{"id":30927,"date":"2026-05-07T12:52:08","date_gmt":"2026-05-07T12:52:08","guid":{"rendered":"https:\/\/www.europesays.com\/ai\/30927\/"},"modified":"2026-05-07T12:52:08","modified_gmt":"2026-05-07T12:52:08","slug":"recursive-self-improvement-edges-closer-in-ai-labs","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/ai\/30927\/","title":{"rendered":"Recursive Self-Improvement Edges Closer In AI Labs"},"content":{"rendered":"

The field of artificial intelligence<\/a> was built on the premise that machines might someday improve themselves. In 1966, the English mathematician I. J. Good wrote<\/a> that \u201can ultraintelligent machine could design even better machines; there would then unquestionably be an \u2018intelligence explosion,\u2019 and the intelligence of man would be left far behind.\u201d AI researchers have long seen recursive self-improvement, or RSI, as something to both desire and fear. Today, advances in AI are raising the question of whether parts of that process are already underway.<\/p>\n

RSI means many things to many people. Some use the idea as a bogeyman to scare up regulation, while others brandish it in marketing. For some, it means a fully autonomous loop, while for others it\u2019s nearly any use of tech to build tech. <\/p>\n

Safest to say it\u2019s a spectrum. At its strictest, researchers use the term to describe systems that can improve not just their outputs, but the process by which they improve\u2014generating ideas, evaluating results, and modifying their own methods with zero human direction. By that standard, many of today\u2019s systems fall short. They can help build better AI, but they still rely on humans to set goals, define success, and decide which changes to keep. The question is not whether self-improvement exists in some form today, but how much of the loop has actually been closed.<\/p>\n

Stepping Stones to Self-Improvement<\/p>\n

Researchers have spent decades putting in place the elements of RSI. Machine-learning (ML) algorithms automatically tune the parameters of programs that can play games or even create new programs. ML methods called evolutionary algorithms diversify and iterate on design solutions, including other algorithms. Over the last decade, \u201cAutoML\u201d has automated aspects of the pipeline in which ML models such as neural networks<\/a> are structured, trained, and evaluated.<\/p>\n

Today, large language models<\/a> (LLMs) such as GPT, Gemini, Claude, and Grok extend this trend. One of their biggest use cases is to write code, including the code to produce future versions of themselves. In February, OpenAI reported<\/a> that GPT\u20115.3\u2011Codex was instrumental in creating itself, helping to debug training, manage deployment, and analyze evaluation results. Anthropic claims<\/a> that the majority of its code is now written by Claude Code. These systems still rely on humans to direct and verify the work.<\/p>\n

Last year, Google DeepMind announced a system called AlphaEvolve<\/a>, \u201ca coding agent for scientific and algorithmic discovery.\u201d It uses LLMs to guide the evolution of solutions, such as optimizing neural-network architectures, data-center scheduling, and chip design<\/a>. It\u2019s not a fully recursive loop, as people still need to decide what problems AlphaEvolve should solve and how to evaluate its performance. But each breakthrough enhances scientists\u2019 ability to make further AI breakthroughs. <\/p>\n

\u201cIt\u2019s also a very collaborative process\u201d between humans and machines, says Matej Balog<\/a>, a computer scientist at Google DeepMind who worked on AlphaEvolve. \u201cOften you look at what the system discovers, and you actually learn from that discovery.\u201d The system has already surprised the team. \u201cOur mission is to use AI to discover new algorithms that have evaded human intuition,\u201d Balog says, and \u201cI think we have the first demonstrations that this is not a wild dream.\u201d <\/p>\n

Meanwhile, the co-leads of Google DeepMind\u2019s earlier chip-design system, AlphaChip<\/a>, have launched a startup called Ricursive Intelligence<\/a> to use AI to design AI chips<\/a>. \u201cWe expect that we can dramatically reduce the design cycle from one or two years to days,\u201d says cofounder Azalia Mirhoseini<\/a>. Phase 1 is to help human designers. Phase 2 is to automate the process for companies without in-house designers. In Phase 3, the company will recursively use AI to design better chips to train better AI\u2014though still under human supervision, says cofounder Anna Goldie<\/a>. <\/p>\n

Other projects focus on AI agents<\/a> modifying their own behavior. Last year, scientists at the University of British Columbia and Sakana AI announced Darwin G\u00f6del Machines<\/a> (DGMs), which use evolutionary algorithms to improve LLM-based coding agents. Critically, agents can alter their own code (though not the underlying LLM), and get better at doing so. A newer version<\/a> can even alter its meta-mechanisms for improving itself.<\/p>\n

Members of the team also developed the AI Scientist<\/a>, reported in Nature in March, which aims to automate the broader research loop. It can generate research ideas, run experiments in software, write up the results in papers, and then review those papers. This project hints at how more of the AI development process\u2014not just coding, but experimentation and evaluation\u2014could be folded into an automated loop.<\/p>\n

Jeff Clune<\/a>, a computer scientist at the University of British Columbia<\/a> who worked on both DGMs and the AI Scientist, says that improving AI with AI is \u201cone of hottest topics in Silicon Valley.\u201d He believes that \u201cwe are right around the corner from recursively self-improving systems,\u201d and argues that RSI will rapidly \u201ctransform science and technology and all aspects of society and culture.\u201d<\/p>\n

Why AI Self-Improvement Still Has Limits<\/p>\n

Many barriers remain. Clune says that AI is merely decent at generating, implementing, and judging ideas. \u201cAll of the key pieces work OK but not great,\u201d he says. Dean Ball<\/a>, a senior fellow at the Foundation for American Innovation<\/a>, says that AI scientists still don\u2019t match the best human scientists. \u201cMaybe eventually they\u2019re going to automate the genius,\u201d he says, \u201cbut not next year. Next year they\u2019re automating the grunt who grinds through the algorithmic efficiency games.\u201d <\/p>\n

Even if those capabilities improve, the process may not compound cleanly. Nathan Lambert, a computer scientist at the Allen Institute for AI, recently wrote an essay arguing that instead of recursive self-improvement, we should expect \u201clossy self-improvement (LSI)<\/a>,\u201d in which increasing friction slows the flywheel. That\u2019s in part because large AI systems are growing more complex, and the job of an AI researcher will be to manage that complexity rather than to refine parts of the system. Further, top systems cost billions of dollars to develop, and no one wants to set an AI loose with that kind of cash. <\/p>\n

There are also broader constraints. Ball has written about RSI and why he\u2019s not a \u201cdoomer\u201d<\/a>\u2014someone who believes the phenomenon will take off and destroy civilization. Taking over the world, he argues, requires many practical steps, from running lab experiments to navigating politics. Further, knowledge is distributed and often tacit, so can\u2019t easily be bundled into one AI mind. For example, the capabilities of the chip-manufacturer TSMC<\/a> emerge from the collective intelligence of its 90,000 interacting employees. <\/p>\n

Full-on RSI might require not just designing software and chips but building data centers, running power plants, and mining metals, all using self-reproducing robots. For these and other reasons, some researchers argue that humans will remain central to the process. Meta<\/a> researchers Jason Weston and Jakob Foerster recently wrote that instead of self-improvement, \u201ca more achievable and better goal for humanity is to maximize co-improvement<\/a>.\u201d Keeping humans in the loop will lead to both faster and safer progress, they write, as people lend their insights and also steer AI toward solutions that benefit humanity.<\/p>\n

Could RSI End the World?<\/p>\n

Still, many scientists haven\u2019t ruled out runaway RSI<\/a>, sometimes called the singularity<\/a>. Last year, researchers interviewed<\/a> 25 AI experts about automating AI R&D. All but two entertained the notion that it could lead to an intelligence explosion. Participants were also more likely to think that AI companies would keep their self-improving models internal rather than deploy them publicly. \u201cIt\u2019s a pretty alarming combination, right?\u201d says David Scott Krueger<\/a>, a computer scientist at the University of Montreal<\/a> who co-authored the paper. He worries about research so risky happening \u201coutside the public eye.\u201d<\/p>\n

Krueger, who founded an AI-safety nonprofit called Evitable<\/a>, advocates for globally pausing AI development. \u201cIt\u2019s gambling with everyone\u2019s lives,\u201d he says. One red line he has suggested for triggering the pause is when 99 percent of code is written by AI. \u201cThat\u2019s one that I think we\u2019re maybe crossing about now.\u201d <\/p>\n

Even though Ball calls the singularity<\/a> \u201ctotally childish sci-fi bullshit,\u201d he believes frontier AI labs conducting RSI research should be closely monitored so that their models don\u2019t fall into the wrong hands, such as bad actors who could use them to accelerate the development of cyberattacks or biological weapons<\/a>. RSI has risks, he says, but they can be managed. <\/p>\n

Society of Artificial Minds<\/p>\n

When people picture RSI, they might envision one big-brained AI growing bigger-brained. But it might look more like evolution, where many diverse agents emerge and act together. Krueger says there could be \u201csomething like a Cambrian explosion of artificial life forms.\u201d They\u2019d have ecosystems, cultures, and economies. <\/p>\n

Clune believes evolutionary algorithms and open-ended processes<\/a>, which explore without a strong objective, will be key to RSI. Collaboration between agents will also help. Systems like the AI Scientist, which packages its findings into formal papers, offer one way for agents to share results and build on each other\u2019s work. \u201cIt\u2019s a pretty good way for the system to communicate with other agents,\u201d Clune says. <\/p>\n

Human scientists might get edged out of AI research, but slowly. First, Clune says, they\u2019ll spend less time on lower-level tasks and become more like professors or team leads, who pick research directions. Then people will be more like program officers or CEOs, who set broader research agendas. Finally, they\u2019ll conduct oversight, a role he hopes humans never forfeit. Clune says he might be sad if a machine replaces him as an AI scientist, a role he finds \u201cexhilarating.\u201d But the payoff could be worth it. \u201cI\u2019ll give up my hobby to cure cancer.\u201d <\/p>\n

From Your Site Articles<\/p>\n

Related Articles Around the Web<\/p>\n","protected":false},"excerpt":{"rendered":"The field of artificial intelligence was built on the premise that machines might someday improve themselves. In 1966,…\n","protected":false},"author":2,"featured_media":30928,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[24,4989,25,20007,2225,3736],"class_list":["post-30927","post","type-post","status-publish","format-standard","has-post-thumbnail","category-ai","tag-ai","tag-ai-safety","tag-artificial-intelligence","tag-evolutionary-algorithm","tag-llms","tag-singularity"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/30927","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=30927"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/30927\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/30928"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=30927"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=30927"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=30927"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}