July 16th news, yesterdayGoogle Deepmind CEO Damien HasabisDemis HassabisThis post is part of our special coverage Syria Protests 2011AGI PROBABLY IN A FEW YEARS. IN HIS VIEW, AGI WOULD BECOME AN IMPORTANT POWER OR FIRE-LIKE TECHNOLOGY THAT WOULD PROMOTE DEVELOPMENTS IN AREAS SUCH AS DRUG RESEARCH AND DEVELOPMENT, CLEAN ENERGY AND ADVANCED MATERIALS, BUT ALSO ENTAIL RISKS SUCH AS CYBERSECURITY AND BIOSAFETY。
Hassabis called for the establishment of a front-line AI standard body responsible for developing and regularly updating model assessment criteria and testing the capabilities of front-line models in high-risk areas such as cybersecurity, biological threats, etc. Models that meet the corresponding benchmarks will be identified as "frontline models" and their development agencies as "frontline laboratories"。
According to his vision, the forward laboratory could initially submit a model assessment on a voluntary basis up to 30 days before the model was released; once the mechanism had matured, the testing could become a precursor to model deployment on the United States market。
The assessment should also focus on whether the model attempts to circumvent security restrictions, whether there are frauds and whether enterprises implement cybersecurity and safety studies。
Hassabis stated that at present competition in AI is accelerating and that technological progress has exceeded society ' s understanding of its risks, and that there is a need to promote innovation while strengthening security responsibilities and promoting international cooperation。