The AI Provisions of the Fair Digital News Bargaining Bill

An Attempt to Legislate a New Technology

This article was written whilst the Fair Digital News Bargaining Bill was still under consideration after the Select Committee Report. It addresses material that was added by the Select Committee about the applicability of the Bill to Artificial Intelligence.

At the time of posting (24 June 2024) it was unknown whether the Bill would proceed. Subsequently the Government announced it will proceed with the Bill But what is of interest is the approach that has been taken to AI. This is perhaps one of the first attempts to cast a legislative net around a slippery technology. The Government has indicated it will not proceed with these proposals but this article is submitted for your interest on a “what might have been” basis.

Introduction

The Fair Digital News Bargaining Bill is a legislative proposal aimed at establishing fair bargaining conditions between news media entities and operators of digital platforms. ​

It outlines processes for compulsory and voluntary bargaining, sets forth a bargaining code, and establishes civil penalties for non-compliance. ​

The purpose of the Fair Digital News Bargaining Bill is to support the sustainable production of New Zealand news content by ensuring that operators of digital platforms make a fair contribution to the cost of producing news content. ​

The bill aims to incentivize operators to enter into news content agreements and other arrangements that contribute to the sustainable production of New Zealand news content. ​

It also aims to facilitate fair bargaining between news media entities and operators about the terms on which news content can be made available on digital platforms. ​

Additionally, the bill allows for the imposition of news content agreements that require operators to fairly compensate news media entities through binding arbitration.

The Bill was introduced into Parliament by the Labour Government in 2023. It was sent to the Economic Development, Science and Innovation Select Committee. After the October 2023 election the Coalition Government decided to continue the Select Committee consideration while at the same time making no commitment as to the future of the Bill.

The Bill was reported back at the end of May 2024. The recommendation of the Select Committee was that the Bill not proceed further citing concerns about its potential impact on the online marketplace and free speech.. However, in the event that recommendation was not accepted, a number of amendments were proposed to the original Bill.

The most significant of these amendments dealt with the inclusion of Artificial Intelligence (AI) in the Bill. These provisions were completely absent from the Bill as originally introduced.

In this article I consider the provisions of the Bill as they relate to AI. I do so because this is an early attempt to provide some form of regulatory recognition of AI.

An Overview of the Bill

The problem that the Bill seeks to address is this. Online digital platforms aggregate and display news content to attract attention to their sites, and make money through advertising and other services. The advertising revenue that they derive from that activity means that advertisers are no longer using mainstream media outlets (MSM) to promote their products.

A further concern, stated in the explanatory statement to the Bill, is that the online platforms do not share that monetisation with the people who create the news content. Attempts by news media entities to bargain and obtain compensation for the value of their news content have been by and large unsuccessful. That said, Google has voluntarily entered into arrangements with content providers. But the concern is that arrangement could come to an end should Google no longer wish to continue it.

The Bill was intended to support sustainable production of New Zealand news content by ensuring that operators of digital platforms make a fair contribution to the cost of producing news content that is made available by their digital platforms.

That purpose would be fulfilled by incentivising operators to enter into news content agreements and other arrangements that contribute to sustainable production of New Zealand news content. The Bill would facilitate “fair bargaining” between news media entities and operators about the terms on which news content produced by the entities may be made available by the operators digital platforms.

If it was necessary the Bill would impose news content agreements that require operators to fairly compensate news media entities for that news content through binding arbitration.

Rather than leave the matter to market forces, creativity and innovation, the proposal underlying the Bill is that the State would step in to level the playing field. In reality, however, what is proposed is a means of shifting financial help for MSM from the State to the Platforms through a form of compulsory bargaining.

The way that the Bill proposed this should be done was to create what was described as a fair bargaining environment through a bargaining code that would be established by an independent regulator and operate as secondary legislation. That means that the Code would have the force of law but could be examined by the Regulatory Review Committee of Parliament.

Once the Code was in place bargaining parties would be required to comply with the bargaining code and to bargain in good faith. There was also a requirement  requiring parties registered under the legislation to participate in the bargaining process.

It was proposed that the Bill would promote voluntary commercial agreements between operators of digital platforms and news media entities, with minimal government intervention. Thus a voluntary solution was preferable.

Where agreement could not be reached, a stepped bargaining process would take place to facilitate, from the perspective of the regulator, fair and equitable outcomes. Civil penalties (as distinct from fines) are provided for non-compliance with the legislation along with provisions inserted by the Select Committee relating to compensatory orders which could be made concurrently with pecuniary penalties.

The Bill focusses on digital platforms that make news content available by those who are other than news media entities – otherwise known as mainstream media or MSM.

The Ministry of Culture and Heritage Departmental Report on the Bill, which was compiled after submissions were received, recommended the the inclusion of artificial intelligence in the Bill.

Incorporating AI

The Select Committee proposed that digital platforms be extended to include those that used AI.

The Bill – in defining a digital platform – addresses the “flavour” of AI applicable to the Bill. It does this by defining an AI system and an AI service.

An artificial intelligence system is defined as a machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments.

Thus the criteria are:

-      a machine based system

-      that infers how to generate outputs such as predictions, content, recommendations, or decisions

-      that can influence physical or virtual environments

-      and that it does so for explicit or implicit objectives

In my view there is a significant amount of unnecessary verbiage in this definition that actually is confusing. A simple and technologically accurate definition should contain the following criteria:

-      a machine based system

-      that infers how to generate outputs

-      for explicit or implicit objectives

-      in answer to a query

The original definition ties in with the definition of an AI service which means an internet service that uses an artificial intelligence system to generate outputs for users of the service.

However, if the simple approach I have suggested were adopted the definition of an AI service could be simplified to state “AI service means an internet service that uses an artificial intelligence system.”

Thus the definition of system tells us how AI actually works and the definition of service locates AI into the Internet environment.

Having defined what an AI system and an AI service is the Bill goes on goes on to link an AI system to news content for the purpose of training the AI system.

The focus is upon the way in which news content may be used to train a digital platform or AI system.

The first element is that an AI system must be trained using news content. This links to the definition of news content in the Bill. The training must generate outputs which happens if the AI system enables or facilitates the generation of outputs.

To try and break it down further:

-      an AI system is trained using news content

-      if the news content is used to train the system to generate outputs.

Thus there is a necessity for outputs. Training an AI system alone would not be sufficient to fulfil the definition. There must be some consequence or purpose for the training – that is the generation of an output for users. When we cross reference that to the definition of a digital platform which includes an AI service we find that training must involve the use of news content produced by news media entities but the output need not make news content available.

What is important for the purposes of capture by the provision of the Bill is that when the output does in fact contain news content, the person viewing the output must be significantly less likely to view the original content

But that is not enough on its own. In addition to those two elements there are other requirements.

Training a digital platform using news content occurs if

-      The digital platform fulfils the definition of an AI service (actual or prospective AND

-      The AI service or AI system is trained using news content

A digital platform is trained to generate outputs if

-      The platform’s AI system enables etc the generation of outputs by the system

All these elements must be present to fulfil the requirement of using news content to train the AI system. If there is an absence of linking the training with the generation of outputs the AI system falls outside the definition.

The section helps to clarify the nature of training. Training includes:

-      Supervised learning where the training is to generate outputs using labelled data – that is data that has metadata associated with it. The subclause describes the metadata as “labels that provide additional information about the data.

-      Unsupervised learning where the system is trained to generate outputs using an unlabelled dataset – that is a dataset that lacks metadata

-      The training of a new system or

-      A new version of an existing system

-      Before the system is used to generate outputs (This would seem to suggest that the training of the system must be complete before queries requiring outputs are provide)d

-      Iterative training of an existing system during its ordinary course of operation – this suggests built in machine learning

-      Any other type of training prescribed by the regulations

Thus the regulations may add another form of training to the categories of training that are already set out. This allows the Minister to alter or add to the Statute without the scrutiny of the legislative process

Finally the training may be carried out by a person (human assisted training) or by the system itself (machine learning) or by a combination of the two.

The critical aspect of the AI provisions is that the training of the AI system and service must use news content. This is critical to bring the AI system or service within the scope of the proposed bargaining provisions of the Bill. An AI system that is NOT trained using news content does not fall within the proposed Bill. The news content – if I read the Bill properly may fall within the 5 categories of learning/training.

There seems to be one shortcoming and that is that the training may take place using news content along with other data from other sources. In such a case is the training to be exclusively with news content or may other content from other sources be included? I would suggest the answer should be yes in that generative AI models (or large language models) scrape their data from across the Internet to generate the predictive outputs in answer to a prompt or query.

Fit for Purpose?

Are the Select Committee proposals regarding AI are fit for purpose. Are they adaptable for any future AI developments that may be forthcoming. Do they deal with a specific “flavour” of AI and thus are limited in scope?

The AI methodology defined in the Bill is a form of generative AI or Large Language Model and is thus within the category of a weak or narrow AI.

The scope of the AI is restricted by the type of training that is prescribed. The training must involve the use of news content. This may well take place within a wider “data-scraping” activity in obtaining training data. However, it would be possible for a “scraping” algorithm to specifically exclude news content sources thus excluding the AI system from the definition.

The definition of AI service and AI system is also very limited and does not allow for adaptation or innovation in AI. Once again this is because the type of AI is restricted to a LLM generative AI model and furthermore the content that is used for training narrows the definition even further, although the regulations may add another form of training in addition to those specified. Nevertheless the foundational problem of the use of news content will remain.

Although the definition may cover the specific flavour of AI – that is generative AI – and may be fit for purpose in that respect, the problem – as is so often the case with legislating for technology – is that the definitional restriction of a technological type means that the statute is frozen in time or, in this case, technology. It ignores, especially in the context of the Digital Paradigm, the properties of continuing and permissionless innovation along with continuing disruptive change.

I have argued elsewhere that I believe that the use of content by digital platforms does not require a specific piece of compelled bargaining legislation. Rather the use of another entity’s content locates the problem within copyright law.

It would be a simple matter to include the utilisation of content for training an AI system or service as a form of infringement and the measure of damages could relate to whether the use of the output meant that it was significantly less likely for the user to access the original content. An additional aspect of the measure of damages could include consideration of whether or not the data was scraped from a “pay-per-view” site rather than a “free-to-air” one. But these are matters which may need to be considered separately.

As matters stand, if the Bill were to proceed, the restrictive nature of the definitions of AI and the limitation to training from news content, together with the requirement that an output in response to a prompt would have to be “news specific” with a significant likelihood that the original content would not be viewed means that the proposed legislation would be so technologically hobbled that it would be of very limited use