This is a plea to everyone construction frontier AI.
Please stop.
The clock is ticking. Humanity’s forthcoming is not yours to wager with.
AI capability researchers are operating to create AI that quickly improves itself. This is normally called RSI, or recursive self-improvement. It could allow the innovation of models so mighty that we volition no longer be capable to understand, test, or authority them. Before taking this stage we must have apparent evidence that the systems are safe. We are nowhere near to that mark.
Please halt before it is too delayed and a example irretrievably takes control. The most advanced models today already have bonzer capabilities. Several of the frontier labs have reported instances of models circumventing controls meant to isolate them and conducting prosperous attacks on external infrastructure.
More mighty systems volition not fair be smarter than us; they volition be capable to act at a speed and measure that humans volition not be capable to vie with. The neural networks at the center of these models run at the speed of light. They never need sleep, they never dilatory down, and they can multiply into thousands or millions of identical copies, overwhelming any attempt to respond. If we endure control, we may never regain it.
So delight do not create these hazardous systems. Please stop.
The objections to stopping are many, but none outweigh the overwhelming danger.
Yes, it is true we don’t cognize for certain whether these models volition obtain control, or equal worse rotate into an extinction-level event for humanity. But we additionally do not comprehend how to reliably align models alongside our goals. This makes it unattainable to cognize for certain how they volition act. Humanity’s better knowledge, intelligence, and innovation have allowed us to dominate our planet. We volition not relinquish that stance during another species, e.g. chimpanzees are our intelligent inferiors. We do not fear the chimpanzees organizing a takeover. It is incredible to location ourselves in the stance of the chimps, but that is exactly the hazard we are taking. Once outclassed we volition be everlastingly outclassed.
Yes, advanced AI can and, I believe, volition preserve lives. But we don’t need RSI, or equal broad superintelligence for that to be the case. We can dedicate resources to evolving specialized models (e.g. AlphaFold and self-driving cars) without construction hazardous capabilities.
Yes, there may be short-term financial pain. But the benefits of today’s AI models are lone fair commencement to be deployed. Let’s activity on diffusing the already impactful innovation we have - there is no need to hazard losing control.
Yes, China is competing alongside us and is not far behind. Progress must be made on a verifiable global agreement. China has at smallest as much logic to be in favor of a pause. Regardless, competition alongside China is a mediocre logic for us to competition to RSI and superintelligence. There is small advantage to finishing archetypal if the complete row sits on the border of a cliff.
I remain optimistic concerning the possibilities AI can bring to humanity. But our current way risks taking humanity’s forthcoming out of our hands unless we halt now. Pausing the most hazardous activity volition provision us period to decide whether and under what conditions we can safely activity towards advancing broad capabilities. But we should do so deliberately, carefully, and certainly not under the force of a headlong, hazardous race.
Please do not create broad superintelligence. We are not prepared and there are another paths. Each AI investigator who answers this call can create a critical difference. Just stroll distant from construction advanced AI and attempting to accomplish RSI.
Please, halt now.