Artificial intelligence

Political deal on the AI Act

The EU has finally agreed the terms of its landmark AI Act following a three-day negotiation marathon. The agreement includes obligations for general purpose AI systems (GPAI), such as providing technical documentation, summaries of the training datasets and complying with copyright law. Providers of GPAI-systems with ‘systemic’ risks, meaning the largest foundation models, will have to comply with more extensive obligations, such as conducting assessments and evaluations of the risks presented by them, mitigating those risks, reporting serious incidents to the European Commission and conducting adversarial red team testing. According to EU commissioner Thierry Breton, ‘the AI Act is much more than a rulebook – it’s a launch pad for EU start-ups and researchers to lead the global race.’ Some technology groups have already expressed concern that compliance with the Act will place a significant burden on companies. French president Emmanuel Macron has also warned that the rules could leave European tech companies at a disadvantage to rivals in the US, UK and China, especially over the strict controls on foundation models. He said: ‘We can decide to regulate much faster and stronger than our major competitors. But we will regulate things that we will no longer produce or invent. This is never a good idea.’

Although a political agreement has been reached, technical meetings are likely to continue into January to finalise the details. The text will still need to be adopted by member states. France, Germany and Italy are thought to be in discussions to seek alterations or to prevent the law from being passed.

Agreements reached at AI safety summit

The world’s first ‘AI Safety Summit’ has ended with a series of agreements reached among the 29 governments that attended. The Bletchley Declaration committed the signatories to ensuring the safe development of AI, recognising risks related to transparency, fairness, accountability, safety, ethics, privacy and data protection. A further agreement on AI safety testing was signed by 10 countries, including the UK, the US and major European states as well as by leading technology companies. Countries also agreed to support a report on frontier AI led by Professor Yoshua Bengio. It will be published ahead of the next summit in France in a year’s time.

---

Competition

Google loses antitrust lawsuit over its app store

A federal jury in the US has found against Google in a lawsuit brought by video game maker Epic. The search giant had been accused of suppressing competition in the Android app market to increase profits from its Play Store. Epic had sought to use an alternative billing mechanism on the Play Store that avoids Google’s fees. The case is seen as significant because it brings into question the 30 per cent commission fee that is widespread in the gaming industry. Google is expected to appeal.

---

Content

Indonesia bans social media companies from online selling

Indonesia has introduced a new regulation prohibiting social media companies from using their platforms for online sales. A government minister accused TikTok of ‘predatory pricing’ and causing damage to local small and medium sized businesses. He said the regulation ‘will justly regulate fair trade online and offline’. Sellers at Jakarta’s largest market had complained of 50 per cent losses in profits, claiming they could not compete with much cheaper imports sold online. Social media companies are said to be considering applying for separate e-commerce licences.

‘Data poisoning’ software will protect content creators

Researchers at the University of Chicago have devised a data poisoning technique designed to disrupt the training of AI models, according to MIT Technology Review. Called ‘Nightshade’, the tool alters images in ways that are invisible to the human eye, but which are sufficient to corrupt an AI model’s training process. The aim of the project is to ‘balance the playing field between model trainers and content creators’ by forcing AI training companies to license image data sets, respect crawler restrictions and conform to opt-out requests. AI researchers’ reliance on data scraped from the web is seen as ethically fraught, but companies with large existing image datasets, such as Getty Images, are at an advantage when using licensed training data.

---

Privacy/online harm

Controversial powers retained in UK online safety bill

The UK Online Safety Bill has now come into force. Controversial powers to force messaging services to examine the contents of encrypted messages for child abuse material have been retained in the Act but the regulator, Ofcom, has indicated that tech firms would only be asked to access messages once ‘feasible technology’ has been developed. A number of companies have said that they cannot access messages without compromising privacy and at least one has threatened to take the government to court if it is asked to do so. Wikipedia has also said that it would not be able to obey some aspects of the Act, including age verification. Ofcom is now expected to consult on proposals for the Bill’s implementation, including codes of practice and enforcement processes.

No evidence that online activity harms mental health

Internet use does not appear to harm mental health, according to a new study. Research from the Oxford Internet Institute said there was no evidence of widespread harm from online activities such as browsing social media, nor that certain groups are more at risk. Professor Andrew Przybylski, one of the paper’s authors, pointed out that the data necessary to establish a causal link was ‘absent’ without more cooperation from tech companies: ‘The best data we have available suggests that there is not a global link between these factors.’ He suggested that any legislation aimed at addressing mental health problems caused by online activity would need to be based on much more conclusive evidence. The paper, ‘Global Well-Being and Mental Health in the Internet Age’, studied data on 2.4 million people aged 15-89 in 168 countries between 2005 and 2022.  The paper can be viewed here.

Infrastructure

New cables to connect Pacific islands

The Gambia will connect to a second fibre optic submarine cable by 2025, the government has announced. The new cable is expected to cost between 30 and 35 million USD and will be financed by the World Bank. The country has relied on the ACE (Africa Coast to Europe) for high-speed internet services since 2012. The Gambian government is also planning to strengthen connectivity using satellites and will approve ‘all necessary licences’ to Starlink.

Microsoft expanding nuclear power in energy mix

Microsoft is expanding its plans to integrate small modular reactors (SMRs) into the energy mix for its datacentres. The company is recruiting a programme manager for nuclear technology who will be tasked with making a technical assessment of how SMRs and microreactors could be used. As well as being deployed at traditional power plants, the technology could also result in reactors being sited adjacent to the datacentres themselves. Datacentres around the world are struggling to source clean, reliable power, resulting in delays to many projects.

In Brief

The EU has reached agreement on the new Cyber Resilience Act, which will require manufacturers to implement cybersecurity measures across a product’s lifecycle. The Act will go for formal approval before entering force, expected to be early next year.

Most NFTs (non-fungible tokens) are now worthless, according to a report from crypto gambling company dappGambl. NFTs are a crypto asset that certifies digital ownership. They were envisaged as a key component of the future metaverse and the subject of a buying frenzy in 2021.

The first of a series of country reports from the Global Media & Internet Concentration Project has been published. ‘Growth and Upheaval in the Network Media Economy in Canada, 1984-2022’ by Professor Dwayne Winseck of Carleton University can be downloaded here.

The Australian competition regulator, the ACCC, has released its latest report for the digital platform services inquiry, concluding that new competition laws were required. The report claims that the risk of harmful behaviour had increased with the growth of the major technology platforms, enabling practices which reduced customer choice and deterred innovation.

Funding for European technology companies will almost halve in 2023, according to venture capital group Atomico. Tech start-ups in Europe are expected to raise $45 billion, down from $82 billion in 2022. The shortfall is due to reduced funding from US investors amid a global fall in venture capital investment since the pandemic, and the impact of rising interest rates.

Iran plans to be a leader in ‘clerical AI’ as the country’s religious establishment looks at ways to harness the technology. The head of a state-linked organisation that encourages the growth of technology businesses, Mohammad Ghotbi, said that ‘robots can’t replace senior clerics’ but they can help them to ‘issue a fatwa in five hours instead of 50 days’.

Sources: The Financial Times, Reuters, Europarl.europa.eu, the BBC, Ars Technica, Bloomberg, Politico, Euronews, telecommpaper