Month: March 2024

After Blocking News in Canada, Meta Challenges Australia (Again)

Image: Shutterstock via AI modification

It was inevitable. After Meta pulled the plug on news content on its platform in Canada as its way of complying with the obligations of the Online News Act, Australia, the model that Canada sought to emulate, was surely next in line. On March 1, Meta announced that it plans to stop paying publishers of news content in Australia, and will not renew its current agreements with Australian media once they expire. Most will expire this year.

Canada had modelled its Online News Act (Bill C-18) on Australia’s News Media Bargaining Code, albeit with “improvements”. Rod Sims, who was head of the Australian Competition and Consumer Commission (ACCC) at the time the Commission designed the Code (later incorporated into legislation as the Treasury Laws Amendment (News Media and Digital Platforms Mandatory Bargaining Code) Act 2021), was invited to testify before the Canadian Parliamentary committee examining Bill C-18. In his testimony, Sims talked about the success of the Code, its benefits for not just large media players but also many smaller “country” outlets, estimating the benefits to be north of A$200 million per year to journalism in Australia. He added that the institution of the Code “has transformed the journalism landscape in Australia. It’s gone from pessimism to optimism.”

Inspired by the results of the Australian legislation (which, by the way, ended up not designating either Google or Facebook under the Code, since they managed to come to sufficient “voluntary” agreements with Australian media), Canada moved ahead, basing its legislation on the Australian law but adding a couple of additional features. One was to increase transparency with regard to deals that would be struck under the law. Another was to require self-designation by platforms (while making it apparent that only Meta/Facebook and Google) met the criteria, allowing them an exemption if they reached acceptable deals with media. In this way, the companies could not avoid designation and would be subject to the law, something they strongly opposed, even though both had already engaged in voluntary programs on their own terms to provide some financial support to selected media outlets.

Just as happened in Australia, (see “Google’s Latest “Stoush” with Australia: What’s the Lesson from Germany’s Failed Effort? and “Facebook in Australia: “READY, FIRE, AIM”) both platforms pushed back strongly against the draft legislation, threatening to block news for Canadian users. (Facebook briefly and disastrously blocked news for Australian users during its campaign against the Code, but ultimately backed down). First, in the fall of 2022 Facebook said it might have to block postings of news on its Canadian platform, followed by Google which  threatened to block search for Canadian news in Canada by Canadian users. By the summer of 2023, when the Online News Act became law without any of the amendments proposed by the platforms, Meta upped the ante by declaring that it would end news availability on Facebook and Instagram for all users in Canada prior to the Act taking effect, set for December 2023. Again, just as in Australia, Canadian government leaders were public in their condemnation, accusing Meta of threatening and irresponsible behaviour. Alas, it was all to no avail. It appears Meta had already made its decision to not provide financial support for news content in Canada, and to end the few existing agreements that it had undertaken in the past. At the time, it indicated it would also be taking similar action elsewhere. Rather than submit to the legislation by negotiating with media entities, it complied (in letter if not in spirit) by blocking links to Canadian media. Negotiations with Google continued and eventually a compromise of sorts was reached whereby Google agreed to contribute to a fund which would be used to support journalism in Canada.

This was a somewhat pyrrhic victory (the fund will be about $100 million, less than half what had previously been estimated), but a victory nonetheless in the eyes of at least some of the news media. One can debate the overall success of the legislation (see MediaPolicy.ca’s “The Online News Act is law: a buzzer-beater win or epic miscalculation?”), but along with more government financial support, the Google funded pot will be welcomed by many smaller media outfits. Ironically, establishing a fund rather than requiring negotiations between the platforms and media for payment for content was an early proposal by some commentators. Now this has come to pass more by accident than design. Criteria for disbursing from the fund have been tweaked so that broadcast media, and in particular the CBC, who employ the bulk of news journalists in the country, get less than their proportional share would otherwise indicate.

The lesson for Canada, and now for Australia, is that the big digital platforms will not hesitate to play hardball if they feel their global interests are threatened. While Australia, followed by Canada, was first off the mark with legislation designed to level the playing field between a stressed journalism sector and the monolithic platforms, the response of the platforms was governed more by potential precedent than the specifics of those markets. The existence of draft legislation at the federal level in the US, (the Journalism Competition and Preservation Act, aka JCPA) as well as at the state level in California and Illinois, has not escaped the attention of Meta and Google. (Even the watered-down compromise settlement that Google made with Canada has led to some lip-smacking speculation in the US as to the amount of funding that could flow to US media). It appears that Meta, in the face of cost cutting and loss of market share in 2022, had made a business decision that if it had to pay for access to news content, it would do without. To what extent this is a wise business decision remains to be seen, but the company has clearly made a business decision in this regard. This decision may or may not affect Meta’s bottom line, but it will have the effect of leaving the platform as a purveyor of less than reliable information from nonprofessional sources. However, doing the most socially responsible thing as opposed to maximizing profits by cutting costs is not what Meta is about.  Having made its decision, it will need to unwind its commitments to Australian media, which it is now in the process of doing.

What does this mean for Australia and what can the Australian government do about it? Writing about this in Canada’s National Post, Rod Sims, now professor at the Crawford School of Public Policy at ANU, outlined some choices the Australian government needs to face. It could move to designate Meta under the Bargaining Code and force it into the negotiation and arbitration process. That would likely lead to Meta taking precisely the action that it took in Canada. The government could amend the legislation, but to what end? It could publicly criticize Meta, accusing it of unfairness and bad behaviour. It has already done this, with Prime Minister Albanese saying that what Meta is doing is “not the Australian way”. That will have zero influence on Mark Zuckerberg and the people who run Meta.

At the end of the day, Australia can stand up to Meta, and let the chips fall as they may, or it can allow Meta to free ride on Australian news content, accepting that there may be social benefits in allowing this to happen. A recent report by the Australian Broadcasting Commission (ABC) points out that Facebook is the largest social media platform for general news and half of Facebook’s users in Australia report using the social media platform for news. (Regardless of this, Meta’s beancounters give news no value to the platform). According to a University of Canberra report cited by ABC, 45% of Australians get their news from social media as opposed to less than 20% from print sources. The largest source of news is still TV at 58%. (The numbers are greater than 100 because many consumers get their news from more than one source).  In one sampling, 14% of Australians got their news from Instagram! While I find this personally appalling (indirectly revealing my age), that is the reality of our society today. Better that consumers find reliable, curated news somewhere–but we still need to recognize that responsible journalism needs to be paid for. Meta, apparently, has no desire to be a part of that equation. Without the infusion of responsible, curated journalism, Facebook will become an even greater home for misinformation than it already is. But does Meta care? Clearly not. Consumers need to be encouraged to find their news sources elsewhere. Easier said than done.

The Australian government is no doubt pondering how to respond in the best interest of Australia. Allowing Meta to wriggle out from its obligations under the Bargaining Code would not necessarily undermine the deals struck with Google, who appears to have accepted that its overall interest is best served by some form of accommodation. Having Microsoft, which has publicly stated it is willing to subject itself to both the Australian and Canadian legislation, breathing down its neck is undoubtedly a factor in this. Even if the Google deal won’t be undone, it is still galling that Meta can get away with it. Canada had to swallow that reality, yet stood up to Meta. What will Australia do? It’s a tough call.

© Hugh Stephens, 2024. All Rights Reserved

DMCA Copyright Infringement? The Perils of Relying on AI

Image: Shutterstock

My curiosity was piqued by the email that popped up in my inbox. “DMCA Copyright Infringement Notice”. Should I open it or ignore it? I took the plunge. Immediately my sphincter puckered, and a frisson went down my spine as I read it was from one Alicia Weber, Trademark Attorney, with Nationwide Legal Services in Austin, TX. She stated that she represented the Intellectual Property division of Claude AI UK and Nationwide Legal had identified an image belonging to their client on this very blog. If I did not take specified action within 5 business days, legal proceedings would be undertaken against me under DMCA Section 512(c) guidelines. What had I done? As one might expect for a copyright blog, I strive to be meticulously careful and to follow the copyright rules, but images can be tricky. I have found that the safest thing is to make your own, or license one from a service like Shutterstock. However on occasion, when Shutterstock does not have what I want, I have resorted to using a Creative Commons (CC) image off the web, being careful to comply with its terms.

Complying with usage terms is really important for works covered by CC licences, as the Independent Journal Review, a US news website, has just learned. In a recent case, the website used a photograph offered through a Creative Commons licence by photographer Larry Philpot, but failed to attribute the work to the author as required by the terms of the licence. There was no requirement to pay for using the photo, just to provide attribution in the correct format. The law firm blog that discussed this case does not say what the damages were, but there have been a number of cases where users who did not comply with the full terms of a Creative Commons licence have found themselves sued for, and sometimes paying out, substantial sums. There is even a “business model”, if you can describe it as such, that consists of putting out images on the web covered by a complicated CC licence, such as the original CC 2.0 licence, and then going after users for statutory damages for willful infringement if even the slightest mistake is made in attribution. This practice is commonly referred to as “copyright trolling”. While I am not passing judgment one way or the other, this website claims that the plaintiff in the Independent Journal Review case fits that description, although in that case the attribution requirements did not seem to me to be particularly onerous.

But back to Nationwide Legal Services of Austin, TX. Did I make an error of attribution? Since the image I used was licensed from Shutterstock, and was indicated as such on the blogpost, clearly not. What then was the problem? According to the complaint, this blog post of mine used an image belonging to Claude AI. To wit, this image.

But wait (as the TV ads say), it is clearly not the same image. Yes, it has some superficial similarities to the one I licensed from Shutterstock. Both have an image of a robot holding a logo outlined in red, although not the same robot. In the case of the image I used, the robot is holding the internationally recognized symbol for copyright, ©. See above. The robot in the Claude AI image appears to be holding a stylized image of a Euro. How could the two have been confused? Did any human actually look at these two images and compare them? If so, they would have seen that they are clearly different. Did one take inspiration from the other? I don’t know, but Claude AI’s image could just as easily have been based on the one offered by Shutterstock or vice versa. So, what happened?

More than likely the supposed match between the two was identified by an AI powered web crawler that scanned thousands of images on the web and then compared them to images held in Claude AI’s repertoire, but apparently no one bothered to actually verify the match. It was easier to simply send out an (automated?) letter demanding action. In this case, although the legal language was threatening, the demand was not for money. All that was asked was that I credit the company by adding a direct and clickable hyperlink to https://claudeai.uk/,  either beneath the image or in the footer of the page. This link leads to a webpage that proudly proclaims, “Maximizing Efficiency with Claude AI: The Future of Workplace Automation”. Claude AI is touted as “an artificial intelligence-powered platform that can automate various tasks in the workplace.”

Like identifying infringing images and sending legal notices?

Well, No-one said AI was perfect.

I politely responded to Ms. Weber, pointing out the discrepancy. So far, I have not heard back. I hope I don’t.

© Hugh Stephens, 2024. All Rights Reserved.

Japan’s Text and Data Mining (TDM) Copyright Exception for AI Training: A Needed and Welcome Clarification from the Responsible Agency

A news headline that reads “AI Guidelines”

Image: iStock

Japan has always been known for its strong creative sector and rich cultural output, from animé to manga to literature, music and film, and for its respect for intellectual property (IP) and the rights of creators. Compared to some of its neighbours in the region, it has been a pillar of respect for IP. In the middle of last year, however, this image was sharply challenged through the interpretation, or misinterpretation, of remarks made to a Diet committee by then Minister of Education, Culture, Sports, Science and Technology (known by its acronym of MEXT), Keiko Nagaoka, with respect to Japan’s Text and Data Mining (TDM) exception. The TDM exception had been introduced in 2018. It was widely, but incorrectly, reported (for example, here and here) that the Minister had stated Japan would not enforce copyrights on data used for AI training. To the dismay of creators everywhere, this statement was pumped up by advocates for the AI industry as an example for others to follow in the competitive race to develop generative AI, even if it meant throwing rights-holders under the bus. But this is not what really happened. To clear up any confusion, Japan’s Agency for Cultural Affairs (ACA), an entity that is part of MEXT and which (as part of its mandate) manages the Copyright Office, has just published a draft discussion and consultation paper on AI and copyright. To understand the true situation regarding TDM for AI training in Japan, and the Japanese government’s position on this issue, read on.

The misinformation about Japan’s position on AI and copyright can be traced back to Article 30(4) of the Copyright Law, the 2018 amendment introduced to deal with text and data mining. At the time, it did not attract much attention, but has come into prominence with the explosion of data mining for AI development. This section permits the unlicensed use of copyrighted data for the purpose of testing, data analysis or data processing. Notably (and unfortunately) it does not make any explicit distinction between legally accessed and non-legally accessed materials, unlike the TDM provisions in the EU, the UK and Singapore. In other words, it does not explicitly prohibit the use of pirated content.

At first blush, Section 30 (4) appears to be the proverbial loophole in copyright protection through which you could drive the generative AI truck. That, however, is not the case despite misunderstandings regarding Minister Nagaoka’s comments.  The provision carefully distinguishes between works where the end use is simply for data analysis and processing purposes, and uses where, according to the English translation of the Japanese law, there is a degree of “enjoyment” of the work by the user, in which case the exception does not apply. We will come back to the meaning of “enjoyment”, as this is a key part of the story. In addition, the use of the work must not “unreasonably prejudice the interests of the copyright owner”. Many readers will recognize this wording as the third element of the Berne Convention “three step test” that governs exceptions to copyright.

Now, let’s look at the meaning of “enjoyment” in the context of Article 30(4). Unfortunately, I don’t speak Japanese, so I don’t know what specific Japanese term is being rendered into English as “enjoyment”. The English term is likely not an exact equivalent. However, the meaning of “enjoyment” in Japanese law is critical as it is the key concept that separates a benign use of data in its most generic form from data where there is perception of, or access to, the actual content (i.e. the copyrighted expression of an idea, translated as “the thoughts and sentiments expressed in the work”). “Enjoyment” also encompasses the idea of benefit or beneficial use. One definition mentioned in the ACA’s discussion paper describes enjoyment as “the act of accepting, savoring, and enjoying something that is mentally excellent or materially beneficial”. According to the Agency’s discussion paper, if the user of the copyrighted data derives benefit from the content, such as by creating output based on that content, and thus creates “a product that allows one to directly sense the essential characteristics of the expression of the copyrighted work of the learning data…”, then “enjoyment” exists. And importantly, if enjoyment exists, the TDM exception to copyright in Article 30(4) does not apply.

Given that at least some AI generated works clearly replicate and display the essential characteristics of works they were trained on, it is clear the claim that Japan has provided a blanket copyright exception for AI training is miles off-base. This conclusion seems to have escaped those who loudly proclaimed last year that Japan had opened the taps to permissionless copying of copyrighted works for AI development purposes. The potential lack of clarity, compounded by misinterpretation of Minister Nagaoka’s remarks, has led to the drafting of the ACA discussion paper, and launch of a public consultation. It is a welcome development.

The latest draft of this paper, dated January 2024, helps clarify a number of questions related to the Article 30(4) TDM exception. The paper makes it clear that far from declaring open season on its creative sector, Japan has defined its TDM exception carefully and narrowly. However, it is not without its problems, notably the omission of any reference to a requirement for lawful access to data. Credit for pressing for clarification of Japan’s position must go to the country’s creative sector, which has worked hard to ensure that Japan’s position on a TDM exception for AI training is properly understood, both domestically and internationally.

On the lawful data issue, the discussion paper recognizes the damage that piracy causes to Japanese rights-holders;

“…the damage to Japan’s content industry caused by pirated versions is enormous, and it goes without saying that countermeasures against piracy (should be) moved forward…”.

More specifically, the paper makes it clear that knowingly collecting data from a site containing pirated content increases the likelihood that AI developers or AI service providers will bear liability for infringement, as this would represent a neglect of duty of care. This statement in the discussion paper does not, of course, have the force of law but is a further positive indication of the intent and future direction of Japanese law and regulation in this area.

Arguably, Japan’s TDM exception for AI training as expressed in Article 30(4) could have been more clearly drafted, or perhaps a more precise translation could have been prepared. The misunderstanding and reporting of Minister Nagaoka’s comments outside Japan was unfortunately distorted by cultural differences and nuances of language—not to mention probably willful distortion by those seeking to weaken copyright protection globally. The ACA’s draft discussion paper is a welcome clarification of the purpose and intent of the law, demonstrating the careful balance of interests in the legislation and reaffirming the protection that rights-holders in Japan enjoy under the Copyright Act. Hopefully, the draft paper will be finalized shortly and become the foundation for further clarification of the law’s intent.

Given the unlikelihood of further legislative change at the moment, the key conclusions of the paper could be put into officially issued guidelines or regulations. These could also include potentially incorporating all three elements of Berne’s three-step test, thus reaffirming Japan’s well-deserved reputation as a rule of law country that fully respects the rights of creators and its international obligations.

© Hugh Stephens 2024.