Appendix 4: Committee visits
Committee visit to ITN
On 12 March 2024, the Committee visited ITN’s offices in London. In attendance were Baroness Stowell of Beeston (Chair), Lord Dunlop, Lord Hall of Birkenhead, Baroness Healy of Primrose Hill, The Lord Bishop of Leeds, Lord Storey and Lord Young of Norwood Green.
The purpose was to learn more about the work of the UK’s public service broadcasters and their approach to producing news. We discussed the work of ITV, Channel 4 and Channel 5 newsrooms in catering to different audiences, and the importance of ensuring news stories are selected to appeal to a broad range of demographic groups. We further discussed the opportunities and challenges of providing news in different formats across multiple platforms—and the implications for needing multi-skilled teams able to deliver on these requirements.
Committee visit to San Francisco
The Committee visited San Francisco from 7 to 11 October 2024. In attendance were Baroness Stowell of Beeston (Chair), Lord Hall of Birkenhead, Baroness Healy of Primrose Hill, Lord Kamall and Lord Knight of Weymouth.
The primary purpose was to learn about the work of US technology firms and media organisations in relation to our inquiry into the future of news. We also used the opportunity to examine issues with generative AI as a follow-up to our recent report on large language models. We further held meetings relating to our inquiry on the challenges facing AI-first and creative technology firms seeking to scale up in the UK. We are grateful to all those who participated in our work.
The section below provides an overview of our view on some key themes that emerged from the visit. The subsequent sections provide a non-exhaustive summary of our engagements.
Our view on key themes
We found much uncertainty about the impacts of technology on news. The roles of large technology platforms, and how they viewed their position in the market, varied considerably. The relationships between them and news providers will continue to evolve as generative AI plays an increasingly influential role in the production, discovery and monetisation of news.
Some trends around the most recent advances in large language models were however becoming clearer. Generative AI tools were thought to be likely to have increasing influence over the types of information people see. Whether the AI applications provide one answer to a question about current affairs, or a variety of views, could prove decisive in shaping the average casual user’s views on a topic. We heard that these new tools would likely drive a shift in consumer behaviour—away from clicking on links, and towards long-form AI-generated summaries that draw on multiple sources (perhaps increasingly involving voice activated tools). News media business models based on users clicking through to viewing a website might therefore suffer if the AI summary is ‘good enough’ for the average reader. Some stakeholders worried that these trends could make it economically unviable for some news outlets to continue producing quality journalism.
The interests of publishers and AI developers appeared to be aligned around the value of high-quality, timely data needed to train and ground AI models. It was less clear though whether AI firms would be prepared to pay sufficient amounts to sustain the continuation of high quality journalism that provides a plurality of viewpoints. We further heard that the value of journalism lay not just in breaking news: much of the information available on the internet is accepted as fact because it has been originally produced and verified through journalistic endeavours. If generative AI tools lead to a situation where quality journalism becomes increasingly uneconomical, it was unclear how well future AI tools would function.
Many news providers were pursuing a variety of licensing deals with AI firms, though how favourable these arrangements were in practice remained unclear. The terms may involve AI firms gaining access to archive or real time news (or both), in exchange for payment and/or exposure. We heard concerns that the types of licensing deals may be, effectively, a time-limited offer that do not guarantee a sufficient link between the growth of the AI marketplace and a commensurate expansion in news providers’ revenue.
Our discussions suggested that we were still in the early stages of generative AI-influenced news. Widespread changes to audience habits may occur over the next few years, though the details of what this might look like remained uncertain. Media organisations were already experimenting with AI and may shift from using ‘AI as an efficiency’ to ‘AI as a service’ over the next few years too. The negative impacts of generative AI on news advertising revenues, exposure and audience engagement could be envisaged, but the full effects were likely to take some time to feed through and newsrooms were likely to adapt along the way. This all suggested that the long-term implications for the financial health of the media sector would play out over the next five years or more, rather than in a matter of months.
Finally, we heard that progress in generative AI capability is not linear: those at the forefront are likely to draw increasingly far ahead. It therefore matters what types of AI firms and services flourish, and where the UK chooses to focus its efforts. Some stakeholders envisaged a future of 5-10 major providers of base models, who effectively act as commodities (comparable to the situation with cloud compute providers). A plethora of smaller open source and specialised models may be built on top. Some stakeholders suggested this second layer is where the transformative value to society will be at, and where the UK should focus its efforts in developing public-interest AI tools and standards.
Meta
We met with Andy O’Connell, Vice President, Product Policy; Matt Scutari, AI Policy Director; Elizabeth Kendall, Global Public Policy Director, Media and IP; Marc Dinsdale, Director of Digital Media Partnerships & News; Megan Thomas, Public Policy Manager and Richard Earley, Head of UK Content Regulation Policy
Meta provided updates on their latest products, including wearable technologies and generative AI tools. Meta discussed developments on their Llama model and the moves towards multimodal and specialist use cases.
We discussed issues relating to the spectrum of ‘open’ and ‘closed’ source models. Meta outlined the importance of openly accessible development, which we heard promoted better innovation, competition and transparency. We further discussed the need for guardrails and safety interventions, particularly around identifying AI-generated content.
We heard about Meta’s in-application AI assistant, Meta AI, and the increased popularity of voice interaction. Meta highlighted the importance of models being trained locally in order to reflect cultural nuances and better serve consumers.
Meta executives reiterated previous comments about the limited extent to which audiences seek out news on its platforms. We discussed Meta’s decision to switch off news tools in Canada, as well as concerns about the use of copyrighted data, licensing arrangements and the differences between the UK, EU and US approaches in this regard.
Finally, we discussed the steps Meta is taking to counter misinformation and disrupt co-ordinated inauthentic activity on its platforms.
We met with executives Kent Walker, President of Global Affairs; Jaffer Zaidi, Vice President for News Partnerships; Prabhakar Raghavan, (then) Senior Vice President, Knowledge and Information Products; and Katie O’Donovan, Director of Government Affairs and Public Policy, UK.
We discussed the merits and challenges of Google’s initiatives to support the news industry, including lump-sum payments, licensing deals and innovation funding. Google said it was testing new approaches for supporting local news, and suggested that more innovative business models might emerge and help attract better talent into journalism.
We heard about the implications of AI for the news industry. Google suggested that news organisations’ internal use of AI may evolve from a focus on efficiency towards exploring opportunities for new products and services. We heard that text-to-speech products, as well as summarisation tools (for example Google’s Notebook LM), could provide opportunities for efficiency. We further discussed future options for publishers to generate and monetise content using AI tools, as well as future trajectories for deeper interaction with generative AI audio content.
We discussed copyright concerns, including the potential challenges and opportunities of adopting an EU approach to copyright. We discussed whether licensing agreements may be most useful for grounding AI tools to ensure they provided accurate answers, rather than for the initial training of foundation models. On regulation more broadly, we discussed the potential tension between competition regulators wanting openness on one hand, and privacy regulators preferring information to be protected.
Finally, we discussed Google’s Search business. We discussed suggestions that some younger audiences are seeking authenticity over authority, which in turn has implications for how younger demographics seek out and use information online. We discussed whether this indicated a commensurate political shift in the prioritisation and ranking of topics and news. Google said that developments among younger audiences did not mean that Google needed to change the way its search engine ranked information.
Apple
We met with executives including Roger Rosner, Vice President, Applications and Nick Ascheim, Head of Apple News Business.
We discussed the role that Apple News played in the media environment, notably as the most used news app in the US. We heard that its top stories were chosen by human editorial teams based in each global region where the service was offered; the choices were intended to provide a balance of outlets and ensure topics were relevant to the user.
Alongside human-selected top stories, we heard about how Apple News provided algorithmically-driven personalised news recommendations. We heard that editorial quality and balance remained an important consideration for this feature. Apple News did not have user engagement tools common to social media platforms (such as ‘likes’).
Apple News featured stories from partner media organisations; Apple told us that it did not crawl the open internet for content. Apple had commercial arrangements so that publishers may earn a share of revenue via engagement time on Apple News+ (the app’s subscription product for paywalled content) or via advertising the publishers sell themselves or that third parties sell on their behalf.
We further discussed the implications of generative AI for news, and how this would affect different types of news outlets, as well as wider trust in the information environment. Apple executives noted the challenges, though highlighted that there were still opportunities for quality news to remain resilient in the face of AI-generated content.
OpenAI
We met with executives including Jason Kwon, Chief Strategy Officer; Tom Rubin, Chief of Intellectual Property and Content; Chris Lehane, Vice President of Global Affairs; and Mark Gray, Copyright Counsel.
We discussed the trajectory for future developments in AI, including the possibility for humans to interact with the AI model throughout its reasoning process and the potential for more ‘agentic’ capabilities (for example asking ChatGPT to research, plan and book a vacation).
We further discussed AI safety issues, including the potential for autonomous artificial agents capable of navigating their environment without oversight. OpenAI told us that significant effort was being dedicated to improving guardrails and ensuring safe and responsible development.
We heard about the partnerships that OpenAI has developed with news providers. OpenAI expressed a commitment to collaborating on ways to drive readers of news content back to publishers. The details of recent licensing deals with news publishers, and opportunities for renewal, remained unclear. OpenAI suggested that there was particular value in using news content to ensure ChatGPT and search provided grounded and accurate answers. We noted that the relative value of news archive content for training base models may be less significant.
Perplexity
We met with Aravind Srinivas, CEO and co-founder of Perplexity, a search engine which provides users with AI-generated summaries of search results with citations to sources.
We discussed competition in the internet search market, and the role of generative search in disrupting the digital advertising landscape which often relied on users clicking through to websites. We heard that consolidation in the browser market made it difficult for start-up companies to strike partnerships with web browsers.
Although Perplexity offered summaries of the news, we heard that it did not see itself as an alternative to news sites. Its selection of the top news sources was AI-driven, though its algorithms would summarise content from websites (rather than social media posts). We discussed the potential implications of AI-generated summaries on news publisher web traffic.
While Perplexity offers a premium subscription, we heard that further monetisation can come from corporate sponsorship of associated queries. Perplexity emphasised that sharing revenue from these activities was a valuable way to develop mutually beneficial partnerships between publishers and tech firms. Perplexity also said that it was looking to engage with more publishers outside the US, where it had made some deals.
Scale AI
We met with Vijay Karunamuthy, Field Chief Technology Officer and Max Fenkell, Head of Government Relations.
Scale AI executives discussed their work on providing businesses, researchers and governments with comprehensive and high quality data to support safe AI development and deployment.
We heard that the availability of well-structured data sets was important for developing the next generation of models, and to train models for specific use cases (such as in health care or finance). Scale AI discussed the importance of good data governance and opportunities for developing in-house proprietary models.
We heard that the particular value of news lay in being authoritative and timely; such content could be used to ground AI models and ensure they provided accurate answers to questions about real-time issues. News sources could also be used to help a model better assess the validity of information it received from other sources on the web and give greater weighting to authoritative sources.
We also discussed the restrictions on data training and the need to avoid AI ‘eating its own tail’. Scale AI executives outlined the difficulties in using synthetic data. We heard there may be better prospects of success in some cases (perhaps in science, maths and coding) and limited likelihood in others (notably creativity and cultural awareness). The prospects for synthetic video and audio material would be an area of future research.
SmartNews
We met with SmartNews executives including Rich Jaroslovsky, Adviser to the CEO at SmartNews, a news aggregation app available in the US and Japan.
SmartNews executives discussed their focus on the value of news to democracy. One example was the news app’s algorithms being designed in ways that promote unexpected content that may not necessarily align with the user’s political outlook. We heard that news aggregators could seek to engage younger audiences by offering them a range of articles that enabled them to test different ideas and viewpoints easily.
We heard that SmartNews was selective in its news partners, using a set of criteria to confirm whether media outlets met its guidelines. Partners entered into a commercial arrangement and worked with an in-house team of former journalists on story selection.
We heard criticism of tech firms offering short-term or ad-hoc payments to publishers. SmartNews argued that a successful approach required long-term licensing deals that accurately reflect the cost and value of the news content. They suggested that the most viable long-term proposition involved aligning the interests of AI firms and publishers, not treating the latter as a temporary engagement partner.
We met with Ben Lee, Chief Legal Officer.
Reddit discussed its position as one of the world’s largest repositories of publicly available human-generated content.
Reddit expressed concern about AI firms obtaining this content for AI training without permission, and outlined the copyright and privacy implications for the users posting in the network. They suggested that unauthorised crawling risked becoming a disincentive for people to engage in the open internet, and raised concerns about the resulting trend towards information and communication moving into ‘walled gardens’ in private apps.
Reddit further discussed the value of human-led content moderation, while noting that AI could also help users apply the rules of engagement established by user communities.
Mozilla
We met with Masayo Nobe, Vice President of Legal and Michael Feldman, Managing Counsel, Product and Privacy.
Mozilla spoke about the need for policy and regulation to be developed in a way which allow AI to be developed for public benefit. Mozilla discussed the steps this might involve, including options for standards and regulations to encourage the development of openly accessible low-cost models that could be used by many people and institutions in society.
Mozilla discussed several public policy gaps. One example was ‘open washing’, which referred to the misapplication of the term ‘open source’ to AI development. We heard that this enabled companies to reap benefits attached to the concept without providing sufficient openness and transparency. Mozilla suggested that better definitions would be helpful.
We also heard about the risks of consolidation between tech firms and limited availability of compute, which in turn underscored the value of competition legislation. Mozilla suggested new rules could be introduced to allow for an amount of compute to be publicly available, or to promote open-access development.
San Francisco Chronicle
We met with Bill Nagel, Publisher and Emilio Garcia-Ruiz, Editor in Chief.
We discussed the continued evolution of news business models in the face of technological and behavioural change. This included changing audience habits and the implications for the presentation of news content. We heard that some US media organisations were considering hiring influencers to promote news among younger audiences. We discussed the challenges of this regarding the retention of influencers and alignment of values. The Chronicle noted that the delivery mechanisms for news may change but the core substance and principles of quality journalism must be preserved.
We further heard that some publishers were driving readers to closed environments (i.e. in applications or paywalled content) in order to build revenue streams that did not rely on intermediaries. The shift towards moving content off the open web might also have been linked to concerns about unauthorised use by AI firms.
We discussed concerns about the relationship between tech firms and news publishers, noting in particular challenges that had arisen in the past from the rise of social media and other online intermediaries. We further discussed uncertainty around the types of deals being struck with AI firms and how these may unfold in future, alongside the issues raised by legal challenges to tech firms’ practices.
Journalism roundtable
We met with a number of journalists and local news providers including Lance Knobel, CEO and co-founder, Cityside Journalism; Neil Chase, CEO, CalMatters; Soo Oh, Editor and data journalist, The Markup; and Shirin Ghaffary, Reporter, Bloomberg.
We heard about the impact of declining local journalism on democracy. For example, we were told that the lack of reporting on local institutions and public interest matters was becoming economically unviable. We also heard that it was important for local news organisations to appeal to a broad demographic, and address criticisms that local news had previously catered largely to older and wealthier groups.
We discussed the challenges facing local news organisations and the need for organisations to adapt and experiment with new business models, alongside other considerations around the rise of non-profit news providers, and the trade-offs around receiving government funding. Participants discussed suggestions about adopting lessons from influencer culture, and the difficulties involved for a news brand in relying on individual journalists as influencers to attract younger audiences.
We heard about AI being used to generate news sites to promote special interests and ‘clickbait’ outlets. Some participants were optimistic about the opportunities for publishers to use AI for journalism and distribution.
Financial Times
We met with Cristina Criddle, Technology Reporter at the Financial Times.
We discussed trends in the work of tech platforms, alongside wider issues that other stakeholders had raised relating to market competition and the level of influence tech firms had over access to information. Other topics discussed included developments in AI, including its uses in the newsroom and relationships between publishers and platforms.
Andreessen Horowitz (a16z)
We met with Anjney Midha, General Partner at Andreessen Horowitz (a16z).
We discussed the different approaches to developing large language models in the US and Europe. There had been suggestions that the US market tended to focus on increasing the capacity of models (larger and more capable), whereas the European market had focused more on efficiency (smaller and nimbler). We discussed the possibility that capability and efficiency may become the two main fronts of AI research.
We further discussed the actions needed from countries seeking to become centres for AI development. Key points included access to low cost energy computing capacity (compute) and forward-looking regulation to attract researchers and developers. Further ideas included making it easier for companies to list in the UK to compensate for the lack of UK growth venture capital funding. We also heard about initiatives in France and Canada that provided access to compute and allowed for closer partnerships between academia and industry.
We discussed developments in the debate about open or closed model development, and heard that open source models can generate revenue by adopting an ‘open core’ model, which requires licensing for commercial development.
Finally we discussed the value of live journalism to generative AI search products which require accurate and authoritative information.
Investor Roundtable on AI growth opportunities
We met with members of GBx, a community of British founders and investors based in San Francisco. Attendees include Tom Blomfield, Group Partner at Y Combinator and co-founder of Monzo; Alastair Paterson, CEO and co-founder of Harmonic Security; Morgan Beller, General Partner at NFX; Doug Bewsher, investor and former CMO at Salesforce and Skype; and Julian Green, entrepreneur and former Director at Google X.
The discussion focused on AI growth opportunities and challenges in the UK, which were examined in our large language models report. The roundtable was also intended to inform our inquiry on scaling up AI and creative tech companies.
Participants were broadly positive about some elements of the business landscape in the UK, including talent and pro-innovation financial policies such as the Seed Enterprise Investment Scheme and research and development tax credits. However, they identified six key factors which limited the opportunities for AI growth in the UK: entrepreneurial talent; a small domestic market; lack of local expertise; limited links between universities and industry; limited government procurement opportunities; and limited support or incentives to launch an initial public offering (IPO) in the UK.
Talent
We heard that the UK had a significant amount of technical talent, and that many companies kept their engineering teams based in the UK even if other parts of the business moved to the US to access a larger market and greater funding. Some attendees suggested that the ambition of many British-founded companies was geared towards eventual acquisition by a US company, rather than competition and market share. We discussed potential solutions to this issue within the education system and the availability of successful entrepreneur role models.
Smaller domestic market
Participants suggested that most AI companies seeking to scale up would need to enter the US market at some point, often as soon as possible. We heard that the UK’s medium-sized market meant that founders focusing on UK growth risked missing opportunities in the US because they had moved too late. Some participants suggested that the UK needed to take steps to reposition itself as the gateway to the European market, which could open access to a wider market closer to the UK, and potentially encourage more international start-up companies to set up as UK limited companies.
Capital
We heard mixed opinions on the availability of growth capital in the UK. Some argued that the arrival of US venture capital funds had injected sufficient investment in growth, and that foreign and domestic capital sources were effectively interchangeable at later stages. Others suggested that domestic patient capital—in pension funds, for example—was not being targeted at growth companies as an asset class. We also heard that a lack of ambition detered US venture capital investors, who were looking for higher return multiples than UK founders suggested they would be able to achieve.
Networks
Participants said that even if capital was available in the UK, some founders moved to San Francisco to benefit from the community of previous founders and investors who had experience of building and scaling up companies. Some noted that this issue was not unique to the UK–founders moved to the Bay Area from elsewhere in the US as well as other countries. We heard that the Government could do more to incentivise successful British founders to return to the UK and incubate a similar environment domestically.
Universities
While participants agreed that the UK’s top universities were producing high quality talent, we heard that they still struggled to commercialise research. Imperial College London was cited as a market leader in this respect. Attendees said that other universities had historically taken large equity stakes in spinout companies, but suggested that this was starting to improve. They further suggested that more could be done to deploy university resources and assets, including endowments, to support start-up companies. Participants highlighted California’s lack of non-competition laws as a key driver of local spinout success, and advocated for the UK to make it easier for academics to move between research and industry.
Procurement
We heard that other countries, including Denmark, the Netherlands and the US had better initiatives to help start-up companies demonstrate proof of concept and engage in contracts with government.
Public markets
Participants noted that fewer UK companies were choosing to list on the London Stock exchange. Higher valuations were available elsewhere, particularly in the US. We heard mixed opinions on whether it mattered where a company launched its IPO, if its headquarters and engineering talent remained in the UK.
Some participants suggested that the capital and compute-intense race to build foundation models may ultimately lead to a situation where a small group of providers offered base models which effectively acted as commodities. Some participants suggested that the UK should therefore focus on innovation at the application layer of AI, rather than try to compete with US tech giants in developing foundation models. Participants suggested that the UK’s existing success in financial services, gaming, media and advertising could form the basis for the growth of smaller AI models designed to solve specific issues in these industries.