The AI alignment problem is not like any other engineering challenge

Buying this for 3 days, hoping this will spark some interest. <br><br>Currently, I am studying for the Professional Machine Learning Engineer (ML Engineer) exam. I have spent thousands of hours studying AI, machine learning, and deep learning, and have worked on at least 100 projects, but this exam is still enough of a challenge to require intense focus. <br><br>Each time I need to research a topic, I have to consider the theoretical and practical aspects, as well as the business and societal implications. I have to be able to reason about how different systems will interact with each other, including how humans will interact with the systems. This is a difficult challenge, but not an impossible one.<br><br>However, when I read about AI Alignment, I feel like I am studying a completely different field. <br><br>There are hypotheses, theories, and models, but as an outsider, I don't see the same rigorous theoretical foundations as you do in ML. <br><br>I am not sure if this is due to the lack of reliable data, but I am having a hard time understanding the field. <br><br>The AI alignment problem is not like any other challenge in the field of AI, as it deals with complex and abstract concepts that are not yet well understood. <br><br>In contrast, other AI challenges, such as computer vision, natural language processing, and decision-making, have more established foundations and methods.<br><br>The alignment problem requires a deep understanding of human values and the potential risks and benefits of advanced AI systems. <br><br>It is a multidisciplinary field that draws on insights from AI research, cognitive science, neuroscience, philosophy, anthropology, and more.<br><br>Currently, there is no consensus on the best approach to aligning AI systems with human values, and different researchers have proposed a variety of methods, such as inverse reinforcement learning, imitation learning, and value-based reinforcement learning.<br><br>However, these methods are still in the early stages of development, and there is ongoing research into the most effective ways to align AI systems with human values.<br><br>Currently, alignment research is mainly theoretical, and there is a need for more experimental work.