I’ve been thinking about AI Safety for roughly 5 months. The more I think deeply about it, how things may play out, possible outcomes, drivers of progress etc etc, the more I realize how bad I am at thinking about it. What’s dangerous is, it’s never immediately obvious that a scenario in my mind is actually not logical. There’s so many variables, factors to consider, new knowledge you consume, new things happening in the AI world that drastically shift my view. I think humans have a tendency to treat the things they read (especially if they’re niche) as prized knowledge that they have access to and can somehow use to get an advantage out of it. While this is how we develop better and unique viewpoints, what this also leads to is people unconsciously using these assumptions or “things they know to be true about the world” when they make predictions about the future. I once read a book called black swan and my biggest takeaway was that we suck at predicting things historically. This has influenced me to place less worth on timeline arguments but has also made me so much more uncertain about the future. As someone engaging with Effective Altruism ideas, I like to think about the most effective path forward in making an impact on the world. However, how do I choose the most effective path when I don’t know the destination? Well then, it may be a good strategy to form your own belief about the destination and the world which will form but inevitably, we will get complacent and not think deeply enough about some of the assumptions we make. This can be a dangerous model to operate out of. We like making long term commitments to things we care about or think are true about the world. For example, a PhD in a topic we think will be relevant in the future. But once faced with the bitter truth that something about our worldview at the time of making the commitment is something we don’t agree with any longer, we can’t help but think we’ve wasted our time. Maybe this is why our actions sometimes seem so stupid in retrospect. An open question I pose is: How then should we choose the best path forward?