The Scaling Hypothesis roughly says that current Deep Learning techniques, given ever more computing power, data, and perhaps some relatively minor improvements, will scale all the way to human-level AI and beyond. Let’s suppose for the sake of argument that the Scaling Hypothesis is correct. How would that change your forecasts or perspectives on anything related to the future of AI?
Would your forecasts for AI timelines shorten significantly?
Would your forecasts change for the probability of AI-caused global catastrophic / existential risks?
Would your focus of research or interests change at all?
Would it change your general perspective on the current and/or future of AI?
Would it change any forecasts or perspectives of yours in areas that aren’t AI themselves but which might be affected by AI?
Would it perhaps even change your perspective on life?
Would your forecasts for AI timelines shorten significantly?
Yes, by 10-20 years, in particular for the first human level AGI, which I currently forecast between 2045-2060.
Would your forecasts change for the probability of AI-caused global catastrophic / existential risks?
Not by much, I give a low estimate to an AI existential risk.
Would your focus of research or interests change at all?
Yes, in the same way that the classic computer vision field has been made pretty much obsolete by deep learning, apart for few pockets or for simple use cases.
Would it perhaps even change your perspective on life?
Yes, positively. We would get faster than expected to the commercialisation of AGI, shortening the gap to a post-scarcity society.
That said, I don’t believe to the scaling hypothesis. Even though NNs appear capable to simulate arbitrary complex behaviours, I think we will hit a wall of diminishing returns soon, making it impractical to proceed this way for the first AGI.