The task doesn’t require strong sim2real transfer. This post doesn’t summarise the report. 2021: Ramesh et al: Zero-Shot Text-to-Image Generation (first DALL-E from OpenAI; blog post). I’m pretty sure the “date of first git commit” and “date of final human git commit” and “date when the system’s first autonomously designed and owned power plant comes online” will each be later in time than the previous milestone. Several scientists and forecasters have been arguing for prioritizing early research into the possible benefits and risks of human and machine cognitive enhancement, because of the potential social impact of such technologies. Before this point, human deployment decisions (influenced by regulation, general caution, slow decision making, etc) limit AI’s impact; afterwards AIs forcibly circumvent these decisions. Many of the above points, on both sides, apply more weakly to the impact of AI on AI R&D than on the general economy. Both of these reasons are more likely to apply if 20%-AI is hard to develop, i.e. if timelines are long. Once there’s been enough algorithmic progress, and training runs are big enough, we can train AGI. My best guess, defended in the report, is that you need 10,000X more effective compute to train AGI.
A lot of recent AI progress has come from increasing the fraction of computer chips used to train AI. But the question is what kind of chaos do we have to live through before we come up with a real answer? For example, its dash cameras spot distracted driving and provide driver alerts and coaching tips in real time. This would probably mean that humans remain a “bottleneck” on AI’s economic impact for some time. My best guess is that the time between these milestones is less than 1 year, the primary reason being the massive amounts of AI labour available to do AI R&D, once we have AGI. Like your location, age and other personal information, the search engine uses artificial intelligence to guess what you’re looking for. Generative AI tools have been adopted ravenously in recent months by a curious, astounded public, thanks to programs like ChatGPT, which responds coherently (but not always accurately) to virtually any query, and Dall-E, which allows you to conjure any image you dream up.
If this happens gradually over many decades, it might feel like “progress as normal” rather than “AI is on the cusp of having a transformative economic impact”. Its comparative advantages might allow it to automate 20% of tasks long before it can automate the full 100%. The bullets below expand on this basic point. By analogy, AI may automate 20% of 2020 cognitive tasks using methods that don’t get AI close to automating 100% of them. If you are using it in certain kinds of high-stakes situations, you can get misinformation easily. The Deloitte survey of AI executives found that more than half have concerns about ethics, liability, and the consequences of using personal data without consent. If you as a tech specialist or team can’t find any more biases in your AI product, can you relax? Most of the analysis is from the perspective of a compute-centric framework, inspired by that used in the Bio Anchors report, in which AI capabilities increase continuously with more training compute and work to develop better AI algorithms.
That’s a pretty big gap in AI capabilities! AI won’t fill a gap in your headcount or resourcing. I do this via the “effective FLOP gap:” the gap between AGI training requirements and training requirements for AI that can readily perform 20% of cognitive tasks (weighted by economic-value-in-2022). Progress will be bottlenecked by the tasks AI can still not perform. But users still need to optimize their prompts, understand the technology’s limitations, and know where and when they can acceptably integrate the application into their workflows. Even if we have compelling reasons to hand over all decisions to AI, I still expect there to be a lot of (perhaps unreasonable) caution – AIs making decisions will feel creepy and weird to many people. “All of that data is generated independently and lives in disconnected formats all over the place,” he says. Superintelligence shows what happens when a superior AI chooses an ordinary person for observation, with the end goal to take over the world. Less incentive to deploy AI before superintelligence.