Tom Upchurch

No Silva-Lining on AI Act Analysis Writing in the Times this week former Conservative adviser Rohan Silva opines that the EU’s AI Act will stifle innovation – both in Europe and the UK – and will overregulate the nascent sector. In a piece remarkable for its disobliging language he calls the AI Act “dunderheaded”, “half-witted”, “dreadful” “barmy” and “thick-headed” (no please, tell us what you really think about it!) The thrust of Silva’s argument is that the EU is imposing “onerous requirements” to provide “detailed summaries of content used to train AI models”. He is attacking the notion that downstream users of General Purpose AI systems – e.g. us – should have full transparency on the content in the systems they are using. And as importantly he is full-square against the idea that AI developers have to adhere to copyright law and tell rightsholders when they have used their content. When it comes to clobbering copyright Mr Silva has form. Readers may recall him in Prime Minister Cameron’s backroom team in 2010 which decided that UK economy really needed to relax copyright law. Thus the creative sector should step out of the way of the Silicon Valley giants like Google (where a number of his colleagues mysteriously ended up working) and the UK would benefit from massive economic gains – according to a non-existent economic impact analysis. Anyway, his diatribe against the AI Act is factually wrong on a number of counts. Silva mistakenly says it was “hastily drafted”. But the first twitch of the EU legislator’s pen was as far back as 2019, with the High-Level Working Group, and the final Act gestated under close scrutiny from that point onwards. He chastises the EU for responding to the development of Large Language Models, the existence of which was known to precious few people five years ago. But to ignore the new technology altogether would have been a far greater sin. He believes that the requirement for sixty pieces of subsequent secondary legislation shows the patchiness of the Act and that such measures are needed to “plug the gaps”. This is disingenuous. He must know, full well, that Acts routinely derogate detailed implementation standards to secondary measures. Silva is also wrong to say that the EU sets an arbitrary upper limit on the size and power of the AI models. The Act is very clear that the number of floating point operations to determine a system as “high risk” is a threshold – a floor not a ceiling; and furthermore that this should be adjusted over time to reflect technological changes. What Silva clearly most dislikes is that the EU is utterly correctly forming a balance between the needs and interests of rightsholders and users on the one hand and the requirements of the tech development sector on the other. As RELX is a company which encompasses this division, I am bound to see it as a great strength of the Act. It is disappointing that again such one-sided thinking is prevalent in the debate. #AI #EU #Copyright

53

7 Comments

I've got a fantastic panel lined up for the #BCSPolicyJam tomorrow discussing #deepfakes and elections: Tim Clement-Jones, a Liberal Democrat peer and their spokesman for digital in the House of Lords. He was the former Chair of the House of Lords Artificial Intelligence Select Committee and is Co-Chair of the All-Party Parliamentary Group on Artificial Intelligence. Tom Bristow, Tech Reporter, Politico. Lisa Forte, Cyber Security Expert and partner at Red Goat Cyber Security. Hannah Perry, Researcher at CASM, is in the digital policy unit at Demos. She leads Demos’s information ecosystem programmes and is interested in how our democratic systems, institutions, and cultures evolve in a digital age. We'll have the results of the latest BCS, The Chartered Institute for IT members survey on what our tech experts think will be the impact of deepfakes on the upcoming UK election, presented by BCS Director of Communications, James Woodward. We'll also broaden the discussion to look globally. In a year when almost half the world’s democracies hold elections, how do governments deal with this? Is regulation the answer – and to what extent? And should political parties themselves be upfront about their own use of #AI in their campaigns? What can be done to help the public spot fake audio, video and images? What part do technical solutions play - watermarking, for instance? What’s the role of mainstream media – at a time when there is growing distrust of traditional news?  There’s still time to sign up – midday Wednesday 24th, via Eventbrite https://hubs.ly/Q02tz_2n0 

18

Some policy areas have had a LOT more exposure than others during the UK #GeneralElection. In this blog for Full Fact I set out what our AI tools have revealed. https://lnkd.in/ddupTuK6

4

The latest #RadioDavos hosts Cecilie Fjellhøy with the best explanation of Romance Fraud that you are likely to find. Whether you want to protect your loved ones or inform policy makers, listen to this. Cyber-enabled fraud is now a globally scaled business. Thankfully, law enforcement and the private sector are getting better and quicker at dealing with it. World Economic Forum Cybersecurity is accelerating responses to cybercrime through the work of its Partnership Against Cybercrime and by hosting the Cybercrime Atlas: https://lnkd.in/d-mrukj4 a collaboration of more than twenty organisations led by Fortinet, Microsoft, PayPal and Santander.

17

3 Comments

Check out my latest UKTN column!

12

Love this piece from the excellent Seth Lazar on summarisation in the context of government and democracy. A huge amount of it is relevant to other contexts where accuracy and nuance matter. "Next time you’re considering buying something from Amazon, have a quick look at the AI-generated summary of the reviews. It basically just states the obvious. If you really want to know whether the product is worth buying, you have to look at the one-star reviews (and filter out the ones complaining that they had a bad day when their parcel was delivered)."

19

1 Comment

Comments from Perplexity AI’s CEO, noting that robots.txt is not a legal framework, and that Perplexity also relies on 3rd party crawlers for data access, will come as no surprise for publishers dealing with industrialised IP theft in the age of AI. https://lnkd.in/ehZi4wNT Robots.txt has at least 3 clear problems: It's voluntary - so no sanction for developers who ignore and scrape anyway. Essentially a parish notice asking billion dollar businesses not to nick the thing that's central to building and grounding their payday LLMs. No wonder many companies are ignoring requests.  The next government should mandate rules with significant penalties attached for crawlers who don’t comply with instructions from website owners. No transparency of crawlers - with no duty on web crawlers to identify themselves they can scrape and index IP anon.  So even with a web application firewall in place to block scrapers, as an IP owner you don't know who's scraping what or why tilll they dain to tell you.  We need a transparent register of first and third party crawlers used or utilised by any AI developer that wants to trade with UK businesses and consumers. Laundering of IP for commercial uses - Various copyright regimes around the world have well-meaning but flawed exceptions for text and data mining for non-commercial purposes.  These exceptions have created loopholes whereby academic institutions and non-profits scrape commercial IP for “non-commercial purposes”, only for these datasets to be repurposed by billion dollar businesses for commercial purposes, without a commercial licence.  The next government should narrow these exceptions, and put in place sanctions where non-commercial users of IP allow that data to be used for commercial purposes. There's lots of joined up thinking going on between different regulators on AI's impact on the UK economy and wider society.  The DRCF AI and digital hub is a good example of this collaboration in action.    But plans such as Peter Kyle + the The Labour Party's to create a Regulatory Innovation Office must also recognise the pivotal role of copyright in training and ground LLMs, and make sure that the Intellectual Property Office has a seat at that table.  There is no doubt that the IP created by companies within the UK’s world class knowledge economy is already being used to underpin AI technologies, at present largely without permission or reward.  Before technology companies get the chance to batter down the doors in Whitehall to integrate their AI into every aspect of the economy and the civil service, the next government needs to put in place a clear and transparent framework that supports a sustainable value exchange between the creators and commercial users of that high quality knowledge and information.

24

1 Comment

EOS at Federated Hermes Limited has just released its Q3 2024 Public Engagement Report. In the cover feature this quarter, engagers Michael Benedict Yamoah, Navishka Pandit and Judi Tseng look at how physical risk insurers have struggled to respond to the increasing number of extreme weather events such as flooding, as the climate rapidly changes. At the other end of the scale, unchecked extraction of water for industrial purposes in regions of water scarcity can create problems for local communities and wildlife. Joanne (Jo) B. and George Watson explain how we engage with companies on water management to address these concerns. As EOS celebrates its 20th anniversary this year, we surveyed clients, colleagues and EOS alumni to assess how the stewardship landscape has changed, and how it might evolve over the next two decades. Amy D'Eugenio reports on the findings. Finally, Shoa Hirosato identifies the key trends from the 2024 voting season in developed Asia and emerging markets, assessing progress on board independence and diversity, and climate-related risk. Thanks to other contributors this issue, including Dana Barnes, Justin Bazalgette CEnv CMgr MBA BEng, Will Farrell, Richard Adeniyi-Jones, George Clark and Andrew Glynne-Percy. #stewardship #globalwarming #climatecrisis #climatechange #corpgov #water #netzero #biodiversity #nature #ESG #impact https://bit.ly/3YIOwob

17

1 Comment

New from me: The Competition and Markets Authority (CMA) does not believe forcing Vodafone & Three to sell off spectrum will alleviate its competition concerns for their proposed merger. It also takes issue with the complexity of spectrum divestment, instead preferring investment commitments. “An MNO entrant remedy would be high risk in the UK and would not present a sufficient degree of certainty of achieving its intended effect, i.e. the entry of a new MNO that compensates for the loss of competition resulting from the Merger,” the paper says. Citing advice from the regulator, Ofcom, the CMA says that spectrum divestment remedies in other countries have been “relatively unsuccessful in restoring a fourth operator,” adding that it remains difficult for any new operator to gain market share, “regardless of how much spectrum is divested”. The paper cites the US as an example of the uncertain impact of this remedy. In 2019, Sprint agreed to sell its 800 MHz spectrum portfolio to Dish Network as part of a merger with T-Mobile. But Dish has yet to emerge as a competitive fourth operator.

7

Pleased to see the UK regulator cleared Amazon's investment in Anthropic (they're both partners of Robin AI). There's plenty of folks in the AI world with dodgy claims and motives and governance -- so it's right to scrutinize innovative partnerships to check they aren't designed to, or capable of, crushing competition. But it's also the case that investing in building safe AI is expensive. We couldn't do it without partners, and neither can a bigger company like Anthropic. The presence of competition in markets can be measured in many ways: from price levels, to the pace of innovation, and also the number of competitors -- and that last example means it's important for regulators watch to ensure that as many as possible can compete in AI development: not just those with the deepest pockets or most gullible VC investors. #competition #antitrust #AI https://lnkd.in/eUCjbneu

32

1 Comment

Yesterday Press Gazette Charlotte Tobitt picked up on a letter I wrote to the UK House of Lords #digital #communications committee correcting the record on a couple of things. https://lnkd.in/ewfz66nA My letter wasn't about AI Overviews - though we all look forward to seeing the much trailed data on user click throughs. It didn't relate to tags like NOSNIPPET to remove snippets from search - the negative user and traffic impacts of doing so were outlined in a now deleted Google Europe blog. [https://lnkd.in/ee4bs5Vr] My point related to Google appearing to sell access to its search index to third party AI developers like Meta to use for the commercial purpose of Retrieval Augmented Generation (RAG). RAG enables AI models to produce outputs that are up to date and accurate. Without RAG, AI chatbots become unreliable, or just make stuff up to fill the gap in their probabilistic database of responses. The concern I raised was that given the need for publishers to continue to appear in legacy search results, there is little practical option for publishers to opt-out of their IP being sold as part of RAG services for third party LLMs. That inability to control how IP is used risks the commoditisation of RAG at a time of growth in a commercial market of the future. We're a week or so away from the Government publishing a paper on copyright and TDM exceptions. The picture on how news and creative IP is being used without informed consent is incredibly confusing. The new Digital Markets Unit may be able to help. Weakening copyright certainly will not. I'm sure many businesses that invest in IP will be pleased to set out to government how the present use of IP without compensation is unsatisfactory. There are huge growth opportunities for the UK's world leading creative and news industries to underpin exciting new AI products and services - if we get a licensing regime right.

30

🌟🗳️Check out our Election Hub not just for our latest analysis of what the election means for tech, but for a whole range of useful resources including our online briefing hub! More info down below 👇

14

The US presidential/ UK PM debates have never taken place in the same week, let alone consecutive days. Setting aside which was the more entertaining (easily US) – a few thoughts on how the different broadcast formats affected the coverage and impacted the viewer: ⏳ Ground rules & visuals Contrasting with the BBC's looser approach, #CNN set explicit rules of engagement, everything from response times, microphone control and stage positioning. The split screen close-up camera effect was revealing – Biden pensive then growing in confidence and Trump looking increasingly needled as the debate progressed. If a similar technique was adopted for the UK debate, it wouldn't have benefited Starmer... 🔊 Moderation & questions Dana Bash and Jake Tapper led for CNN vs Mishal Husain alone for the #BBC. Two moderators enabled closer control and smoother transition through questions. Lack of audience in Atlanta meant a strangely sanitized atmosphere but removing audience reactions – along with the use of zoom-in – perhaps helps the viewer to focus more on the candidates and be less influenced by variables. The debate in Nottingham was less structured, giving both men – mainly Sunak – more chance to interrupt and go on the attack. There was also something very British about a willing audience stumbling through their questions amidst shouts of protest from outside! 🎯  Fact-checking Husain planted the flag early by highlighting BBC Verify and both candidates appealed to the service when disputed claims arose. She was also more assertive at clarifying statements and presenting official statistics especially during the immigration and fiscal debates. There was little pre-emptive or real time fact-checking in Atlanta, for which CNN are being heavily criticized. In the closing stages, the moderators struggled just to get the candidates – especially Trump - to engage with the question asked. Personal accusations, counter-claims and a ‘he started it’ line of attack increased as the debate wore on. As for the overall optics, Trump and Sunak appeared the more forceful and vigorous. Biden and Starmer will have to hope their more measured, statesmanlike performances are not overly perceived as tired or meek. #presidentialdebate #2024election

16

The Times 🗞 + Fairer Finance 💷 + Citizens Advice 👫 + Instinctif Partners 🤝 = an event not to be missed if your brand is grappling with reputational challenges in highly regulated sectors. Sign up below to save your seat and put your questions to our expert panel on 12 November ⤵ #reputationmeetsregulation

12

1 Comment

Suppliers to CogX — one of the UK’s largest tech conferences — have told Sifted they’re resorting to the courts and turning to debt collectors in an attempt to recoup money they say they are owed. Following up on a story we published in April looking into claims that suppliers to CogX were owed "tens of thousands" in unpaid fees, more reached out to Sifted to say they’re overdue payment from the event. Five suppliers have now contacted Sifted to say they were owed money by CogX. Sifted approached a sixth after an anonymous tip off. Many are now intensifying their efforts to recoup debts as CogX gears up for its next London event on October 7. There are currently eight court orders, dating back to 2022, for CogX to pay creditors money — ranging from £2k-£15k — with six issued since the start of this year, according to TrustOnline, a service setup by the Ministry of Justice to provide access to court records. One supplier, which says they are owed “almost six figures", hired a debt collector earlier this year. A CogX spokesperson told Sifted that it acknowledges “a number of suppliers are awaiting payment” and that it apologises for the inconvenience caused. “CogX is in the process of raising a substantial new funding round to clear creditors and provide capital for long term growth of its CogX events business,” the spokesperson said. “We remain firmly committed to making sure that all outstanding invoices are paid in full." Read the full piece from Freya Pratty and myself here:

6

Drawing on my NYC emergency response days in the Mayor's office, I wrote about the role of debriefing and learning from previous experiences to inform the next one. Some may know it as the "after-action report." I provide a few recommended tips to build your readiness and infrastructure as we head into increasingly turbulent times. But hands down the biggest step any CCO or CMO can take is to build a collaborative culture across teams so that you don't fall victim to unforced errors.

93

1 Comment

Earlier, I commented on a (leaked) draft of Council Conclusions on Digital Policy. At that stage, there were many elements to support, including the call to focus on implementation rather than open new files on tech regulation...although later on in the same document, there were calls to do exactly the opposite, to regulate this or that aspect of the digital sector. Along the way, it has been great to see more consistency. And ahead of the approval of the Conclusions next week, new leaks bring some colour to what different countries have highlighted in the process. Great to see, for example, that: - Germany wants to extend the role of the European Centre for Algorithmic Transparency (ECAT) beyond the Digital Services Act (DSA), including in assessing the relevance of future digital initiatives. This is something we at Wolt have also mentioned - algorithmic transparency shouldn't be limited to VLOPs and one regulation! We will soon publish the 3rd edition of our Algorithmic Transparency report - Estonia has been pushing for further reducing compliance costs for smaller businesses Hopefully, the final outcome reflects well also the priorities of the European Tech Alliance - European tech can support the push to improve our competitiveness (and we need to improve the operating & regulatory environment for European tech), better regulation that is fit for purpose, and better enforcement.

25

Section 3.1 of the Apple Intelligence LLMs model card notes how Apple has scraped training data without the knowledge of permission of rights owners. https://lnkd.in/e8BC9QGS The card notes that, "AFM pre-training dataset consists of a diverse and high quality data mixture. This includes data we have licensed from publishers, curated publicly available or open-sourced datasets, and publicly available information crawled by our web-crawler, Applebot." For the avoidance of doubt - Microsoft execs have repeated the same line in recent weeks - just because data is publicly available on publisher properties does not mean it can be scraped and used for commercial purposes. The fact that Apple and Microsoft make their devices "publicly available" in their retail properties on Regent Street does not entitle me to go in an take one without paying. The card goes on to suggest that Apple respects "the right of webpages to opt out of being crawled by Applebot, using standard robots.txt directives." Many publishers might question whether providing the identity of this bot AFTER data had been scraped by Apple, is a moral or ethical way to approach to data collection. But the kicker really comes in relation to data privacy. The card notes that given Apple's "focus on protecting user privacy, we note that no private Apple user data is included in the data mixture." This seems to suggest that data privacy laws only apply to the use of data that Apple holds on a first party basis. Yet it is not clear that the scraping and use of "publicly available" data (whether scraped by Apple of acquired from a third party scraper) that was published for one purpose, only to be stored, processed and used by Apple for a different commercial purpose, exempts them from data privacy obligations?

24

1 Comment

Quoted in this Graham Lanktree / POLITICO analysis of the UK's current approach to Chinese EVs. In short: the time for decisions on EV tariffs is approaching. That will involve discussing trade offs - which British politicians are remarkably bad at doing - and showing your homework. What's the balance between 'security' v 'green', or the trade-off between 'climate concerns' v 'data exploitation'? How do you communicate that to your voters, peers and industries? Good thing it's not me in the driver's seat (sorry) https://lnkd.in/eTsahCVN

29