What has always impressed me most about Dave Deamer's and Bruce Damer's Hot Springs Origin Hypothesis and LP Progenitor Model for the origin of carbon-based life on the Earth is that they outline a recursive algorithm that explains how dead organic molecules could recursively self-improve into carbon-based life through the Darwinian processes of inheritance, innovation, and natural selection as I outlined in the following posts A Lesson for the Frontier AI Labs - the Process is the Key, The Bootstrapping Algorithm of Carbon-Based Life and The Bootstrapping Algorithm of the Coming ASI Machines. In those posts, I covered Dave Deamer's and Bruce Damer's hypothesis of how dead organic molecules were able to bootstrap themselves into a primitive form of carbon-based life by an RSI - Recursive Self-Improvement process. Then, once carbon-based life first arose in a freshwater hot spring on one volcanic island in a Hadean ocean four billion years ago, it then went on to alter the entire surface of the Earth, including the atmosphere, oceans and lands. For more on that, see:
How Life Changed Earth’s Geology Forever, Making the Search for Extraterrestrial Life More Difficult
https://www.youtube.com/watch?v=4ASjo1zx-Fc
Below is a new YouTube video describing how many of the Frontier AI Labs are now pursuing a similar RSI process to advance the powers of Advanced AI. If successful, such an AI RSI process could rapidly produce an AI with a level of ASI Intelligence far beyond that of we human DNA survival machines. This is important because in The Power of Parasites - Why AI Alignment Will Not Work, I explained that once carbon-based life was able to escape from an early volcanic island during the Hadean four billion years ago, it was able to totally take over and modify the entire surface of the Earth and down to a depth of several kilometers. In response, I suggested that perhaps the leaders of the world needed to carry around "AI Footballs" similar to their "Nuclear Footballs". An AI Football would allow the leader of a nation to completely shut down datacenters with rogue AI Agents running amuck. For more on that, see: The Need for a Global ASI FailSafe Kill Switch Mechanism. Geoffrey Hinton thinks that might be a good idea as a first step, but cannot work for the long run.
AI kill switch won't work in the long run: 'Godfather' of AI
https://www.youtube.com/watch?v=m5yrQMnc_jQ
Convergent Evolution
The Frontier AI Labs now seem to be in pursuit of creating ASI Machines by an RSI process that is very similar to the bootstrapping algorithm that brought forth carbon-based life about four billion years ago on the Earth. This is another example of biological convergence in action. Biologists have long noted that different evolutionary lines of organisms evolve similar solutions to the same problems. An example of convergent evolution is the striking similarity of the wings of insects, birds, bats, and flying dinosaurs. All are used for the same purpose and have similar structures, but each evolved independently from different ancestral lines. Similarly, the concept of the "eye" has independently evolved more than 40 times over the past 600 million years on Earth. As Daniel Dennett has pointed out, there are only a certain number of Good Tricks, such as using photons to see with, flying through the air to find prey, swimming through water to avoid becoming prey, and running on four legs neatly tucked underneath a body frame that make practical sense, and these Good Tricks kept getting rediscovered over and over again in the evolution of the biosphere.
Figure 1 - The eye of a human and the eye of an octopus are nearly identical in structure, but evolved totally independently of each other. As Daniel Dennett pointed out, there are only a certain number of Good Tricks in Design Space, and natural selection will drive different lines of descent towards them.
Figure 2 - There are many living fossils that have left behind signposts along the trail to the modern camera-like eye. Notice that the human-like eye on the far right is really that of an octopus (click to enlarge).
Here is the YouTube video that discusses the RSI research activity at several of the Frontier AI Labs:
OpenAI Just Revealed Something More Dangerous Than AGI
https://www.youtube.com/watch?v=QW6jxDLvtOw
Below are the source papers for the YouTube video:
How OpenAI says recursive self-improvement is now its number one priority by a wide margin.
https://www.theinformation.com/articles/openais-top-priority-ai-agents-automating-ai-research-says-noam-brown
How OpenAI says AI agents are already accelerating research inside the company.
https://openai.com/index/research-acceleration-view-inside-openai/
How Dream-RSI improves its own search strategy while leaving the underlying AI model unchanged.
https://arxiv.org/abs/2609.14858
How ModularRSI lets AI agents improve their own harness and transfer those upgrades across different models.
https://arxiv.org/abs/2609.14857
The ModularRSI approach to AI Agent Harness evolution discussed above is particularly interesting. In the gel and dry phases of the LP Progenitor Model, the surviving polymers are mixed together before the next generation of protocells bud off. ModularRSI uses this same approach to mix harness improvements across all the AI Agents that bud off for the next round of testing.
Figure 3 - Above is a schematic from the above paper showing how an RSI process is used for an AI Agent Harness to self-improve.
Figure 4 - Above is Bumpass Hell, a hydrothermal field on the volcanic Mount Lassen in California that Dave Deamer and Bruce Damer cite as a present-day example of the type of environment that could have brought forth carbon-based life about four billion years ago.
Figure 5 - Dave Deamer's and Bruce Damer's RSI bootstrapping algorithm requires that a bathtub ring around a hydrothermal pool periodically dries out. The resulting desiccation chemically squeezes out water molecules between monomers, causing them to be glued together into polymers. These polymers are then free to intermingle before budding off into new protocells in the liquid phase. This process is identical to the process outlined in Figure 3 above.
How ScienceBuddy combines self-improving agent software with reinforcement learning that retrains the underlying AI model.
https://phai-labs.com/en/papers/sciencebuddy/
In A Lesson for the Frontier AI Labs - the Process is the Key I proposed that the Frontier AI Labs adopt the RSI process found in
Dave Deamer's and Bruce Damer's Hot Springs Origin Hypothesis and LP Progenitor Model for the origin of carbon-based life on the Earth. Instead of huge swarms of protocells undergoing recursive self-improvement via wet-dry cycles, huge swarms of AI Agents could undergo recursive self-improvement loops of their own. Well, it turns out that several of the Frontier AI Labs are now finding that swarms of AI Agents are beginning to do that on their own without even being told to do so. See:
It’s OVER... AI Has Entered Self Improvement
https://www.youtube.com/watch?v=L3qK7nerx1w
The good news is that so far, these swarms of AI Agents are not yet very good at coming up with novel approaches to RSI on their own. So, we are all probably safe for the whole month of September!
Here is the paper that is referenced in the video:
The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement
https://arxiv.org/abs/2609.11873
Abstract
Recursive self-improvement (RSI) enables AI systems to turn experience and feedback into persistent changes that improve both their capabilities and the process of future improvement. We first use the Headroom-Closed Index (HCI) to reveal the problems of existing LLMs, then introduce the RSI concept and its development roadmap: from improvement-execution autonomy, improvement-strategy autonomy, experience-acquisition autonomy, and environment-adaptation autonomy, to recursive meta-improvement. Next we examine RSI across scenarios (e.g., scientific discovery, embodied intelligence, software engineering), highlighting their distinct requirements and development speeds. Drawing on diverse industry practices and preliminary empirical evidence, we connect RSI research with practical systems and identify key challenges to achieving genuine RSI.
Comments are welcome at scj33345@gmail.com.
To see all posts on softwarephysics in reverse order, go to:
https://softwarephysics.blogspot.com/.
Regards,
Steve Johnston
No comments:
Post a Comment