March 2024 - nexa - server-nexa.polito.it

L'AI e il cambio di paradigma
by Antonio March 24, 2024

March 24, 2024

Immaginate che il prossimo primo d'aprile, le BigTech facciano al mondo un bel pesce d'aprile. Si mettano d'accordo e spengano i loro server. Una disconnessione, non di un paio d'ore come è successo il 5 marzo con Facebook e consorelle [1], ma per sempre, o comunque per un periodo molto lungo. A parte lo shock iniziale, la comunità di sviluppatori open source sarebbe in grado, nel giro di poco tempo, di mettere su delle alternative valide, considerata l'enorme quantità di software libero in grado di coprire la quasi totalità dei servizi "offerti" dalle Big. Già, la quasi ... Ma con l'intelligenza artificiale come la mettiamo? Non è questione di software, l'AI è basata su una manciata di linee di codice (v. il mio nanoLLaMA [2]) disponibile in rete liberamente. Non è questione di dati in input. Common Crawl [3] è lì per questo, basta una connessione ed un hard disk decente e i dati sono pronti. E' ovviamente una questione di potenza di calcolo e di corrente elettrica, insomma, dei "chip" che riescano ad effettuare miliardi di miliardi di "moltiplicazioni" al secondo e accesi per mesi e mesi [4]. Quindi ... siamo "bloccati". Il miglior lock-in che l'industria del hardware (Nvidia) e del software è riuscita a mettere in atto. A. [1] https://notizie.tiscali.it/cronaca/articoli/anomalie-siti-meta/ [2] https://github.com/opensignature/stories [3] https://commoncrawl.org/ [4] https://www.saela.eu/flops/

5 19

La società del XXI secolo
by Antonio March 23, 2024

March 23, 2024

Noi siamo il nostro profilo Facebook? Le intelligenze artificiali sono tecnologie o soggetti sociali? Com’è possibile che il bitcoin abbia creato una crisi in Kazakistan? Come ha fatto un chatbot su Twitter ad aver imparato frasi razziste, violente e antisemite senza bisogno di programmazione? La società del XXI secolo è il prodotto di un panorama sociale che vede interconnessi persone (e le loro relazioni), dati (nella forma di informazioni e algoritmi) e tecnologie (dispositivi e servizi sempre nuovi). Le scienze sociali, da un punto di vista teorico e metodologico, ci aiutano ad analizzare questi fenomeni e a immaginare risposte. Ultimo manuale (è uscito un paio di mesi fa) di Davide Bennato. https://www.laterza.it/scheda-libro/?isbn=9788859300755 Recensione: https://www.doppiozero.com/lavoro-povero-con-vita-digitale-o-vita-povera-co… (comprende pure il commento al saggio di Alessandro Somma "Abolire il lavoro povero" [1]) [1] https://www.laterza.it/scheda-libro/?isbn=9788858153505

1 0

Nuova rivista su AI "a gratis"
by Diego Giorio March 23, 2024

March 23, 2024

Credo che possa interessare i nexiani. Però non trovo il costo dell'abbonamento, dato che immagino solo il primo numero sia in omaggio. https://aire.lexxion.eu/current_issue/AIRe Buon fine settimana D.

2 2

Can Large Language Models Reason and Plan?
by Daniela Tafani March 21, 2024

March 21, 2024

Can Large Language Models Reason and Plan? Subbarao Kambhampati Large Language Models (LLMs), essentially n-gram models on steroids that have been trained on web-scale lan- guage corpora (or, effectively, our civilizational knowledge), have caught our collective imagination with linguis- tic behaviors that no one expected text completion systems to possess. By training and operation, LLMs are perhaps best seen as giant non-veridical memories akin to an external System 1 for us all (see Figure 1). Their seem- ing versatility has however led many researchers to wonder whether they can also do well on planning and reasoning tasks typically associated with System 2 competency. Nothing in the training and use of LLMs would seem to suggest remotely that they can do any type of princi- pled reasoning (which, as we know, often involves computationally hard inference/search). What LLMs are good at is a form of universal approximate retrieval. Unlike databases that index and retrieve data exactly, LLMs, as n- gram models, probabilistically reconstruct completions for the prompt word by word–a process we shall refer to as approximate retrieval. This means that LLMs can’t even guarantee memorizing complete answers, something that is the flip side of their appeal about constructing “novel” prompt completions on the fly. The boon (“creativity”) and bane (“hallucination”) of LLMs is that n-gram models will naturally mix and match–and have almost as much trouble strictly memorizing as we do. It is indeed the very basis of their appeal. https://nyaspubs.onlinelibrary.wiley.com/doi/10.1111/nyas.15125 oppure https://arxiv.org/pdf/2403.04121.pdf

2 1

Sharenting: influencer e rischi per i minori, proposta di legge in arrivo
by 380° March 21, 2024

March 21, 2024

Buongiorno, che ignorante che sono, non immaginavo nemmeno che il fenomeno avesse un proprio nome: "sharenting". «Sharenting: influencer e rischi per i minori, proposta di legge in arrivo» 20 Mar 2024 - 11:55 https://www.orizzontescuola.it/sharenting-influencer-e-rischi-per-i-minori-… --8<---------------cut here---------------start------------->8--- La condivisione di foto e video dei figli sui social media, pratica conosciuta come “sharenting”, è un fenomeno in preoccupante aumento. Sebbene motivato dall’amore e dall’orgoglio genitoriale, questo comportamento può nascondere diverse insidie. La condivisione di foto e video dei figli sui social media, pratica conosciuta come “sharenting”, è un fenomeno in preoccupante aumento. Sebbene motivato dall’amore e dall’orgoglio genitoriale, questo comportamento può nascondere diverse insidie. = Pericoli per i minori = * Pedofilia: I contenuti pubblicati online, anche se apparentemente innocenti, possono essere utilizzati da pedofili per alimentare i loro abusi. * Cyberbullismo: Le foto e i video dei bambini possono essere usati per creare meme, insulti o altri contenuti offensivi, con gravi conseguenze sulla loro reputazione e autostima. * Diritto alla privacy: I minori non hanno la capacità di esprimere il proprio consenso alla condivisione della loro immagine online. Lo sharenting viola il loro diritto alla privacy e li priva del controllo sulla propria identità digitale. = Proposta di legge = La social media strategist e giornalista Serena Mazzini, impegnata da anni nello studio dello sharenting, presenterà alla Camera dei Deputati i dati raccolti sulle sue ricerche. L’obiettivo è quello di sensibilizzare il Parlamento sui rischi di questa pratica e di promuovere una proposta di legge per vietarla. = Misure alternative = Esistono modi alternativi per condividere i momenti speciali con i propri figli senza esporli a rischi online. Alcune semplici precauzioni: * Limitare la condivisione a reti private: Condividere foto e video solo con familiari e amici fidati, utilizzando piattaforme con impostazioni di privacy adeguate. * Nascondere il volto del bambino: Offuscare il volto del bambino o utilizzare emoji per coprire i tratti distintivi. * Non condividere informazioni personali: Evitare di pubblicare dati sensibili come il nome completo, la data di nascita o la scuola frequentata. * Sensibilizzare i figli: Crescendo, è importante educare i figli sui rischi della condivisione online e coinvolgerli nelle decisioni che riguardano la loro privacy. --8<---------------cut here---------------end--------------->8--- Per approfondimenti in merito allo "sharenting": https://en.wikipedia.org/wiki/Sharenting# Saluti, 380° -- 380° (Giovanni Biscuolo public alter ego) «Noi, incompetenti come siamo, non abbiamo alcun titolo per suggerire alcunché» Disinformation flourishes because many people care deeply about injustice but very few check the facts. Ask me about <https://stallmansupport.org>.

1 1

Affidabilità delle traduzioni automatiche
by Diego Giorio March 19, 2024

March 19, 2024

Questa è decisamente carina. E non ditemi che sono out of topic, perché riguarda l'affidabilità dei sistemi di traduzione automatica. Il paese in questione è naturalmente la cittadina di BRA, opportunamente tradotta in italiano... Saluti a tutti D.

2 1

MEMENTO | Convegno "Tra condivisione e controllo: intelligenza artificiale generativa e licenze aperte nel diritto d'autore" | 20 marzo 2024, ore 17.00
by Nexa - Media March 19, 2024

March 19, 2024

Gentilissime, gentilissimi, Vi ricordiamo che mercoledì 20 marzo, dalle ore 17.00 alle ore 19.00, si terrà il convegno organizzato dal Centro Nexa in collaborazione con l'Ordine degli Avvocati di Torino, dal titolo: "Tra condivisione e controllo: intelligenza artificiale generativa e licenze aperte nel diritto d'autore". Parteciperà all'incontro l'Avv. Deborah De Angelis (Nexa Fellow), in dialogo con il Prof. Maurizio Borghi (co-direttore del Centro Nexa) e l'Avv. Marco Ciurcina (Nexa Fellow). L'evento si terrà IN PRESENZA presso l'Aula Croce della Curia Maxima (Via Corte d'Appello 16, Torino) E ONLINE al seguente indirizzo: https://pyx-is.zoom.us/meeting/register/tZEqcOqgrTMtGdYyL0ziaS92NFo9pzKN3cw2 Per maggiori dettagli è possibile consultare la pagina web dell'incontro: https://nexa.polito.it/tra-condivisione-e-controllo Cordiali saluti, -- Valeria Bergantino Communication Officer Nexa Center for Internet & Society Politecnico di Torino - DAUIN Via Pier Carlo Boggio, 65/A - 10138 Torino web: https://nexa.polito.it/ mail: valeria.bergantino(a)polito.it<mailto:valeria.bergantino@polito.it> tel: 3473443585

1 0

Joe Biden has just dealt a big defeat to big tech | Joseph Stiglitz | The Guardian
by Alberto Cammozzo March 19, 2024

March 19, 2024

<https://www.theguardian.com/business/2024/mar/18/joe-biden-has-just-dealt-a…> Joe Biden has just dealt a big defeat to big tech | Joseph Stiglitz Joseph Stiglitz Last year, Joe Biden’s administration infuriated lobbyists representing big tech firms and others that profit from our personal data by denouncing a proposal that would have gutted domestic data privacy, online civil rights and liberties, and competition safeguards. Now, the US president’s new executive order on Americans’ data security reveals that the lobbyists had good reason to worry. After decades of data brokers and tech platforms exploiting Americans’ personal data without any oversight or restrictions, the Biden administration has announced that it will ban the transfer of certain kinds of data to China and other countries of concern. It is a small, but important, step towards protecting Americans’ sensitive personal information, in addition to government-related data. Moreover, the order is likely a precursor to additional policy responses. Americans are rightly worried about what is happening online, and their concerns extend well beyond privacy violations to a host of other digital harms, such as mis- and disinformation, social media-induced teenage anxiety, and racial incitement. The firms that make money from our data (including personal medical, financial, and geolocation information) have spent years trying to equate “free flows of data” with free speech. They will try to frame any Biden administration public interest protection as an effort to shut down access to news websites, cripple the internet, and empower authoritarians. That is nonsense. Tech companies know that if there is an open, democratic debate, consumers’ concerns about digital safeguards will easily trump concerns about their profit margins. Industry lobbyists thus have been busy trying to short-circuit the democratic process. One of their methods is to press for obscure trade provisions aimed at circumscribing what the US and other countries can do to protect personal data. It might seem obvious that a US president should protect Americans’ privacy and national security, both of which could be jeopardised depending on how and where the vast amounts of data that we all generate are processed and stored. Yet, amazingly, former President Donald Trump’s administration sought to prohibit the US from placing any restrictions on “the crossborder transfer of information, including personal information” to any country if such transfers were related to the business of any investor or service provider operating in the US or other countries signing the agreement. The Trump administration’s proposal to include this rule in the World Trade Organization did provide for one exception, which ostensibly would allow some regulation “necessary to achieve a legitimate public policy objective”, but it was designed not to work in practice. While big tech lobbyists cite the exception to rebut criticism of the broader proposal, the provision’s language comes straight from a WTO “General Exception” that has failed in 46 of 48 attempted uses. The ban on crossborder data regulation was just one of four proposals that big tech lobbyists convinced Trump officials to slip into the revised North American Free Trade Agreement and to propose at WTO-related talks. Written in arcane jargon and buried among hundreds of pages of trade-pact language, these provisions were misleadingly branded as “digital trade” rules. By prohibiting governments from adopting certain policies, the proposal’s industry-written terms threatened bipartisan efforts in the US Congress to counter big tech’s abuses of consumers, workers, and smaller businesses. They also undercut the US regulatory agencies responsible for protecting our privacy and civil rights, and for enforcing antitrust policy. In fact, had the Trump-era rules banning government restrictions on data flows gone into effect at the WTO, they would have barred the Biden administration’s own new data-security policy. Few people realised the Trump-era proposal even existed – except, of course, for the lobbyists who had been quietly commandeering trade talks. While no previous US trade pacts had included provisions pre-empting executive and congressional authority over data regulation, digital platforms suddenly would have been granted special secrecy rights. The kinds of algorithmic assessments and AI pre-screenings that Congress and executive-branch agencies deem critical to protecting the public interest would have been banned. After Trump’s loss in the 2020 election, industry lobbyists still hoped to make these anomalous rules a new norm. Their plan was to get the same provisions added to a Biden administration agreement called the Indo-Pacific Economic Framework. But instead of going with the lobbyists, Biden administration officials worked with Congress to determine that the Trump-era proposals were not compatible with congressional and administration goals for digital privacy, competition, and regulation. Now, we can understand why tech lobbyists were so incensed by the Biden administration’s decision to withdraw support for the Trump-era proposal. They recognised that by casting aside big tech’s favoured “digital trade” handcuffs, the Biden administration was reasserting its authority to regulate the large platforms and data brokers that Americans across the political spectrum think have too much power. Trade agreements have gotten a bad name precisely because of this kind of behaviour by corporate lobbyists. The US needs a robust debate about how best to regulate big tech, and over how to maintain competition while preventing the digital harms that are stoking political polarisation and undermining democracy. Obviously, the debate should not be tethered by constraints imposed surreptitiously by big tech through trade agreements. US trade representative Katherine Tai is exactly right when she says that it would be “policy malpractice” to lock in trade rules limiting action on these matters before the US government has established its own domestic approach. Whatever one’s position on the regulation of big tech – whether one believes that its anti-competitive practices and social harms should be restricted or not – anyone who believes in democracy should applaud the Biden administration for its refusal to put the cart before the horse. The US, like other countries, should decide its digital policy democratically. If that happens, I suspect the outcome will be a far cry from what big tech and its lobbyists were pushing for. Joseph E Stiglitz is a Nobel laureate in economics, university professor at Columbia University and a former chief economist of the World Bank. Ⓒ Project Syndicate

1 0

112° Nexa Lunch Seminar | 27 marzo 2024, ore 13.00
by Nexa - Media March 18, 2024

March 18, 2024

Gentilissime, gentilissimi, Vi invitiamo a partecipare al 112° Nexa Lunch Seminar che si terrà mercoledì 27 marzo, dalle ore 13.00 alle ore 14.00, con un incontro dal titolo: "'Pay or ok': you cannot be serious". Ospite dell'incontro: Marco Giraudo (Università degli Studi di Torino). L'incontro si terrà IN PRESENZA e ONLINE. SEDE FISICA dell'incontro: Campus Luigi Einaudi, Lungo Dora Siena, 100 Edificio D4, II piano, Aula H1 (MAPS<https://www.google.com/maps/place/Campus+Luigi+Einaudi/@45.0734963,7.699408…>) STANZA VIRTUALE dell'incontro: https://didattica.polito.it/VClass/NexaEvent Di seguito maggiori dettagli: https://nexa.polito.it/lunch-112 Cordiali saluti, -- Valeria Bergantino Communication Officer Nexa Center for Internet & Society Politecnico di Torino - DAUIN Via Pier Carlo Boggio, 65/A - 10138 Torino web: https://nexa.polito.it/ mail: valeria.bergantino(a)polito.it<mailto:valeria.bergantino@polito.it> tel: 3473443585

1 0

Leading adviser quits over Instagram’s failure to remove self-harm content | Instagram | The Guardian
by Alberto Cammozzo March 17, 2024

March 17, 2024

<https://www.theguardian.com/technology/2024/mar/16/instagram-meta-lotte-rub…> Leading adviser quits over Instagram’s failure to remove self-harm content A leading psychologist who advises Meta on suicide prevention and self-harm has quit her role, accusing the tech giant of “turning a blind eye” to harmful content on Instagram, repeatedly ignoring expert advice and prioritising profit over lives. Lotte Rubæk, who has been on Meta’s global expert group for more than three years, told the Observer that the tech giant’s ongoing failure to remove images of self-harm from its platforms is “triggering” vulnerable young women and girls to further harm themselves and contributing to rising suicide figures. Such is her disillusionment with the company and its apparent lack of desire to change, the Danish psychologist has resigned from the group, claiming Meta does not care about its users’ wellbeing and safety. In reality, she said, the company is using harmful content to keep vulnerable young people hooked to their screens in the interest of company profit. In her resignation letter, she wrote: “I can no longer be part of Meta’s SSI expert panel, as I no longer believe that our voice has a real positive impact on the safety of children and young people on your platforms.” In an interview with the Observer, Rubæk said: “On the surface it seems like they care, they have these expert groups and so on, but behind the scenes there’s another agenda that is a higher priority for them.” That agenda, she said, was “how to keep their users’ interaction and earn their money by keeping them in this tight grip on the screen, collecting data from them, selling the data and so on.” A Meta spokesperson said: “Suicide and self-harm are complex issues and we take them incredibly seriously. We’ve consulted with safety experts, including those in our suicide and self-harm advisory group, for many years and their feedback has helped us continue to make significant progress in this space. “Most recently we announced we’ll hide content that discusses suicide and self-harm from teens, even if shared by someone they follow, one of many updates we’ve made after thoughtful discussion with our advisers.” Rubæk’s warning comes as new research by Ofcom published last week found that violent online content is “unavoidable” for children in the UK, many of whom are first exposed when still in primary school. Among the main apps mentioned by those interviewed was Instagram. Rubæk, who leads the self-injury team in child and adolescent psychiatry in the Capital Region of Denmark, was first approached about joining the select group of experts – which has 24 publicly listed members – in December 2020. The invite came after she publicly criticised Meta, then known as Facebook, over an Instagram network linked to suicides of young women in Norway and Denmark following a documentary by Danish broadcaster DR. She agreed to join in the hope of helping to change the platform to make it safer for young people. After a couple of years of having her suggestions ignored – the original network she was critical of still exists – she came to the conclusion that the panel was just for show. Now she believes the invitation could have been an attempt to silence her. “Maybe they wanted me to be a part of them so I wouldn’t be so critical of them in the future.” In emails seen by the Observer Rubæk raised the difficulties users faced in trying to report potentially triggering images with Meta in October 2021. In correspondence with Martin Ruby, Meta’s head of public policy in the Nordics, she said she had tried to report an image of an emaciated female but received a message from Instagram saying they did not have enough moderators to look at the image, which stayed on the platform. In response, Ruby said in November 2021: “Our people are looking at it, but it is not that simple.” In the same email, he mentioned the secret Instagram network that Rubæk had originally criticised, saying that Meta was “taking a closer look”. But despite its well-documented links to suicides, Rubæk says the network remains up and running today. Rubæk’s patients tell her they have tried to report self-harm images on Instagram but they often remain. One client said that after reporting one image, it vanished, but she later saw it via a friend’s account, suggesting it had only been moved from her view. Meta “does a lot of tricks” to get around removing content, said Rubæk. “The AI is so clever, finding even the smallest nipple in a photo.” But when it comes to graphic pictures of self-harm that are proven to inspire others to harm themselves, she added, it appears to be a different story.

1 0