DeepMind Framed Early AI Safety Work in ML Language
Pedro Ortega recalls DeepMind's early strategy of framing AI safety in machine learning terms.
Pedro A. Ortega, former leader of the AI Safety Team at Google DeepMind, posted about the group's initial approach to the topic. He stated that AI safety was viewed as fringe at the time. The team therefore cast problems in machine learning language and reduced explicit references to philosophy. Ortega noted that a link to thinkers such as Foucault would not have been viable. The post references the Medium article 'Building safe artificial intelligence: specification, robustness, and assurance' written with Vishal Maini and the DeepMind safety team.
Combined views
59
2 posts, first seen 2h ago
DeepMind Framed Early AI Safety Work in ML Language
Pedro Ortega recalls DeepMind's early strategy of framing AI safety in machine learning terms.
Pedro A. Ortega, former leader of the AI Safety Team at Google DeepMind, posted about the group's initial approach to the topic. He stated that AI safety was viewed as fringe at the time. The team therefore cast problems in machine learning language and reduced explicit references to philosophy. Ortega noted that a link to thinkers such as Foucault would not have been viable. The post references the Medium article 'Building safe artificial intelligence: specification, robustness, and assurance' written with Vishal Maini and the DeepMind safety team.
Reactions
2 postsPA 
Pedro A. Ortega@AdaptiveAgentsAt the time, AI safety was still considered a fringe topic, not really a legitimate part of AI. Our strategy was to formulate the problems in ML language and minimize explicit references to philosophy. Establishing a link to, say, Foucault would have been a non-starter. If you…
- likes: 0
- replies: 1
- bookmarks: 0
- reposts: 0
SL Seth Lazar@sethlazar@AdaptiveAgents yep GDM safety (as well as sociotechnical alignment with William Isaac) would have been the exception that proves the rule for sure.
- likes: 0
- replies: 0
- bookmarks: 0
- reposts: 0