Summary of Edward Snowden and Ben Goertzel on the AI Explosion and Data Privacy

00:00:00 - 00:30:00

In this video, Edward Snowden and Ben Goertzel discuss the implications of AI for data privacy and the need for a decentralized ecosystem to prevent the misuse of AI by malevolent elites. Snowden emphasizes the violation of human rights caused by mass surveillance, and warns that increasing power concentration into fewer hands, alongside facial recognition and unique identifiers, could lead to accountability issues and mistakes. The two discuss the limitations of machines in grasping fundamental concepts and the potential dangers of using unquantifiable data sets for deep learning. Despite the challenges, they remain optimistic that AI can be harnessed for good rather than evil.

• 00:00:00 In this section, the host introduces his guests and sets the stage for the discussion on AI and data privacy, with Ben Goertzel and Edward Snowden. Snowden discusses his take on the increasing capacity of AI and its implications for data privacy, pointing out that mass surveillance and data gathering are omnipresent, whether it's government or private. Snowden emphasizes that these practices are violations of basic human rights and that the authorities involved did not stop, but instead continue to do it even more now. He questions what lesson others have learned from this impunity for lawbreaking and the policy assumption that this is necessary and legitimate.
• 00:05:00 In this section, Snowden describes how the NSA's XKeyscore used a Federated query system to decentralize data and process queries against all of it, with only the results being sent back. He explains that the decentralization was being used for both good and evil, as the centralized system relied on a crude bash-scripting type system, whereas decentralization enabled them to automate their process. He warns that we're seeing more power being concentrated into fewer hands, and with the rise of facial recognition, license plate readers, and unique identifiers like phone numbers and email addresses, we need to move away from the machine doing the thinking and reasoning probabilistically, as there may be accountability issues leading to mistakes. Additionally, Ben Goertzel explains why big corporations like Google and Meta have an advantage in building AI models, as there are many different approaches in the AI field that have been prioritized by big tech.
• 00:10:00 In this section, Snowden and Goertzel discuss the rise of neural net architectures and the data-intensive nature of AI algorithms. Big companies have an advantage over others because they have more money for processing and loads of data, which is necessary for current neural net models. Government agencies have the data, but they don't have sophisticated tech teams to gather intelligence, as proven with their lack of intelligent mining of data. As AI works better, various government agencies could have open-ended searches against all the gathered data with more facility than before, but this entails either surveillance or surveillance. Lastly, possibilities arise in which individuals can encrypt data with different AI protocols so that different people have differing levels of visibility over the information.
• 00:15:00 In this section, Snowden and Goertzel discuss the potential of AI technology and the implications it holds for data privacy. They acknowledge that current AI technologies have limitations, but they also see how they can be overcome to create hybrid cross Paradigm AI systems that are even smarter than large language models. However, they also highlight the anti-social behavior of leaders in the space and the difficulty in legislating restraint on these forces, particularly since we are becoming more legible and malleable to corporations and governments. Snowden notes that policy prescriptions legislatively will be difficult, but they are looking towards decentralized infrastructure for beneficial applications for the good of humanity.
• 00:20:00 In this section, Edward Snowden and Ben Goertzel discuss the idea of open-source AI models, which would grant companies exclusive commercial usage rights for a limited time. After that period, companies could still use the models but would have to publish their models freely for academic or non-commercial use. Snowden argues that this model would help control institutions controlling us, while Goertzel points out that control over AI is key while it is still under human power structures. The two discuss the capabilities of government spy agencies and how access to similar capabilities could be valuable for society. Ultimately, they touch on the subject of super-intelligence and the current AI landscape.
• 00:25:00 the heart of the AI discussion the issue of control. Ben Goertzel believes that the only way to prevent the misuse of AI by malevolent elites is to create a decentralized ecosystem where the smartest and most effective AI is also based in a decentralized platform, so that it is not easily controlled. He also notes that data must be volunteered and underpinned by cryptographic technology that gives individuals sovereignty and transparency over their data. Edward Snowden agrees with him but thinks that researchers should concentrate on training the machines to be better than humans, not just a replica of us. He warns that this is not an easy task as it requires a fundamental change of mindset from the AI safety community. Despite the challenges, both are optimistic that AI can be harnessed for good rather than evil.
• 00:30:00 In this section, Edward Snowden and Ben Goertzel discuss the limitations of machines in grasping fundamental concepts and the potential dangers of using unquantifiable data sets for deep learning. While machines can learn from observation in the same way humans do, they do not understand the costs of their actions, which arise from our personal experiences of pain and growth. The over-reliance on deep learning techniques that use giant unquantifiable data sets, without understanding how machines make decisions, creates a potential danger in situations where machines make incorrect decisions. Snowden and Goertzel suggest alternative models to deep learning and emphasize the importance of teaching machines from better data inputs.