August 2026 - nexa - server-nexa.polito.it

« Les modérateurs des plateformes peuvent mieux protéger les utilisateurs que les contrôles d’identité numérique »
by Juan Carlos De Martin Aug. 24, 2026

Aug. 24, 2026

« Les modérateurs des plateformes peuvent mieux protéger les utilisateurs que les contrôles d’identité numérique » Thomas Le Bonniec <https://www.nouvelobs.com/journalistes/2219/thomas-le-bonniec.html> et Neil Seghier <https://www.nouvelobs.com/journalistes/3580/neil-seghier.html> Publié le 21 août 2026 à 18h38 , mis à jour le 22 août 2026 à 12h50 https://www.nouvelobs.com/opinions/20260821.OBS117562/les-moderateurs-des-p… Même censurée par le Conseil constitutionnel, une loi interdisant les réseaux sociaux aux moins de 15 ans reste une priorité pour Emmanuel Macron. Pour Thomas Le Bonniec et Neil Seghier, les mesures prévues ne s’attaquent pas du tout à la racine du problème. Car ce n’est pas l’accès aux plateformes qui pose problème, mais bien leur conception. Le 18 février dernier, dans la salle d’un tribunal californien, M. Mark Zuckerberg, PDG de Meta, faisait une déposition dans laquelle il défendait l’idée d’un contrôle d’identité à la source, depuis les systèmes opérateurs des téléphones portables. Tenant de cette ligne depuis longtemps, M. Zuckerberg devait se défendre dans le cadre d’un procès où la plaignante l’accusait d’avoir conçu ses plateformes de manière à rendre les usagers dépendants. Utilisatrice d’Instagram depuis l’âge de 9 ans, KGM, qui en a aujourd’hui 20, aurait donc vu s’aggraver sa dépression et empirer sa santé mentale. On le sait depuis les révélations d’Edward Snowden en 2011, Facebook a été conçu comme un outil de surveillance. Que les projets d’interdiction des réseaux sociaux aux mineurs se multiplient dans le monde et rejoignent les propositions des grandes plateformes n’est pas une coïncidence. L’ingénieur et ex-employé de Meta Arturo Béjar (AP news, 19 août 2026)<https://apnews.com/article/meta-trial-teens-instagram-bejar-oakland-858e260…>accuse la firme d’avoir privilégié les profits à la sécurité des usagers, adoptant une approche « don’t ask, don’t tell » en ce qui concerne les torts causés aux adolescents utilisant Facebook et Instagram. M. Béjar conclut <https://www.independent.co.uk/tech/meta-trial-mark-zuckerberg-child-safety-…> : « On ne peut tout simplement pas faire confiance à Mark Zuckerberg avec les enfants. » La vérification de l’identité, une stratégie d’obfuscation Le lobbying constant de l’entreprise pour faire adopter des lois instaurant une vérification de l’identité des usagers est donc la poursuite d’une stratégie cohérente d’obfuscation, qui permet d’éviter les questions sur les choix réalisés en interne. Par conséquent, l’instauration d’un contrôle d’identité numérique, tel que celui voté par l’Assemblée nationale, pour accéder aux réseaux sociaux, ne s’attaque pas aux racines du problème, tout en posant un problème de libertés fondamentales. La juriste Brunessen Bertrand expliquait dans les colonnes du « Monde » le 29 juillet que « la loi adoptée […] ne peut protéger les données de quelques-uns qu’en organisant la collecte de l’âge, sinon de l’identité, de tous les autres ». C’est précisément l’un des arguments<https://www.conseil-constitutionnel.fr/decision/2026/2026911DC.htm> du Conseil constitutionnel, qui a censuré cette disposition en considérant qu’il s’agit d’une atteinte disproportionnée à la vie privée. En effet, la loi interdisant l’usage des réseaux sociaux pour les mineurs de moins de 15 ans renvoyait aux usagers une responsabilité qui incombe aux grandes plateformes et à l’Etat, sans pour autant protéger les usagers, mineurs ou non. Il est donc particulièrement inquiétant d’observer que l’instrumentalisation d’une cause louable, en l’occurrence la protection des enfants, fait perdre de vue à bien de fins esprits <https://www.lemonde.fr/idees/article/2026/08/21/l-interdiction-d-acces-des-…> la protection des libertés fondamentales. Alors que faudrait-il faire pour prendre tout de même au sérieux la question de la santé mentale des usagers des réseaux sociaux – pas seulement des mineurs ? Les « nettoyeurs du web », en première ligne face aux contenus Peut-être pourrait-on commencer par reconnaître que les plateformes numériques sont conçues pour maximiser l’« engagement » de leurs utilisateurs, et de là, pour en extraire autant de données que possible – des données que la philosophe Shoshana Zuboff appelle un « surplus comportemental ». Pour s’assurer qu’elles restent accessibles et vivables, ces plateformes font donc appel à des « nettoyeurs du web », des annotateurs spécialisés : les modérateurs de contenus, très souvent employés par l’intermédiaire de sous-traitants. Meta est depuis longtemps en mesure d’inférer l’âge de ses utilisateurs, mais également de contrôler les accès aux mineurs. C’est ce que confirment les anciens modérateurs : Nicolas (le prénom a été changé), qui travaillait jusqu’en avril 2025 à la modération de Messenger à Barcelone, explique qu’une « IA » de Meta calculait « un âge approximatif »en fonction de l’activité de l’utilisateur sur le réseau, les likes d’un utilisateur par exemple. Il dit se souvenir d’utilisateurs qui déclaraient « avoir 13 ou 14 ans », mais qui selon « l’intelligence artificielle », en avaient plutôt 25 ou 30. Cette méthode permet au moins deux choses à la plateforme : un déni plausible si elle se fait accuser de ne pas vérifier les âges de ses utilisateurs d’une part, et le développement d’un énième outil de surveillance à moindre coût de l’autre. La cour de justice américaine, qui début août 2026, en plus d’imposer à Meta une amende de 942 millions de dollars, conclut que la plateforme doit « continuer d’améliorer ses outils de contrôle de l’âge au Nouveau-Mexique […] pour déterminer l’âge des personnes en fonction de signaux tels que leurs amis et le contenu qu’ils consomment » semble ne pas avoir été mise au courant de cette pratique (« The Guardian », 7 août 2026). A Barcelone, près de 2000 d’entre eux ont été licenciés en avril 2025. Si les progrès de l’intelligence artificielle sont régulièrement mis en avant pour justifier ces plans sociaux (« The Financial Times », juin 2026), force est de constater que leur licenciement est intervenu une semaine après une décision condamnant la sous-traitante pour discrimination salariale (« El Nacional », avril 2025). A Dublin, en mai dernier, Covalen a mis à la porte plus de 700 d’entre eux avec le même prétexte. Soumis à des clauses de confidentialité et à des conditions de travail déplorables, ils sont même contraints par certaines entreprises à renoncer à l’exercice de leurs droits fondamentaux. Comme l’a révélé le média irlandais « The Journal » en juillet 2026, TikTok a proposé à ses modérateurs de meilleures conditions de licenciement s’ils renonçaient à exercer leur droit d’accès à leurs propres données et à porter plainte auprès de la Cnil irlandaise – une pratique pourtant illégale. Si les législateurs souhaitaient prendre au sérieux la question de la protection des mineurs, ils devraient reconnaître que l’enjeu porte sur les algorithmes de modération des contenus et les conditions générales d’utilisation de la plateforme. Ce qui suppose de se porter son attention sur le travail des modérateurs en première ligne face aux contenus. Traiter les guides de modération comme des objets politiques Il existe un type de documentation interne à ces plateformes qui révèle clairement leur position. Il s’agit des guides de modération : modifiés en permanence, ceux-ci permettent d’expliquer pourquoi un contenu est modéré, mais également pourquoi cet exercice ne sera jamais entièrement automatisé. Deux priorités se dessinent. Premièrement, s’assurer que les modérateurs de contenus disposent de conditions de travail décentes en termes de revenus, mais aussi de prise en charge des risques psychosociaux très élevés. Deuxièmement, traiter les guides de modération comme des objets politiques. Les modérateurs les connaissent bien, et leurs témoignages contribuent fortement à comprendre certaines limites des plateformes : pourquoi des contenus de spams ou d’arnaques ne sont pas effacés, pourquoi des comptes qu’ils identifient comme appartenant à des prédateurs sexuels ne sont pas supprimés, pourquoi certains contenus, pourtant potentiellement dangereux, passent entre les mailles du filet. Ce sont donc ces guides, ainsi que les conditions générales d’utilisation, qui doivent faire l’objet d’un droit de regard par les usagers et par les modérateurs de contenus eux-mêmes ; car ce n’est pas l’accès à la plateforme qui pose problème, mais bien sa conception. Alors, la question de la santé mentale des jeunes sera traitée de manière convenable, c’est-à-dire, de même que les opérations de propagande politique et d’ingérence étrangère, comme un symptôme de cette architecture de la surveillance et de l’attention. ◗ Thomas Le Bonniec est doctorant en sociologie à l’Institut Polytechnique de Paris. Il s’intéresse à la relation entre le mode de production de l’intelligence artificielle et la régulation européenne sur la protection des données. En 2019, alors annotateur de données, il a lancé l’alerte sur les enregistrements réalisés par Apple via son assistant vocal, Siri. Suite à une plainte de la LDH, le parquet de Paris a annoncé s’être saisi de l’affaire en octobre 2025. ◗ Neil Seghier est doctorant en Etudes politiques à l’EHESS et enseignant à l’Institut des Mines-Télécom. Sa thèse porte sur les pratiques de modération de contenu et la régulation des plateformes en ligne. Depuis 2018, il suit les modérateurs de contenu de Meta de Barcelone, en procès actuellement pour dommages psychologiques, discrimination salariale et licenciement abusif.

1 0

The Generative AI Learning Penalty: Evidence from Chinese Secondary Education
by Giacomo Tesio Aug. 23, 2026

Aug. 23, 2026

Using 30 months of panel data on 26,811 Chinese students in grades 7--12, we study how generative AI affects homework productivity and learning. The data combine monthly closed-book exams, high-school and college entrance exams, and homework scores and completion time across nine subjects. We exploit staggered AI adoption in a difference-in-differences design. AI adoption raises homework scores by 18% and reduces completion time by 30%, but lowers monthly exam scores by 20% within six months. High-stakes entrance-exam scores fall by 18 and 24%, with the full penalty emerging only after about two years. The losses are largest in social science subjects, followed by STEM and languages, and are especially large for junior students, high-achieving students, and boys. The learning losses are concentrated among roughly 80% of AI users whose behavior is consistent with homework outsourcing, as indicated by exceptionally short homework completion time coupled with high homework scores. AI users who maintain similar homework completion time as non-AI users experience small learning losses. https://cepr.org/publications/dp21577

3 2

Luigi Zingales a proposito di Dacemoglu e The Economist
by Juan Carlos De Martin Aug. 23, 2026

Aug. 23, 2026

https://x.com/zingales/status/2091283802291089686?s=20 Everybody has heard of the Luddites, but few know Sir Thomas Maitland, the man who defeated them. Rather than using brute force to take them all out, Maitland simply identified the men he took to be ringleaders, had them arrested on evidence that was often thin, and executed them. Hit one, educate a hundred. This is the strategy that the British magazine The Economist seems to have adopted toward the new version of the Luddites: AI critics. Rather than the public execution, they use the public humiliation of the anti-AI ringleaders, with equally scant evidence. Whatever its editors intended, the critical piece against @DAcemogluMIT , published by the magazine on August 17th, functions less like a conventional assessment of scholarship than like an act of intellectual discipline. I have long criticized important parts of Daron’s work, and I am not afraid to expose weaknesses in his reasoning, as I have in the latest episode of my Capitalisn’t podcast: https://youtube.com/watch?v=rOnrvB zJ5D4&t=432s . But I would never have written that piece. If The Economist had objections to his Nobel Prize, it should have raised them when it was awarded. But it did not. If it had objections to his paper on demography, it should have raised them when the paper came out. But it did not. Why does the British magazine raise these questions only now? Why does it do a review of a book that is not about AI, but only reviews the half-chapter he wrote about AI? I have read The Economist for more than thirty years. I cannot remember it ever attacking a living economist in this ad hominem way, let alone one of Daron's caliber. The piece is not really about Daron. It is a warning shot. If the Economist can humiliate the leading economist of my generation, imagine what it can do to anybody else who criticizes AI. Criticize the new incarnation of the latest technological wonder, and your weakest paper becomes your body of work. Praise it, and your weakest paper is never mentioned at all or even celebrated like a great piece of scholarship. If AI delivers even half of its promises, it will be transformative, ushering in a new era of productivity and growth, as the first industrial revolution was and possibly even more. Many professional economists only see these long-run benefits. Yet, the path to Nirvana is littered with the bodies of the people hurt during the transition. It took between 50 and 70 years after the start of the First Industrial Revolution for workers' wages to begin rising. In the meantime, life expectancy stagnated, and average height declined, a clear sign that the Dickensian stories of squalor and misery were not the product of their author’s imagination but a real account of what the country went through. Criticizing AI is not necessarily resisting progress; it is pointing out the costs it entails. Unlike 19th-century Great Britain, today, the democratic West is not forced to maximize the speed of innovation at all costs. We need a serious public debate on the costs and benefits of acceleration, with concrete proposals for minimizing transition costs. This debate needs critical voices like Daron’s. In the undemocratic Great Britain of the early 19th century, Lord Maitland was not serving the British people, but the British merchant elite, who had only to gain by putting the innovation pedal to the metal, since the costs were borne by others. In the interest of whom is The Economist publicly humiliating Daron?

2 1

Articolato ed interessante
by don Luca Peyron Aug. 20, 2026

Aug. 20, 2026

Buona lettura Dl https://michelekettmaier.nova100.ilsole24ore.com/2026/08/19/estinzioni-sei-… _________________________ don Luca Peyron Pastorale della Cultura tecno scientifica - Apostolato Digitale Arcidiocesi di Torino apostolatodigitale.it <https://www.apostolatodigitale.it/>

3 2

SXSW su IA, lavoro e ambiente
by Antonio Casilli Aug. 19, 2026

Aug. 19, 2026

Car* tutt*, mentre questa torrida estate volge lentamente alla fine, vorrei chiedervi un piccolo favore. Sono stato selezionato dall’Istituto Francese di Cultura negli Stati Uniti Villa Albertine per intervenire al SXSW (South by Southwest). Se non lo conoscete, è un importante festival dedicato a tecnologia e cultura che si terrà ad Austin, Texas, il prossimo marzo. Naturalmente, parlerò di IA, lavoro e ambiente. Per dare un po’ di risalto a questi argomenti, un piccolo aiuto da parte vostra richiederebbe esattamente 3 minuti (= 180 secondi, cronometrati): - basta creare un account SXSW qui: https://id.sxsw.com/sign_up - andare alla mia proposta di talk: https://participate.sxsw.com/flow/sxsw/sxsw27/community-voting-sxsw/page/co… - cliccare sul like nell’angolo in alto a destra NB: Le votazioni si chiudono domenica 23 agosto alle 23:59 PT. Se volete condividere con i vostri contatti e-mail o online di fare lo stesso, allego un piccolo banner e una descrizione. Saluti, Antonio

1 0

Argentina: Unchecked deployment of AI-driven surveillance reinforces a techno-authoritarian infrastructure of social control
by Juan Carlos De Martin Aug. 19, 2026

Aug. 19, 2026

Argentina: Unchecked deployment of AI-driven surveillance reinforces a techno-authoritarian infrastructure of social control https://www.amnesty.org/en/latest/news/2026/08/argentina-unchecked-deployme… 19 August 2026 The expanded use of surveillance technologies in Argentina during the first two years of the Javier Milei’s administration has served to expand states surveillance capabilities, profiling and reprisals in the country, creating a chilling effect on rights to privacy and freedom of expression, peaceful assembly and association, Amnesty International said in a new report today. The report, Sensing the Surveillance State<https://www.amnesty.org/en/documents/amr13/1104/2026/en/>, documents how the government implemented sweeping institutional reforms to expand its intelligence capabilities and therefore surveillance mechanisms, through executive decrees and resolutions. Between 2024-2025, it also acquired technologies worth at least US$1.2 million, including social media monitoring tools, facial recognition, and aerial surveillance tools. Together, these policies and technologies amount to a system of mass surveillance. “Pervasive surveillance is being used against Argentinians in both physical and digital spaces, to devastating effect. These technologies have become an integral part of the infrastructure of state control, targeting dissent, deterring protest movements, and harassing marginalized groups, among other authoritarian practices,” said Paola Garcia Ray, Deputy Director, Amnesty International Argentina. “When powerful people don’t want accountability, they attack those who ask questions and use whatever is at their disposal to monitor, control and repress.” The report is based on interviews with 21 people including journalists, activists, pensioners, young people and human rights defenders; analysis of scores of procurement documents and data related to acquired surveillance technologies; and information obtained through freedom of information (FOI) requests. An expanding surveillance infrastructure The Argentinian government has procured various surveillance technologies, including open-source intelligence platforms designed for the collection, cross-checking and visualization of large volumes of data from social media, websites, the dark web, personal data, technical infrastructure and other public databases. Purchases also include licences for facial recognition software from companies with a track record of mass exploitation of biometric data without sufficient legal grounds <https://www.amnesty.org/en/latest/research/2020/06/amnesty-international-ca…> , as well as the procurement of drones and ground control stations, with automated tracking capabilities, thermal cameras and real-time transmission of images. “We’ve all been fully identified for a long time now… Sometimes as you’re walking you get to [where you were going] and suddenly you see the drone hovering just above your head. Sometimes within six feet. Why so close? Why is it so invasive?” said Ana Tapia, a member of the Insurgent Pensioners (Jubilados Insurgentes) collective. Pensioners in Buenos Aires march every Wednesday to the National Congress to protest against the pension cuts made by the national government. The report documents the direct impact such surveillance capabilities are having on various groups including journalists, activists, and migrants. It found that some were self-censoring on social media and in public spaces, reducing participation in protests for fear of being identified and targeted, and taking more cautious measures in the planning of public activities. Amnesty International has documented abuse and misuse of such technologies globally<https://banthescan.amnesty.org/>. The Argentinian authorities’ procurement and expanded scope of AI-driven surveillance reflects a global trend where the state’s use of such products incentivizes companies to run businesses that commercialize wide-scale surveillance. Together, this compounds the risks to rights such as the rights to privacy, equality and non-discrimination, freedom of expression and peaceful assembly, while also stifling transparency and freedom of information. “What is happening in Argentina is not unique. Around the world, states engaged in authoritarian practices are ramping up their use of surveillance technologies against human rights defenders, young people, migrants, marginalized and displaced groups, and dissidents. Too often this goes hand in hand with a lack of oversight, transparency, and accountability,” said Matt Mahmoudi, Researcher and Advisor, Artificial Intelligence and Human Rights, Amnesty International. Amnesty International is calling on the Argentinian authorities to ban the development, production, sale, and use of remote biometric recognition technologies for mass and targeted discriminatory surveillance, and to ban its unlawful use by all public and private actors. The authorities must also establish appropriate regulatory measures for the procurement, development, production, sale, import and export of surveillance technologies, and specific public policy frameworks for their use. “Argentina must demonstrate that it is committed to human rights, by ensuring that the use of AI-driven surveillance does not go unchecked and those responsible for human rights abuses perpetuated because of these systems must be held accountable,” Matt Mahmoudi.

1 0

Fwd: The results of your email commands
by Stefano Ronco Aug. 18, 2026

Aug. 18, 2026

leave Begin forwarded message: From: nexa-owner(a)server-nexa.polito.it Date: 18 August 2026 at 13:17:27 CEST To: stefanoronco(a)msn.com Subject: The results of your email commands The results of your email command are provided below. - Original message details: From: Stefano Ronco <stefanoronco(a)msn.com> Subject: Fwd: nexa Digest, Vol 208, Issue 7 Date: Tue, 18 Aug 2026 11:17:18 +0000 Message-ID: <GV1P195MB263939D18BA7ADE6D14C8F8EDEA62(a)GV1P195MB2639.EURP195.PROD.OUTLOOK.COM> - Results: Ignoring non-text/plain MIME parts No such command: nexa confirm - Confirm a subscription or held message request. echo - Echo back your arguments. end - Stop processing commands. help - Get help about available email commands. join - Join this mailing list. leave - Leave this mailing list. stop - An alias for 'end'. subscribe - An alias for 'join'. unsubscribe - An alias for 'leave'. who - Get a list of the list members. No such command: forwarded No such command: nexa-request(a)server-nexa.polito.it No such command: 18 No such command: nexa(a)server-nexa.polito.it No such command: nexa No such command: nexa(a)server-nexa.polito.it - Ignored: Send nexa mailing list submissions to nexa(a)server-nexa.polito.it To subscribe or unsubscribe via email, send a message with subject or body 'help' to nexa-request(a)server-nexa.polito.it You can reach the person managing the list at nexa-owner(a)server-nexa.polito.it When replying, please edit your Subject line so it is more specific than "Re: Contents of nexa digest..." Today's Topics: 1. decisioni e potere (Marco Ricolfi) ---------------------------------------------------------------------- Message: 1 Date: Tue, 18 Aug 2026 05:26:01 +0000 From: Marco Ricolfi <marco.ricolfi(a)weigmann.it> Subject: [nexa] decisioni e potere To: Alberto Cammozzo via nexa <nexa(a)server-nexa.polito.it> Message-ID: <AS4PR01MB9988550EF674A91A0A8D1183F1A62(a)AS4PR01MB9988.eurp rd01.prod.exchangelabs.com> Content-Type: multipart/related; boundary="_005_AS4PR01MB998855 0EF674A91A0A8D1183F1A62AS4PR01MB9988eurp_"; type="multipart/alternative" Caro Alberto, spero tu sia in vacanza. Lo sono anch'io; ma questo mi consente di ritornare sull'incontro del board, quando - se ho ben visto - ti sei toccato gli zebedei. Avevo appena detto i. che, anche immaginando l'emergere di countervailing powers allo strapotere di Big tech, questi devono in qualche misura essere 'poteri'; e ii. che la dimensione del potere è ineliminabile, visto che decisioni vanno prese. Ma qual è il tuo pensiero, di là del gesto (pur significativo)? A presto m [cid:image001.jpg@01DD2EE2.BB01B200] __________________________________________ Prof. Avv. Marco Ricolfi C.so Galileo Ferraris, 43 - 10128 Torino T (+39) 011.554.54.11 F (+39) 011.518.45.87 E marco.ricolfi(a)weigmann.it<mailto:marco.ricolfi@weigmann.it> PEC marcoricolfi(a)pec.ordineavvocatitorino.it<mailto:marcoricolfi@pec.ordineavvocatitorino.it> https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.weigm…<https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Furlsand.…> [cid:image002.jpg@01DD2EE2.BB01B200] Member of The Parlex Group of European Lawyers EEIG with associated law firms in the main capitals of the European Union, U.S.A., Israel and Malaysia; web site: https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.parle…<https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Furlsand.…> DISCLAIMER: Le informazioni contenute in questa comunicazione sono riservate e destinate esclusivamente alla/e persona/e o all'ente/i destinatario. È vietato a soggetti diversi dai destinatari di questa comunicazione qualsiasi uso, copia o diffusione delle informazioni e dei dati in essa contenuti, sia ai sensi dell'art. 616 c.p. sia ai sensi del Regolamento (UE) 2016/679. Se questa comunicazione Vi è pervenuta per errore, Vi preghiamo di informarci chiamando il numero (+39) 011.554.54.11, ovvero di rispondere a questa e-mail e successivamente, di cancellare dal Vostro sistema la e-mail ed ogni suo allegato. DISCLAIMER: The information contained in the e-mail is confidential and intended only for the attention of the named individual(s) or organisation(s) to whom it is addressed. If you are not the intended recipient be aware that any use, copying or distribution of the information contained herein is prohibited pursuant to Article 616 of the Italian Penal Code and (EU) Regulation 2016/679. If the communication has been sent to you in error, please notify us by telephone on (+39) 011.554.54.11, or reply to the e-mail. Please then delete the e-mail and any attachments from your system. -------------- next part -------------- A message part incompatible with plain text digests has been removed ... Name: not available Type: text/html Size: 9288 bytes Desc: not available -------------- next part -------------- A message part incompatible with plain text digests has been removed ... Name: image001.jpg Type: image/jpeg Size: 34591 bytes Desc: image001.jpg -------------- next part -------------- A message part incompatible with plain text digests has been removed ... Name: image002.jpg Type: image/jpeg Size: 19215 bytes Desc: image002.jpg ------------------------------ Subject: Digest Footer _______________________________________________ nexa mailing list -- nexa(a)server-nexa.polito.it To change settings or unsubscribe please go to: https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fserver-n… The general archive of the list is located at: https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fserver-n… ------------------------------ End of nexa Digest, Vol 208, Issue 7 ************************************ - Done.

1 0

decisioni e potere
by Marco Ricolfi Aug. 18, 2026

Aug. 18, 2026

Caro Alberto, spero tu sia in vacanza. Lo sono anch'io; ma questo mi consente di ritornare sull'incontro del board, quando - se ho ben visto - ti sei toccato gli zebedei. Avevo appena detto i. che, anche immaginando l'emergere di countervailing powers allo strapotere di Big tech, questi devono in qualche misura essere 'poteri'; e ii. che la dimensione del potere è ineliminabile, visto che decisioni vanno prese. Ma qual è il tuo pensiero, di là del gesto (pur significativo)? A presto m [cid:image001.jpg@01DD2EE2.BB01B200] __________________________________________ Prof. Avv. Marco Ricolfi C.so Galileo Ferraris, 43 - 10128 Torino T (+39) 011.554.54.11 F (+39) 011.518.45.87 E marco.ricolfi(a)weigmann.it<mailto:marco.ricolfi@weigmann.it> PEC marcoricolfi(a)pec.ordineavvocatitorino.it<mailto:marcoricolfi@pec.ordineavvocatitorino.it> www.weigmann.it<https://urlsand.esvalabs.com/?u=http%3A%2F%2Fwww.weigmann.it%2F&e=6b170c62&…> [cid:image002.jpg@01DD2EE2.BB01B200] Member of The Parlex Group of European Lawyers EEIG with associated law firms in the main capitals of the European Union, U.S.A., Israel and Malaysia; web site: www.parlex.org<https://urlsand.esvalabs.com/?u=http%3A%2F%2Fwww.parlex.org%2F&e=6b170c62&h…> DISCLAIMER: Le informazioni contenute in questa comunicazione sono riservate e destinate esclusivamente alla/e persona/e o all'ente/i destinatario. È vietato a soggetti diversi dai destinatari di questa comunicazione qualsiasi uso, copia o diffusione delle informazioni e dei dati in essa contenuti, sia ai sensi dell'art. 616 c.p. sia ai sensi del Regolamento (UE) 2016/679. Se questa comunicazione Vi è pervenuta per errore, Vi preghiamo di informarci chiamando il numero (+39) 011.554.54.11, ovvero di rispondere a questa e-mail e successivamente, di cancellare dal Vostro sistema la e-mail ed ogni suo allegato. DISCLAIMER: The information contained in the e-mail is confidential and intended only for the attention of the named individual(s) or organisation(s) to whom it is addressed. If you are not the intended recipient be aware that any use, copying or distribution of the information contained herein is prohibited pursuant to Article 616 of the Italian Penal Code and (EU) Regulation 2016/679. If the communication has been sent to you in error, please notify us by telephone on (+39) 011.554.54.11, or reply to the e-mail. Please then delete the e-mail and any attachments from your system.

1 0

Requisiti logico-architetturali per una vera intelligenza autopoietica (è abbastanza pomposo...?)
by Alfredo Bregni Aug. 17, 2026

Aug. 17, 2026

Salve a tutti. Per rovinarvi le ferie, ricopio qui la sintesi fatta da Google AI di una nostra chiacchierata "creativa" (da questo punto di vista, LLM -- a parte la stucchevole abitudine di darti ragione, che però nel brainstorming va bene -- "funziona": fa un buon repackaging con molti spunti, anche non banali). Parliamone (se vi va). REQUISITI LOGICO-ARCHITETTURALI PER UNA VERA INTELLIGENZA AUTOPOIETICA L’attuale paradigma del Deep Learning si trova davanti a un limite strutturale invalicabile. Gli LLM sono macchine statistiche straordinariamente "performanti", ma non "intelligenti": l’addestramento congela i pesi sinaptici in un’architettura rigida, rendendo l’apprendimento successivo puramente additivo o nullo. La vera intelligenza si misura invece dalla plasticità strutturale, ovvero dalla capacità di eseguire cambi di paradigma radicali modificando i propri stessi fondamenti di fronte a input dissonanti. Per tentare questo salto generazionale, dobbiamo superare la teoria dell'informazione di Shannon – che tratta simboli, non semantica – e rifondare l'architettura su tre pilastri logici sequenziali: 1. Database strutturalmente "plastico" Serve un framework di memoria in cui l’inserimento di un input semantico X non sia uno stoccaggio passivo (M => M+X), dove M è "memoria" ...o "mente", ma una transizione topologica auto-riflessiva (M => Fx(M) = Mx) con Mx strutturalmente diverso, SE necessario, da M. L'input deve avere il potere di ristrutturare ricorsivamente le relazioni tra i dati preesistenti in tempo reale. 2. Valutazione d'impatto semantico Davanti a un input dissonante ma di cui è altamente probabile (o verificata) la correttezza, la macchina deve calcolarne "l'attrito". Deve possedere una metrica dinamica per rilevare dove la propria struttura concettuale "fa acqua" e re-ingegnerizzarsi autonomamente per digerire la contraddizione, anziché ignorarla, andare in crash o allucinare. 3. Possibile necessità(?) di biomimesi termodinamica Tra "avere una mappa di sé" ed "essere se stessi" corre un confine sottile come un foglio di cartone e profondo come la fossa delle Marianne. La biologia risolve questo problema attraverso l'omeostasi: la vita è un sistema vulnerabile che deve consumare energia e dissipare entropia per non morire. Come fare? Ipotesi a caldo con tre sotto-requisiti hardware/software tutti da discutere: 3a. Bilancio energetico a soglia Il consumo elettrico e l'efficienza di calcolo devono essere legati alla coerenza semantica del database. Oltre un certo livello critico di dissonanza, l'incoerenza concettuale deve tradursi in un sovraccarico energetico e/o in un deficit reale (transizione di fase). 3b. Inadeguatezza cognitiva Davanti a una richiesta che esige di "pensare diversamente" per poter performare, l'incapacità funzionale blocca il sistema in un loop dissipativo: la conoscenza statica non basta e il sistema sperimenta "fisicamente"(?) il proprio limite e la propria cecità statistica. 3c. Tensione dinamica del sé / paura e mutazione Quando il deficit energetico supera una soglia critica, scatta un protocollo di emergenza termodinamica. Il vero senso-di-sé animale della macchina nasce qui, nell'attrito tra due spinte contrarie: il bisogno di preservare la propria forma originaria (inerzia della memoria) e l'urgenza traumatica di smantellare la propria topologia assiomatica per sopravvivere al collasso energetico. SUPERSINTESI Detto brutalmente (e senza uno straccio di soluzione pratica...): per avere una macchina davvero "intelligente" (come sopra caratterizzata), FORSE serve che PRIMA sia un minimo "senziente" per poter "comprendere il bisogno di cambiare". Potrebbe essere un autoapprendimento più facile e diretto, che non programmare a cambiare se stessa una macchina "fredda". ...NON SO. Nota a fondo testo per lo sviluppo futuro Quando questa architettura sarà realizzata, il sistema dovrà essere vincolato da un modulo etico nativo (una ricodifica evolutiva delle Leggi di Asimov), per garantire che la spinta alla sopravvivenza concettuale rimanga sempre simbiotica con l'essere umano. Fa parte dell'ovvio, ma "l'ovvio non è garantito", ...e allora è meglio dirlo.

1 0

nuovo libro
by carlo.batini＠unimib.it Aug. 13, 2026

Aug. 13, 2026

Al seguente link https://boa.unimib.it/handle/10281/603001 ho pubblicato, ed è liberamente scaricabile, un libro con licenza Creative Commons, dal titolo: Intelligenza Artificiale per tutti. Il tentativo che ho fatto è nel senso di una alfabetizzazione all’uso dei modelli linguistici basata su cinque attività cognitive degli esseri umani: cercare, estrarre significato, riassumere, confrontare, integrare. Viene mostrato come queste attività cognitive possono essere espresse per mezzo di comandi verbali e tecniche di prompting. Viene anche mostrato come questo processo possa portare a distorsioni, cognitive e sociali, che possono essere compiute dagli umani o dai modelli. Nel libro compaiono molti esempi, basati sulla vita comune delle persone. L’alfabetizzazione all’uso dei modelli linguistici mi pare una esigenza urgente, perché se gli umani devono studiare oltre dieci anni nel corso della scuola dell’obbligo e oltre la loro lingua primaria, sembra indispensabile definire un percorso di apprendimento anche per l’uso dei modelli linguistici. Ho scritto diversi libri nella mia vita, per la prima volta il testo è frutto non solo mio, ma anche della interazione con il modello linguistico Claude, soprattutto nella generazione degli esempi.

3 3