Tag Archives: Authors

Under Threat: Human Subtitle Authors Facing AI Challenges in Film

Posted on by
Reply

Is artificial intelligence poised to dismantle the SDH [subtitles for the deaf and hard of hearing] industry? While SDH remains the standard subtitle format across most platforms, the individuals behind it raise a valid concern as the sector, like many creative fields, faces increasing devaluation in the AI era. “SDH is an art; the industry often overlooks this. Many see it merely as transcription,” remarked Max Deryagin, chairman of Interface Activities, a nonprofit for freelance subtitlers and translators.

<p class="dcr-130mj7b">While AI promises to streamline subtitle creation, it misses the mark, according to Meredith Canela, a committee member. "There's a notion that AI tools mean we should work less. Yet, having spent 14-15 years in this field, I can attest that the time taken to complete projects has not changed significantly over the past five to six years."</p>

<p class="dcr-130mj7b">"Automatic transcription shows some positive advancements," Cannela adds. However, the overall efficiency does not represent a net gain compared to previous software, as extensive corrections are necessary.</p>

<figure id="8a1af689-3e30-498c-9401-81b7bc4d8a2d" data-spacefinder-role="inline" data-spacefinder-type="model.dotcomrendering.pageElements.ImageBlockElement" class="dcr-173mewl">
    <figcaption data-spacefinder-role="inline" class="dcr-fd61eq">
        <span class="dcr-1inf02i">
            <svg width="18" height="13" viewbox="0 0 18 13">
                <path d="M18 3.5v8l-1.5 1.5h-15l-1.5-1.5v-8l1.5-1.5h3.5l2-2h4l2 2h3.5l1.5 1.5zm-9 7.5c1.9 0 3.5-1.6 3.5-3.5s-1.6-3.5-3.5-3.5-3.5 1.6-3.5 3.5 1.6 3.5 3.5 3.5z"/>
            </svg>
        </span>
        <span class="dcr-1qvd3m6">"You can't overwhelm your audience"... Barbie's open caption screening for deaf and hard of hearing audiences in Westwood, California in 2023.</span> Photo: Allen J. Shaven/Los Angeles Times/Getty Images
    </figcaption>
</figure>

<p class="dcr-130mj7b">Moreover, the quality of AI-generated SDHs is often subpar, requiring significant effort to meet standards. Unfortunately, human subtitlers frequently find themselves taking on "quality control" roles with minimal compensation. Many in the field state that earning a sustainable income is currently a challenge.</p>

<p class="dcr-130mj7b">"The fees for SDH work were never great, but they've dropped to a point where it's hardly worth the effort," says Rachel Jones, an audiovisual translator and committee member. "This seriously undermines the value we provide."</p>

<p class="dcr-130mj7b">This value is crucial. "We're thrilled to welcome Teri Devine, associate director of inclusion at the Royal National Institute for Deaf and Deaf," a representative stated. “For those who are deaf or hard of hearing, subtitles are an essential service."</p>

<aside data-spacefinder-role="supporting" data-gu-name="pullquote" class="dcr-19m4xhf">
    <svg viewbox="0 0 22 14" style="fill:var(--pullquote-icon)" class="dcr-scql1j">
        <path d="M5.255 0h4.75c-.572 4.53-1.077 8.972-1.297 13.941H0C.792 9.104 2.44 4.53 5.255 0Zm11.061 0H21c-.506 4.53-1.077 8.972-1.297 13.941h-8.686c.902-4.837 2.485-9.411 5.3-13.941Z"/>
    </svg>
    <blockquote class="dcr-zzndwp">The same sound can mean a million different things. As humans, we interpret how it should feel.</blockquote>
</aside>

<p class="dcr-130mj7b">Deaf and hard of hearing communities are diverse, meaning subtitles must accommodate various needs in crafting SDH. Jones explains, "While some believe that naming songs in subtitles is pointless, others might resonate with it because of the song's title."</p>

<p class="dcr-130mj7b">Subtitles involve numerous creative and emotion-driven choices—qualities AI currently lacks. When Jones first watches a show, she notes her emotional reactions to sounds and determines how best to express those in words. She then decides which sounds to subtitle and which may be excessive: "You can't overwhelm the audience," she points out. It's a delicate balancing act. "I want to avoid over-explaining everything to the viewers," Cannela adds.</p>

<figure id="880e1917-c3ac-492b-829a-737f8a57f715" data-spacefinder-role="supporting" data-spacefinder-type="model.dotcomrendering.pageElements.ImageBlockElement" class="dcr-a2pvoh">
    <figcaption data-spacefinder-role="inline" class="dcr-9ktzqp">
        <span class="dcr-1inf02i">
            <svg width="18" height="13" viewbox="0 0 18 13">
                <path d="M18 3.5v8l-1.5 1.5h-15l-1.5-1.5v-8l1.5-1.5h3.5l2-2h4l2 2h3.5l1.5 1.5zm-9 7.5c1.9 0 3.5-1.6 3.5-3.5s-1.6-3.5-3.5-3.5-3.5 1.6-3.5 3.5 1.6 3.5 3.5 3.5z"/>
            </svg>
        </span>
        <span class="dcr-1qvd3m6">"Algorithms cannot replicate the level of professional work."</span> Photo: Milan Sulkara/Arami
    </figcaption>
</figure>

<p class="dcr-130mj7b">AI struggles to discern which sounds are crucial. "It’s far from achieving that now," Deryagin notes, emphasizing the importance of understanding the broader context of a film rather than just individual images or scenes. For instance, in *Blow Out* (1981), a mysterious sound recurs, enhancing viewers' understanding of the main plot points. "SDH must create these connections rapidly without over-informing the audience initially," he explains. "The same sound can have countless meanings, and as a human, it’s my job to interpret those nuances."</p>

<p class="dcr-130mj7b">"You can't simply feed an algorithm a soundtrack and expect it to get it right. Providing metadata will not bridge the gap to professional quality."</p>

<p class="dcr-130mj7b">Netflix provided a glimpse into its "SDH process" following the subtitles for *Stranger Things*—for example, "[Eleven pants]" or "[Tentacles squelching wetly]"—in an <a href="https://www.netflix.com/tudum/articles/stranger-things-season-4-captions" data-link-name="in body link">interview with the subtitler</a>. The company chose not to comment further on AI in subtitle production. The BBC informed the *Guardian* that "we do not use AI for TV subtitles," though much of that work was outsourced to Redbee Media last year. <a href="https://www.redbeemedia.com/news/red-bee-medias-artificial-intelligence-captioning-workflows-bring-costs-down-for-network-10/" data-link-name="in body link">A statement was issued</a> regarding the use of AI for creating SDHs for the Australian Broadcasting Network 10.</p>

<p class="dcr-130mj7b">Jones notes that linguists and subtitlers aren't inherently opposed to AI, but at this juncture, it complicates rather than simplifies their work. "In every industry, AI tends to replace the creative aspects that bring us joy, rather than alleviating the tedious tasks that we’d rather avoid," she concludes.</p>

Source: www.theguardian.com

MetaExpose Authors Risk Bankruptcy Following Company Criticism Ban | Meta

Posted on by
Reply

The former Meta executive, who authored a provocative book highlighting social media companies’ interactions with China and their treatment of teenagers, is reportedly facing bankruptcy after its release.

Lawmakers in Congress have contended that Mark Zuckerberg’s company is trying to “silence and punish” Sarah Wynn Williams, the former director of global public policy at Facebook, Meta’s predecessor.

Former Labor Transport Secretary Louise Hayes stated that Wynn Williams may incur a fine of $50,000 (£37,000) for each breach of an order obtained by Meta.


In her book, Eardaling People, published this year, Wynn-Williams made several claims regarding the conduct and culture of social media firms, including allegations of sexual harassment that the company denied. She asserts that her dismissal was due to “poor performance and toxic behavior.”

Nevertheless, the former diplomat has been prohibited from publishing memoirs after Meta secured a ruling against her. She later testified before the US Senate Judiciary Subcommittee, claiming Meta collaborated “with gloves” with Beijing regarding censorship tools.

Pan Macmillan, which published the memoir, reported over 150,000 copies sold across all formats. The book was also recognized as a Sunday Times bestseller in Hardback for 2025, with a paperback edition due for release early next year.

Haigh pointed out Wynn-Williams’ situation during a House of Representatives debate on employment rights on Monday, asserting that her decision has led to significant financial jeopardy.

“Despite previous official statements indicating that Meta had ceased using NDAs [non-disclosure agreements] in cases of sexual harassment,” she noted, “Sarah is being pushed towards financial ruin within the UK arbitration system.

“Meta has given Sarah a disturbing order and is gearing up to impose a $50,000 fine for any violations. She is on the brink of bankruptcy, and I am confident that the home and government will push this legislation to protect individuals with moral courage.”

It’s understood that the $50,000 figure pertains to damages Wynn-Williams must pay for violating a separation agreement she signed when leaving Meta in 2017, with Meta asserting that she voluntarily agreed to the terms.

Mehta indicated that, as of now, Wynn-Williams has not been compelled to adhere to the contract.

The company refrained from commenting on Hayes’ intervention. Senate testimony from Wynn-Williams previously asserted that the company has been “disconnected from reality” and is plagued by false claims.

Meta characterized the book as “an outdated, previously reported compilation of company claims and unfounded allegations against executives.” She claimed she was dismissed for “poor performance and toxic behavior,” with investigations concluding that she made misleading harassment allegations.

The ruling that barred her memoir’s publication affirmed that “the false narrative should never have seen the light of day.”

The order dictated that Wynn-Williams must halt promotion of the book and minimize any further publications, though no actions were mandated against Pan Macmillan.

Since her Senate hearing in April, Wynn-Williams has remained publicly silent. In a statement this month, she expressed gratitude for the continued investigation into Meta’s actions by the US Senate.

“I wish I could elaborate,” she stated. “I urge other tech employees and potential whistleblowers to share their insights before more harm comes to children.”

Her attorney mentioned that Wynn-Williams “will remain silent regarding the matters currently under Congressional investigation.”

Source: www.theguardian.com

Famous Authors Sue Microsoft Over AI Training Using Their Books

Posted on by
Reply

A coalition of authors has accused Microsoft of utilizing nearly 200,000 pirated copies to develop an artificial intelligence model. This accusation adds to the ongoing legal struggles surrounding copyright issues between creative professionals and tech companies.

Kai Bird, Jia Tolentino, Daniel Okrent, and others argue that Microsoft intends to use a well-known digital version of their book to train Megatron AI for generating responses to user queries. Their lawsuit, filed in federal court in New York on Tuesday, is among several crucial cases initiated by authors, news outlets, and other copyright holders against tech firms regarding alleged misuse in AI training.

The authors are seeking a court order to prohibit statutory damages of up to $150,000 for each work that Microsoft is accused of misusing.

Generative AI products like Megatron can produce text, music, images, and videos based on user input. To develop these models, software engineers gather expansive databases of media and train AI to produce similar outputs.

The authors claim that Microsoft has utilized a trove of nearly 200,000 pirated books for training Megatron, which generates text responses to prompts. The complaint states that Microsoft employed these pirated datasets to “build not only computer models from the works of numerous creators and authors but also to produce a variety of representations replicating the syntax, sound, and themes of the copyrighted works.”

A Microsoft representative has yet to respond to inquiries about the lawsuit, while the authors’ attorney declined to comment.

This lawsuit against Microsoft was filed just after a federal judge in California ruled that the use of copyrighted material for AI training could be considered fair use, but acknowledged that they might still be liable for the utilization of pirated book versions. This marked the first US legal decision addressing the legality of using copyrighted materials without authorization for AI training. On the same day the complaint against Microsoft was filed, a California judge ruled in favor of Meta in a similar copyright dispute, attributing the decision more to the plaintiff’s weak argument than to the strength of the tech company’s defense.

The conflict over copyright and AI emerged soon after the launch of ChatGPT, encompassing various forms of media. The New York Times has taken legal action against OpenAI for copyright infringement related to article archives. Similarly, Dow Jones, the parent company of the Wall Street Journal and the New York Post, has filed a lawsuit against the perplexed AI. Major record labels are pursuing legal action against companies producing AI music generators. Getty Images has also sued Stability AI concerning a startup’s text-to-image product. Just last week, Disney and NBC Universal initiated legal proceedings against Midjourney, a company operating popular AI image generators that are believed to misuse iconic film and television characters.

Tech companies argue that being compelled to use copyrighted materials fairly to create new, transformative content and to compensate copyright holders could hinder the burgeoning AI industry. OpenAI CEO Sam Altman has stated that the development of ChatGPT was “impossible” without incorporating copyrighted works.

Source: www.theguardian.com

Meta Prevails in AI Copyright Lawsuits as US Ruling Favors Company Over Authors

Posted on by
Reply

Mark Zuckerberg’s Meta has secured judicial backing in a copyright lawsuit initiated by a collective of authors this week, marking a second legal triumph for the American Artificial Intelligence Industry.

Prominent authors, including Sarah Silverman and Ta-Nehisi Coates, claimed that the owners of Facebook utilized their books without authorization to train AI systems, thereby violating copyright laws.

This ruling comes on the heels of a decision affirming that another major AI player, Humanity, did not infringe upon the authors’ copyrights.

In his ruling on the Meta case, US District Judge Vince Chhabria in San Francisco stated that the authors failed to present adequate evidence that the AI developed by tech companies would harm the market to establish an illegal infringement under US copyright law.

However, the judgment offered some encouragement to American creators who contended that training AI models without consent was unlawful.

Chhabria noted that using copyrighted material without permission for AI training is illegal in “many situations,” contrasting with another federal judge in San Francisco who recently concluded in a separate case that Humanity’s AI training constituted “fair use” of copyrighted works.

The fair use doctrine permits the utilization of copyrighted works under certain conditions without the copyright holder’s permission, which serves as a vital defense for high-tech firms.

“This ruling does not imply that Meta employs copyrighted content to train language models,” Chhabria remarked. “It merely indicates that these plaintiffs presented an incorrect argument and failed to establish a supportive record for their case.”

Humanity is also set to face further legal scrutiny this year after a judge determined that it had illegally utilized over 7 million books from the Central Library, infringing on the authors’ copyrights without fair use.

A representative for Boys Schiller Flexner, the law firm representing the authors against Meta, expressed disagreement with the judge’s ruling to favor Meta despite the “uncontroversial record” of the company’s “historically unprecedented copyright infringement.”

A spokesperson for Meta stated that the company valued the decision and characterized fair use as a “critical legal framework” for developing “transformative” AI technology.

In 2023, the authors filed a lawsuit against Meta, asserting that the company exploited unauthorized versions of their books to train the AI systems known as Llamas without consent or remuneration.

Copyright disputes are placing AI firms in opposition to publishers and creative sectors on both sides of the Atlantic. This tension arises because generative AI models, which form the foundation of powerful tools like ChatGPT chatbots, require extensive datasets to be trained, much of which is comprised of copyrighted material.

Skip past newsletter promotions

This lawsuit is part of a series of copyright cases filed by authors, media organizations, and other copyright holders against OpenAI, Microsoft, and companies like Humanity regarding AI training.

AI enterprises claim they are fairly using copyrighted materials to develop systems that create new and innovative content, while asserting that imposing copyright fees on them could threaten the burgeoning AI sector.

Copyright holders maintain that AI firms are unlawfully replicating their works and generating rival content that jeopardizes their livelihoods. Chhabria conveyed empathy toward this argument during the May hearing, reiterating it on Wednesday.

The judge remarked that generative AI could inundate the market with endless images, songs, articles, and books, requiring only a fraction of the time and creativity involved in traditional creation.

“Consequently, by training generative AI models with copyrighted works, companies frequently produce outputs that significantly undermine the market for those original works, thereby greatly diminishing the incentives for humans to create in the conventional manner,” stated Chhabria.

Source: www.theguardian.com

Authors in London protest Meta’s theft of book and use of ‘Shadow Library’ to train AI

Posted on by
Reply

A demonstration will be held today outside Meta’s London office by authors and other publishing industry experts protesting the organization’s use of copyrighted books for training artificial intelligence.

Notable figures like novelists Kate Moss and Tracy Chevalier, poet Daljit Nagra, and former chairman of the Royal Literature Society, are expected to be present outside Meta’s Kings Cross office.

Protesters will gather at Granary Square at 1:30 pm, with hand-written letters to Meta by the Authors Association (SOA) planned for 1:45 pm, also to be sent to Meta’s US headquarters.

Earlier this year, Meta CEO Mark Zuckerberg allegedly approved the use of Libgen, known as the “Shadow Library,” which contains over 7.5 million books. The Atlantic recently released a searchable database of the titles in Libgen, suggesting that authors’ works may have been used to train Meta’s AI models.

SOA Chair Vanessa Fox O’Loughlin condemned Meta’s actions as “illegal, shocking, and devastating for writers.”

Vanessa added, “Books take years to write, and Meta stealing them for AI replication threatens authors’ livelihoods.”

In response, a Meta spokesperson claimed they respect intellectual property rights and believe their actions comply with the law.

Skip past newsletter promotions

Several prominent authors, including Moss, Richard Osman, Isiguro Kawako, and Val McDermid, signed a letter to Culture Secretary Lisa Nandi asking for Meta executives to appear before Congress. The petition garnered over 7,000 signatures.

Today’s protest is led by novelist AJ West, who expressed dismay at seeing their work in the Libgen database without consent.

A court filing in January revealed a group of authors suing Meta for copyright infringement, noting the impact on authors’ rights by using unauthorized databases like Libgen.

SOA’s chief executive Anna Gunley emphasized the detrimental effect of companies exploiting authors’ copyrighted works.

Protesters are encouraged to create placards and use hashtags like #MetaBookThieves, #DothewRiteThing, #MakeItfair.

Source: www.theguardian.com

Australian authors suggest Meta might have used their book to train AI without permission

Posted on by
Reply

The Australian author expresses being “lively alive” and feels violated knowing their work was allegedly included in a pirated dataset used to train AI.

Parents company of Facebook and Instagram faces a copyright infringement lawsuit from US authors like Ta-Nehisi Coates and comedian Sarah Silverman.

In a court application from January, CEO Mark Zuckerberg reportedly approved using the book’s online archive, Libgen Dataset, to train the company’s AI models, despite warnings from the AI executive team of its pirated nature.

In the Atlantic, Searchable databases have been released for authors to check if their work is in the Libgen Dataset.

Books by notable Australian authors, including former Prime Ministers Malcolm Turnbull, Kevin Rudd, Julia Gillard, and John Howard, are among those published.

Holden Sheppard, author of Invisible Boys, a popular young adult novel adapted to a Stan series, expressed disappointment that his work was utilized in training meta AI.

He expressed his disapproval of his books being used without consent to train generative AI systems, considering it unethical and illegal and calling for fair compensation for the authors.

He emphasized the need for AI-specific laws in Australia to ensure compliance with existing copyright laws by generative AI developers or deployers.

Journalist and author Tracey Spicer discovered two of her books, including one that addresses artificial intelligence, were included in the dataset without her consent.

She called for a class-action lawsuit in Australia and urged affected authors to contact local federal lawmakers.

Skip past newsletter promotions

She criticized big technology companies for profiting while reducing writers to a serf-like status, highlighting the financial struggles of many authors.

…

Alexandra Heller-Nicholas, an award-winning film critic and author of several books, expressed her frustration and called for government action.

The Australian Authors Association urged Facebook to advocate for authors whose work was used without permission.

Society Chair Sophie Cunningham contacted affected authors and condemned the treatment of writers by large companies profiting from their work.

Cunningham criticized Meta’s dealings with writers as exploitative and called for fair treatment and compensation for authors.

Mehta declined to comment on the ongoing lawsuit and is reportedly lobbying for AI training on copyrighted data via executive orders.

Previously, Melbourne publisher Black Inc. Books raised concerns about the use of AI in the industry, with some companies entering agreements with publishers for content use.

Source: www.theguardian.com