tags. We'll keep them. Let me write the translation paragraph by paragraph. Original first paragraph: "
Microsoft CEO Satya Nadella warned that any company surrendering control of its artificial intelligence data risks ceasing to exist as a firm. In a July 12 blog post and subsequent interview with Fareed Zakaria, Nadella laid out what he called a 'reverse information paradox' — businesses pay for AI intelligence twice, once with money and again with their proprietary knowledge.
" Translation: "Microsofts CEO Satya Nadella advarede om, at enhver virksomhed, der giver afkald på kontrollen over sine AI-data, risikerer at ophøre med at eksistere som virksomhed. I et blogindlæg den 12. juli og et efterfølgende interview med Fareed Zakaria fremlagde Nadella det, han kaldte et 'omvendt informationsparadoks' — virksomheder betaler for AI-intelligens to gange, én gang med penge og igen med deres egen viden.
" Second paragraph: "The 'reverse information paradox'
Nadella argued that companies currently pay for AI services in cash and also feed the systems their own data, which the models then learn from. That means the same firm that buys AI tools also gives away the unique information that makes its business valuable. 'A firm that gives up control of its AI data will not remain a firm,' Nadella wrote. He said outsourcing a company's 'thinking' to an AI provider without retaining control of the resulting data is a fundamental mistake.
" Translation: "Det 'omvendte informationsparadoks'
Nadella argumenterede for, at virksomheder i dag betaler for AI-tjenester med kontanter og samtidig fodrer AI-modellerne med deres egne data, som modellerne så lærer af. Det betyder, at den samme virksomhed, der køber AI-værktøjer, også giver afkald på den unikke information, der gør dens forretning værdifuld. 'En virksomhed, der giver afkald på kontrollen over sine AI-data, vil ikke forblive en virksomhed,' skrev Nadella. Han sagde, at det er en grundlæggende fejl at outsource en virksomheds 'tænkning' til en AI-udbyder uden at bevare kontrollen over de resulterende data.
" Third paragraph: "Metadata as a new asset class
In the interview with Zakaria, Nadella said companies should keep metadata from every interaction with an AI model. That data, he argued, can be used to train the company's own models over time. Without it, the firm loses the ability to improve its own AI systems and becomes dependent on the provider.
" Translation: "Metadata som en ny aktivklasse
I interviewet med Zakaria sagde Nadella, at virksomheder bør gemme metadata fra hver interaktion med en AI-model. Disse data, argumenterede han, kan bruges til at træne virksomhedens egne modeller over tid. Uden dem mister virksomheden evnen til at forbedre sine egne AI-systemer og bliver afhængig af udbyderen.
" Fourth paragraph: "Legal changes needed
Nadella called for new legal protections for AI data buyers. He noted that existing laws — patents, copyrights, trademarks — protect the sellers of intellectual property. But AI, he said, is the first technology where the buyer also needs protective rights. 'We need legal changes,' he said, without specifying exactly what those changes should look like.
" Translation: "Juridiske ændringer nødvendige
Nadella opfordrede til nye juridiske beskyttelser for købere af AI-data. Han bemærkede, at eksisterende love — patenter, ophavsrettigheder, varemærker — beskytter sælgere af intellektuel ejendom. Men AI, sagde han, er den første teknologi, hvor køberen også har brug for beskyttelsesrettigheder. 'Vi har brug for juridiske ændringer,' sagde han, uden at specificere præcis, hvordan disse ændringer skulle se ud.
" Fifth paragraph: "A technical route to control
Nadella also outlined a technical approach: keep the 'harness, context, and memory' separate from any single AI model. That way, a company can switch between models without losing its proprietary data or the ability to fine-tune its own AI. The goal, he said, is to maintain control over the data pipeline even when using third-party AI services.
" Translation: "En teknisk vej til kontrol
Nadella skitserede også en teknisk tilgang: hold 'værktøj, kontekst og hukommelse' adskilt fra en enkelt AI
