the “Gboard” functionality developed for Android smartphones: When Gboard displays a suggestion, your phone locally stores information about the current context and whether you clicked on the suggestion (was the suggestion helpful or not?), the federated learning model trains locally on this history and millions of Android devices share their results (new weights and parameters) to have improvements in the next iteration of Gboard’s AI model.</p><figure><img alt="How the Google Keyboard service works [5]" src="https://cdn-images-1.medium.com/max/955/1*Fta4NXeqHa7gdm8b3Gfmxg.gif" /><figcaption>How the Google Keyboard service works [5]</figcaption></figure><p><strong>Differential Privacy transforms data from its raw state into a format that allows organizations to learn from the majority of that data</strong> while simultaneously ensuring that the results will not distinguish or re-identify the data of an individual.</p><p>Since the early 2000s, research has shown that <strong>87% of the American population can be uniquely identified by the combination {date of birth, gender, zip code}</strong> [6]. In 2007 an anonymized database published by Netflix was reverse-engineered by researchers who unmasked the preferences and opinions of 500,000 users of the platform [7].</p><p>Major players such as Microsoft, Google, and Apple have turned to differential privacy to help ensure the confidentiality of sensitive data. This attention from large tech companies has helped push differential privacy out of research labs and integrate it into application design and product development.</p><p>Differential confidentiality is now a technique adopted by SMEs (small and medium-sized enterprises) as well as by software startups because they find it a great added value.</p><figure><img alt="Example of anonymized dataset" src="https://cdn-images-1.medium.com/max/930/1*ePSZtYPsSpFW7YxYoC2vDw.png" /><figcaption>Example of an anonymized dataset</figcaption></figure><p>The differential privacy mechanism essentially adds noise (typically Gaussian or Laplacian) to the raw data to achieve a quantifiable level of confidentiality. By knowing this level, we can estimate the maximum amount of information that can be disclosed in our dataset. There are two main methods:</p><p><strong>Local differential privacy =</strong> noise is added to each individual data point in the dataset (either by one of the company employees once the data has been obtained or by individuals themselves before releasing their data. available to the company).</p><figure><img alt="Diagram of local differential confidentiality" src="https://cdn-images-1.medium.com/max/1000/1*CYws_gATIaN9IhgDlP3g7A.jpeg" /><figcaption>Diagram of local differential confidentiality</figcaption></figure><p><strong>Global Differential Privacy =</strong> Noise needed to protect individual privacy is added to the output of the raw data query.</p><figure><img alt="Diagram of Global Differential Privacy" src="https://cdn-images-1.medium.com/max/1000/1*9_fF1RgB8N1q3tN11GwAaA.jpeg" /><figcaption>Diagram of Global Differential Privacy</figcaption></figure><p>In general, Global Differential Privacy can lead to more accurate results compared to Local Differential Privacy while maintaining the same level of privacy. On the other hand, when using Global Differential Privacy, people donating their data should be confident that the receiving entity will add the necessary noise to maintain their privacy.</p><h3>Conclusion</h3><p>The alliance between the performance of AI and respect for data privacy is a difficult quest but full of opportunities and promise. It is clear that no measure applied in isolation can be completely effective in preventing abuses. So, where the decisions of intelligent algorithms are consequential, it makes sense to combine measurements at several levels to make them work together.</p><blockquote><strong>Note: This article was originally published in it’s French version </strong><a href="https://datavalue-consulting.com/ia-et-vie-privee-regles-bonnes-pratiques/"><strong>here</strong></a></blockquote><h4>References :</h4><ul><li>[1] <a href="https://gdpr-info.eu/art-4-gdpr/">Article 4, paragraph 1 of the GDPR</a></li><li>[2] <a href="https://gdpr-info.eu/recitals/no-71/">Recital 71 of the GDPR</a></li><li>[3] <a href="https://memeburn.com/2020/09/twitter-investigating-cropping-algorithm-after-users-flag-racial-bias/">Twitter cropping algorithm</a></li><li>[4] <a href="https://medium.com/sap-machine-learning-research/privacy-preserving-collaborative-machine-learning-35236870cd43">Privacy-preserving Collaborative Machine Learning;</a> by Robin Geyer, Moin Nabi, and Tassilo Klein</li><li>[5] <a href="https://ai.googleblog.com/2017/04/federated-learning-collaborative.html">Federated Learning: Collaborative Machine Learning without Centralized Training Data</a>, Google AI Blog</li><li>[6] <a href="http://www.latanyasweeney.org/work/identifiability.html">Policy and Law: Identifiability of de-identified data</a>, by Latanya Sweeney, Ph.D.</li><li>[7] <a href="https://arxiv.org/abs/cs/0610105">How To Break Anonymity of the Netflix Prize Dataset</a>; by Arvind Narayanan, Vitaly Shmatikov</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7859541c4977" width="1" height="1" alt=""><hr><p><a href="https://pub.towardsai.net/conciliating-ai-privacy-rules-and-good-practices-7859541c4977">Conciliating AI &amp; Privacy: Rules and Good Practices</a> was originally published in <a href="https://pub.towardsai.net/">Towards AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>