Oly, I enjoyed your talk at the Technical AI Safety (TAIS) conference and this essay that corresponds to it. As a novice in AI with a QA background, this provides a nice background for me. May I ask about loss of control? I agree once an AI is replicating itself and once it gets to a certain level of automation, we have lost control. If the policy makers in Washington, Beijing, or elsewhere are not paying attention, not caring, or hoping for this level of AI control, does any of this matter? I do not mean to be hopeless, but realistic and concerned about our collective future.
A concern! This shows that the 'safety' systems keeping democratic representatives aligned with society's wellbeing are, to some extent, compromised. (We knew this already, of course, it's just that the stakes are rising - i.e. the 'environment' is getting more hazardous.)
From this angle, we might want to focus attention on engineering better control at the society-government interface. How? To grab my rule of thumb from this article: look at society's sensing, understanding, deciding, and acting apparatus. How are they compromised? How might they be fortified? That motivates many potential interventions - among them: 'AI for Human Reasoning'. https://www.oliversourbut.net/p/ai-for-human-reasoning-for-you
Another thing to look at is: recognising our roles in the existing decision-making apparatus, are there communications we can make (letters, essays, speaking to friends, journalists, politicians, ...) which will move that societal decision-making forward productively?
Similarly, are there control actions we have available (protest, boycott, voting, ...) but haven't exerted yet, or not to the appropriate extent?
For me, I'm working on AI for human reasoning since that looks like a good fit for my skills and team for now. I also participate in conversations with political decision-makers from time to time when I'm invited.
The 'AI for Human Reasoning' essay is intriguing to think about, using compute and existing technology to increase and ultimately intersect AI into human biology. Bringing all of us collectively into the future aligned with AI, which thinking about out loud means that the political policies and guidelines may be less of a concern, but still a concern, assuming we are all working together with the AI systems interchangeably.
Glad to hear you have had conversations with political decision-makers. I am hopeful for the mid-terms in the US, but with a probably Republican-led senate and current president, I don't see much progress toward AI regulation.
Oly, I enjoyed your talk at the Technical AI Safety (TAIS) conference and this essay that corresponds to it. As a novice in AI with a QA background, this provides a nice background for me. May I ask about loss of control? I agree once an AI is replicating itself and once it gets to a certain level of automation, we have lost control. If the policy makers in Washington, Beijing, or elsewhere are not paying attention, not caring, or hoping for this level of AI control, does any of this matter? I do not mean to be hopeless, but realistic and concerned about our collective future.
A concern! This shows that the 'safety' systems keeping democratic representatives aligned with society's wellbeing are, to some extent, compromised. (We knew this already, of course, it's just that the stakes are rising - i.e. the 'environment' is getting more hazardous.)
From this angle, we might want to focus attention on engineering better control at the society-government interface. How? To grab my rule of thumb from this article: look at society's sensing, understanding, deciding, and acting apparatus. How are they compromised? How might they be fortified? That motivates many potential interventions - among them: 'AI for Human Reasoning'. https://www.oliversourbut.net/p/ai-for-human-reasoning-for-you
Another thing to look at is: recognising our roles in the existing decision-making apparatus, are there communications we can make (letters, essays, speaking to friends, journalists, politicians, ...) which will move that societal decision-making forward productively?
Similarly, are there control actions we have available (protest, boycott, voting, ...) but haven't exerted yet, or not to the appropriate extent?
For me, I'm working on AI for human reasoning since that looks like a good fit for my skills and team for now. I also participate in conversations with political decision-makers from time to time when I'm invited.
Thanks for the reply!
The 'AI for Human Reasoning' essay is intriguing to think about, using compute and existing technology to increase and ultimately intersect AI into human biology. Bringing all of us collectively into the future aligned with AI, which thinking about out loud means that the political policies and guidelines may be less of a concern, but still a concern, assuming we are all working together with the AI systems interchangeably.
Glad to hear you have had conversations with political decision-makers. I am hopeful for the mid-terms in the US, but with a probably Republican-led senate and current president, I don't see much progress toward AI regulation.
Sure, perhaps into human biology - but I'd say into our sociology first! Biohacking carries it's own risks.
Agreed, as does the sociology intertwining.
Ha! Touche