yahoo news | The Man Who Thought He Could Keep AI Safe

Demis Hassabis, the co‑founder of DeepMind, has spent his career trying to shape artificial‑intelligence into a force that benefits humanity while averting the existential threats it could pose. From the moment Google acquired DeepMind in 2014, he insisted on strict safety controls, an independent oversight board, and a ban on military applications, framing his work as a moral quest rather than a purely commercial venture. Early on he even imagined a “singleton” scenario—a secret bunker‑based research enclave that would develop superintelligence under a single, trusted authority, protecting the world from a rogue AI.

When the vision of a unified, safe AI effort began to crumble—sparked by Elon Musk’s founding of OpenAI and deep‑seated competition among tech giants—Hassabis accelerated DeepMind’s progress, achieving landmark breakthroughs such as AlphaGo and protein‑folding predictions that earned a Nobel Prize. Yet his attempts to embed nonprofit‑style governance inside Google faltered, and a controversial NHS partnership triggered privacy backlash, leading to internal retreats and the departure of co‑founder Mustafa Suleyman. The emergence of powerful chatbots like ChatGPT forced Hassabis to confront a new moral crossroads: either cede the chatbot market to rivals or engage in an all‑out “wartime” race for AI dominance.

By 2026 Hassabis has accepted that formal governance structures alone cannot guarantee safety; instead he seeks personal influence at the highest decision‑making tables. He argues that real safety comes from having a seat at the table when critical issues arise, allowing him to steer outcomes while adapting from idealism to pragmatic realism. Though his original values appear compromised, his presence offers a modest “scaffolding of reassurance” amid a global, multi‑lab AI arms race, until broader governmental restraints can be imposed.

Read more: https://www.theatlantic.com/ideas/2026/03/ai-google-deep-mind-hassabis/686527/

#demishassabis #deepmind #google #alphago

800.000 Neuronen lernen in 5 Minuten Ping-Pong – und spielen #Doom.
Klingt wie #Sci-Fi? Ist aber echte Forschung von Cortical Labs. DishBrain lernt schneller und energieeffizienter als #DeepBlue oder #AlphaGo. Revolutioniert biologisches Computing nun KI?
Der #BlogArtikel von Henrik Müller ist kostenlos online verfügbar: https://www.laborjournal.de/blog/

#Laborjournal #LifeSciences #Neurowissenschaften #BioComputing #neuronalePlastizität #KI

ML_Bear (@MLBear2)

제프리 힌튼의 주장 요약: 스케일링 법칙은 한계에 가까워지고 있으며, 가치 높은 데이터는 기업 내부에 갇혀 공개 인터넷은 소진되었음. 해결책으로 모델이 자체 트레이닝 데이터를 생성하는 방안을 제시하며, 이는 AlphaGo가 인간을 이긴 방식과 유사하다고 언급.

https://x.com/MLBear2/status/2003970497873432911

#research #scalinglaws #geoffreyhinton #alphago

https://blog.gslin.org/archives/2025/12/24/12817/alphago-%e6%9d%8e%e4%b8%96%e4%b9%ad%e7%89%88-%e8%a6%81%e5%be%a9%e5%87%ba%e8%b7%9f%e7%94%b3%e7%9c%9f%e8%ab%9d%e5%b0%8d%e5%a5%95%ef%bc%9f/

AlphaGo (李世乭版) 要復出跟申真諝對奕？

#alphabet #alphago #deepmind #go #google