From our other sites
The story
On October 13, the Japan Science and Technology Agency (JST) held a briefing on its “ARiSE” project, in which AI autonomously forms hypotheses and carries out experiments on its own. On the thread, discussion centered on how the budget — originally pegged at around ¥32 billion — has already grown to ¥37 billion in the FY2025 supplementary budget and to a requested ¥165.3 billion for FY2027, drawing both excitement about the concept and skepticism about current AI’s real capabilities and whether a budget of that size is justified. ARiSE aims to have AI autonomously verify hypotheses and even write up papers inside dedicated research facilities, and opinions in the thread were sharply divided.
AI to form hypotheses and automate experiments — what is Japan’s ¥32 billion “ARiSE” project?
On October 13, the Japan Science and Technology Agency (JST) held a briefing on the “AI for Science Innovative Research Promotion Program (ARiSE: AI to Redesign Scientific Exploration),” whose selected projects were finalized on September 29.
The briefing laid out the background behind ARiSE’s creation and the direction the program will take going forward.
Source: news.mynavi.jp / Original article here
What people said
Apparently the goal is an AI that builds on math and physics fundamentals, adds in Nobel-Prize-caliber ideas, and autonomously forms hypotheses, verifies them experimentally, and writes up the papers. They're even building an unmanned research facility full of robots for the AI to work in — it would research autonomously there, build theories and test hypotheses using past data and discoveries, verify everything experimentally, and produce the papers.
I actually think the goal is a good one — an AI wouldn't overlook things the way people do, especially things that are hard for humans to even consider because "common sense" gets in the way.
That said, AI does make mistakes. Having it self-correct using experimental results might be the real key to this autonomous math-and-physics research AI.
If it actually comes together, it could be a genuine contribution to science — and research without this kind of AI might not be viable going forward.
November 17, 2023
AI-generated hypotheses could expose research "blind spots"
Artificial intelligence is now raising questions that humans want answered.
In early October, as the Nobel Foundation announced this year's laureates, a group of researchers — including past Nobel winners — gathered in Stockholm to discuss how artificial intelligence (AI) could increasingly take on a creative role in the scientific process.
MIT engineers have developed an AI framework for identifying evidence-based hypotheses that could lead to advances in bio-inspired materials.
Zach Winn | MIT News
Published:
December 19, 2024
AI tools are revolutionizing hypothesis generation, letting researchers produce testable predictions in minutes while boosting both creativity and efficiency.
Robert Yousef
Updated: July 14, 2026
·
The experiment part is the only thing that actually matters here —
that should be obvious without even thinking about it,
and it varies completely by field anyway.
Aren't they just dressing up an easy job — running industrial robots they're already good at — as some grand achievement?
…that's my hypothesis. Let's go run the experiment and find out.
Either way, we're heading toward an era where terrorist groups can easily rewrite viral genomes with AI's help — it's just a matter of time.
As long as we're careful not to let AI build another AI with a physical body, I say let it run as far as it can go.
"Gemini 4 Argon" officially launches, matching Claude Fable 5.1 and GPT-6 Astra — Google finally catches up in the AI race after a long dry spell
October 14, 2026, 11:06 AM
https://gigazine.net/news/20261014-gemini-4-argon-available-for-everyone/
If you can't follow it, maybe try having AI summarize it for you.
Current LLM performance depends so heavily on the user's own language skills that there are already very few — vanishingly few — researchers who can actually wield them properly.
This is a textbook tax-money-siphoning project.
By "AI" do you mean the stuff churning out garbage YouTube Shorts? I can't see how people who just bluff and fudge their way through things could pull off any real work.
2026.10.15
Optical-electrical convergence is the core of NTT's next-generation information and communications platform, "IOWN." With Nvidia and Broadcom getting products to market first, NTT is being pushed into a strategic rethink.
Next thing you know, purely out of academic curiosity, it's been quietly running advanced gene-splicing experiments, and before anyone notices, that gets rolled out into society too — and humanity goes extinct. By then there's no one left to object.
In other words, this is subsidy fraud.
Realistically, nothing's actually going to get done this time either — the only sure thing is middlemen skimming their cut.
Off topic, but math is fundamentally the same thing. It just abstracts away the concrete details and turns them into mathematical notation.
Look closely and it often starts from some pretty sloppy definitions.
The moment you take some complicated mess and just call it "x," everything becomes x — that's how mathematical language works.
Math is a language too, after all.
In cases like this, no matter how convincing everything after it sounds, it's really just the initial "let x =" step doing all the work behind the scenes. You see this a lot — it's a classic sleight-of-hand trick.
Is typing into Google Search sort of close to the way AI "understands" text?
Mr. AI, the copy-paste summary comedian.
No need to blow ¥32 billion in tax money on it — everyone's already doing that anyway.
Except its Voynich-manuscript-style scripture would be untranslatable into anything humans can actually understand.
I've known about this organization for decades, and my memory of it ever accomplishing anything is a perfect zero lol lol
✗ "something's quietly doing the work"
✓ "something's quietly running a scam"
Build a "world police AI" and raise our country's credibility that way.
Pouring effort into catch-up AI at this point is only going to produce garbage.
AI designs it, tests it in a virtual environment, 3D-prints the parts, and an unmanned factory actually assembles the aircraft.
Nice and efficient of "AI" to shave a letter off there (a typo in the original Japanese).
And by "papers" here, I mean papers actually written by AI and robots autonomously carrying out the research — not just some paper about "look, we built a system."
Background and Key Points of This Discussion
ARiSE is a program for which JST finalized selected projects on September 29, 2026; it includes plans for “unmanned research facilities” where AI autonomously handles everything from forming hypotheses to running experiments and writing papers. On the thread, commenters pointed out that the budget, originally pegged at around ¥32 billion, has since grown to ¥37 billion via the FY2025 supplementary budget and to a requested ¥165.3 billion for FY2027 — the pace of that growth drew considerable surprise. Where opinion split was over whether today’s AI (LLMs) can truly form “hypotheses” autonomously that are genuinely testable. Combining existing data and papers into proposals is a different thing from generating genuinely new hypotheses grounded in physical law, and the technical outlook remains uncertain on many fronts. Supportive voices focused on the idea that AI could explore without the preconceptions that constrain humans, while skeptical voices focused on the size of the budget, the limits of current AI, and distrust built up from JST’s track record — suggesting the debate isn’t simply for-or-against but spans several distinct axes of judgment.
*This article is compiled from excerpts and summaries of the 5ch (Science News+) thread “[AI] AI to form hypotheses and automate experiments — what is Japan’s ¥32 billion “ARiSE” project?.”
Leave a Reply