
Hi and welcome to AI. In this edition … news media is struggling with AI; Trump mandates us the safety efforts of AI to focus on the fight against “ideological bias”; Distributed training gains increasing traction; The increasingly strong AI could incline the scale to the totalitarianism.
AI is a potentially disturbing business models of many organizations. In several sectors, however, the threat is as seemingly existential as intelligence. It will become a company in which I am, so I hope you forgive a somewhat benevolent newsletter. However, the reports should be up to all of us, because the functioning free prints play a crucial role in democracy – it develops the public and helps to keep power to account. And there are some similarities between the manager of the workers – and critically – do not provide challenges and opportunities that AI represents, of which the leaders of enterprises in other sectors can also learn.
Last week I spent a day at the Aspen Institute called “AI & News: Course Mapping”, which was hosted at Reuters headquarters in London. The conference was attended by the best executives from a number of British and European intelligence organizations. This took place according to the Chatham House rules, so I can't tell you who said exactly what, but I can pass on what was said.
Tools for journalists and editors
The reports spoke about using AI primarily in internally oriented products to make their teams more efficient. AI helps write search machine optimized subtitles and translate content-library organizations Reach a new audience In places she traditionally did not serve, although most emphasized the maintenance of people in a loop to monitor accuracy.
One editor described by AI to automatically produce short articles from press releases and liberated journalists for more original messages, retaining human quality control editors. Journalists also use AI to summarize documents and analyze large data sets – such as landfills of government documents and satellite images – and allow investigative journalism that would be difficult without these tools. These are cases of good use, but lead to a modest impact – mostly from the efficiency of existing workflows.
Bottom up or from top to bottom?
Among the leaders of editors and technicies have appeared an active debate on whether intelligence organizations should approach access from bottom-up-up-to-date generative tools in the hands of every journalist and editor, allowing these people to operate your own data or “Vibe code” Widgets driven by AI that help them in their work or whether effort should be down from top, leading the management.
The bottom-up approach has merit-democratizes access to AI, seizes first-line employees who often know pain points and can often see good cases than the executors at a high level, and free the limitation of AI developers to be spent only on projects that are larger, more complex and more important.
The disadvantage of the bottom -up approach is that it can be chaotic, which makes it difficult for the organization to ensure adherence to ethical and legal policies. It can create a technical debt, and the tools are built at a run that cannot be easily maintained or updated. One editor feared the creation of a two -stage editorial office, some of which the editors received a new technique and others lagging behind. Also, the bottom does not ensure that the solution creates the best return on investment-shell consideration, as AI models can quickly be expensive. Many of them demanded a balanced approach, although there was no consensus on how to achieve this. From the conversations I had with the executors in other sectors, this dilemma is known across industries.
Alerts of threatening trust
Messages are also cautious when building AI tools oriented to the audience. Many of them have begun to use AI to produce articles that can help busy and increasingly impatient. Some built chatbots AI who can answer questions about a particular narrow subset of their coverage – such as stories about the Olympics or climate change– But they tended to call them “experiments” to help readers that the answers may not always be accurate. Little in terms of the content of generated AI. They are afraid that hallucinations produced by AI gene undercut confidence in the accuracy of their journalism. Their brands and their businesses eventually depend on this trust.
Those who hesitate will be lost?
This caution, although understandable, is in itself a colossal risk. If the intelligence organizations themselves do not use AI to summarize messages and make them more interactive, technology companies are. People are increasingly turning to AI and Chatbota search engines, including confusion, Openai's Chatgpt and Google Gemini, and Google's “AI” overviews now provide many search and many more. Several executives at the conference said that “Disintermediac” – a loss of direct connection with their audience – was their biggest fear.
They have a reason to be afraid. Many news organizations (including Luck) They are at least partially dependent on Google search to bring the audience. A recent Tollbit study—KO that sells software that helps protect websites from web search engines – see that clicks for Google AI reports were 91% lower than the traditional Google search. (Google has not yet used AI reports for intelligence questions, although many think it is only a matter of time.) Another study of clicks from Chatbot's conversations are equally abysmal. Cloudflarewhich also offers help protect the publisher of news from scratching on the web, found it Openi scraped the news site 250 times for each page of the recommendation that this site sent.
The intelligence organizations have so far responded to this potentially existential threat through a mixture of legal reverse – New York Times sued Openi for violating copyrights while Dow Jones and New York Post they sued the confusion– and partnership. These partnerships included multi -year, seven -digit license stores for news content. (Luck Many of the executions at the conference have said that the license agreements were a way to earn the income of content that technology companies have probably already “stolen”. They also saw the partnership as a way to build relationships with technology companies and use their expertise to help them build AI products or train their employees. No one saw relationships as particularly stable. Everyone was aware of the risk of becoming too dependent on AI licensing income because they were previously burned when the media industry let Facebook become the main driving force of operation and advertising from advertising. Later this money disappeared virtually overnight when Meta After the US presidential election in 2016, CEO of Mark Zuckerberg decided to de-emphasize news in people.
Ferrari powered by an artificial intelligence
The executives have recognized that they need to build direct relationships in the audience that cannot be divided by AI, but few of them had clear strategies. One expert at the conference honestly said that “the intelligence industry does not take AI seriously”, which focuses on “incremental adaptation rather than structural transformation”. He compared the current approaches to the three -stage process, which had “Ferrari powered by AI” at both ends, but “a horse and a cart in the middle”.
He and other media industry advisors urged news organizations to get from the structuring of their approach to reports around “articles”. Instead, they encouraged executives to think about ways in terms of source materials (public data, transcripts of interviews, documents obtained from sources, raw videos, sound recordings and archive intelligence stories) could turn into a series of outputs, short-term video, waste summary or traditional news. with generation of AI technology. They also urged news organizations to stop thinking about the production of messages as a linear process, and began to think about it as a circular loop, perhaps the one in which there was no man in the middle.
One person at the conference said that intelligence organizations had to become less island and look more at knowledge and lessons from other industries and how they adapted to AI. Others said it might require startups – perhaps incubated by intelligence organizations themselves – to promote new business models for AI age.
Bets could not be higher. While AI represents existential challenges for traditional journalism, it also offers unprecedented opportunities to expand the range and potentially reopen with the audience who “turned off messages” – if the leaders are courageous enough to introduce what messages can be in the AI era.
There's other AI news.
Jeremy Kahn
Jeremy.kahn@fortune.com
@Jeremyakahn
Repair: Last week Tuesday's Eye On AI edition Incorrectly identified the country where TrustPilot has a seat. It's Denmark. The message in this edition also Miss appointed the name of the Chinese startup for the AI Manus viral model. The start name is the effect of butterflies.
This story was originally listed on Fortune.com