Despite his many accomplishments, at the age of 60, Schmidhuber feels mounting time pressure toward building an Artificial General Intelligence within his lifetime and remains committed to pushing the boundaries of AI research and development.
, CEO of OpenAI himself, calling for AI regulation. From your perspective, is there an existential threat?It is true that AI can be weaponized, and I have no doubt that there will be all kinds of AI arms races, but AI does not introduce a new quality of existential threat. The threat coming from AI weapons seems to pale in comparison to the much older threat from nuclear hydrogen bombs that don’t need AI at all.
You said something important, in that some people prefer to talk about the downsides rather than the benefits of this technology, but that's misleading, because 95% of all AI research and AI development is about making people happier and advancing human life and health. Let’s take the example of a technology, GANs, General Adversarial Networks, which today has been used in applications for fake news and disinformation. In actuality, the purpose in the invention of GANs was far from what it is used for today.Yes, the name GANs was created in 2014 but we had the basic principle already in the early 1990s. More than 30 years ago, I called it artificial curiosity. It's a very simple way of injecting creativity into a little two-network system.
Through this zero-sum game the first network gets better and better at producing these convincing fake outputs, which look almost realistic. So, once you have an interesting set of images by Vincent Van Gogh, you can generate new images that leverage his style without the original artist having ever produced the artwork himself.
In the beginning, however, the model of the world knows nothing, so how can we motivate the first network to generate experiments that lead to data that helps the world model learn something it didn’t already know? That’s what artificial curiosity is about. The dueling two-network systems effectively explore uncharted environments by creating experiments so that over time the curious AI gets a better sense of how the environment works.
Jones: How long have these AIs, which can set their own goals, existed? To what extent can they be independent of human interaction?Neural networks like that have existed for over 30 years. My first simple adversarial neural network system of this kind is the one from 1990 described above. You don’t need a teacher there; it's just a little agent running around in the world and trying to invent new experiments that surprise its own prediction machine.
United States Latest News, United States Headlines
Similar News:You can also read news stories similar to this one that we have collected from other news sources.
Source: DenverChannel - 🏆 239. / 63 Read more »
Source: KSLcom - 🏆 549. / 51 Read more »
Source: FOX29philly - 🏆 570. / 51 Read more »
Source: nypost - 🏆 91. / 67 Read more »
Source: Refinery29 - 🏆 26. / 68 Read more »
Source: ABC7Chicago - 🏆 284. / 63 Read more »