Thanks for writing this, it realy clarifies the vision you've been sketching in previous parts. How do we ensure this meaningful control scales robustly to the wider global society?
Glad you liked it! I would look at your question in two ways:
(1) The political element will be really important. There will always be human bad actors, so those with the most power need to align on consistently implementing pro-control infrastructure and enforcing its use.
(2) There will always be errors and mistakes as you implement and scale anything complex. Strong defence in depth is critical, so that mistakes have as limited an impact as possible. And you want a pervasive culture of error-correction, so that you learn properly and grow more robust after survivable errors.
Ultimately, you need to system to consistently improve over time as the AIs get more powerful and the stakes higher. And if a perfect solution does exist out there somewhere, hopefully you eventually find it.
Thanks for writing this, it realy clarifies the vision you've been sketching in previous parts. How do we ensure this meaningful control scales robustly to the wider global society?
Glad you liked it! I would look at your question in two ways:
(1) The political element will be really important. There will always be human bad actors, so those with the most power need to align on consistently implementing pro-control infrastructure and enforcing its use.
(2) There will always be errors and mistakes as you implement and scale anything complex. Strong defence in depth is critical, so that mistakes have as limited an impact as possible. And you want a pervasive culture of error-correction, so that you learn properly and grow more robust after survivable errors.
Ultimately, you need to system to consistently improve over time as the AIs get more powerful and the stakes higher. And if a perfect solution does exist out there somewhere, hopefully you eventually find it.