Rick Rosner On, 42: Trump, AI Breakthroughs, Fusion Energy, Mathematics, and Artificial Consciousness
Author(s): Scott Douglas Jacobsen
Publication (Outlet/Website): Personal SubStack
Publication Date (yyyy/mm/dd): 2026/09/23
Rick Rosner and Scott Douglas Jacobsen discuss U.S. political messaging, Greenland security arrangements, petroleum reserves, and election strategy before examining advances in artificial intelligence. They explore mathematical reasoning, fusion plasma control, weather forecasting, machine metacognition, artificial consciousness, ethical obligations toward synthetic minds, and humanity’s uncertain relationship with increasingly capable AI.
Rick Rosner: Early voting has already started in some states, with more starting in the next two weeks. The leadership in Washington is still throwing bullshit at us. Trump announced this new deal with Denmark over Greenland.
People quickly countered that U.S. security arrangements with Greenland go back about 85 years. In 1941, after the Nazis occupied Denmark in 1940, the U.S. signed an agreement with Denmark’s envoy to help defend Greenland.
Seventy-five years ago, Harry Truman signed a new defence agreement. During the Cold War, the U.S. maintained numerous installations and thousands of military personnel in Greenland. The idea that Trump newly secured permission for us to do security stuff in Greenland is mostly bullshit. Those rights already existed, although the new agreement reportedly extends the arrangements.
That’s what we’re getting from fucking Trump. We’re getting threats of tearing down the Kennedy Center. He’s not doing anything actually to help people.
We’ve got this Strategic Petroleum Reserve, which has an authorized capacity of 714 million barrels of oil. It was drawn down under Biden and now under Trump, and it’s down to about 285 million barrels.
The U.S. uses roughly 20 million barrels of petroleum a day. Trump has sharply reduced withdrawals from the reserve. About 130 million barrels have been released since the current release program began in March, ostensibly to help reduce fuel prices. Recent weekly withdrawals have fallen substantially. About 40 million barrels remained uncommitted under the announced release program. That isn’t the same as saying only 40 million barrels can physically be withdrawn.
That makes me wonder whether operational limits constrain further withdrawals. I was thinking about the caverns and whether the remaining oil might be hard to extract, but I don’t know enough to say.
The other possibility is that, in an October surprise, Trump might try to release another few tens of millions of barrels to bring down gas prices. Maybe he’s hoping for something like a 20% drop, though that’s speculation.
Then he can say, “Look, it’s already going down.” As I said, he’s got 43 days to go. Trump has a lot of bullshit to go, and most of it will be nonsense. But there’s a latency period between Trump saying something and it being proven nonsense.
He’s a king of that fucking latency. We’ll see how much bullshit comes out. It’s going to be a lot as we get closer to the election, as he persuades the easily persuadable that he’s actually doing shit.
That’s it. Next topic.
Jacobsen: Some areas of advancement for AI in recent years. This one called AlphaProof generates mathematical proofs that are formally checked in Lean. The research was published in Nature in 2025 and demonstrates strong performance on competition-level mathematical problems.
There’s also been AI control of fusion plasma in a physical tokamak, published in Nature in 2022. That provides experimental evidence for AI-based plasma control.
Rosner: Let’s talk about that for a second. Let’s explain what that is. A tokamak uses magnetic fields to contain hot plasma in a fusion experiment. It’s hard to contain this shit. You get fusion when the temperature, density, and confinement are sufficient for nuclei to fuse. Deuterium and tritium, both forms of hydrogen, can fuse to produce helium and a neutron, releasing a huge amount of energy.
The plasma is difficult to keep stable and contained. We’ve been containing it for decades, but maintaining the conditions needed for practical fusion power remains a challenge. You’re saying AI can adjust the magnetic fields fast enough to help control the plasma and keep it contained.
That could be a big deal. If you can, fusion has major advantages. Fission leaves you with radioactive byproducts that you have to do something with, and certain failures can cause meltdowns.
With fusion, losing confinement stops the reaction. You don’t get a fission-style meltdown, although there are still hazards, including radioactive tritium, activated reactor materials, and potential damage from plasma disruptions. Fusion would be great if you can pull it off.
Next thing you were talking about?
Jacobsen: AI weather forecasting, also published in Nature. They ran extensive quantitative comparisons against an established forecasting system under specified evaluation conditions. This program is called GenCast. It outperformed the European Centre for Medium-Range Weather Forecasts’ ensemble system on 97.2% of the targets evaluated in the study.
Gemini Deep Think has reached Olympiad-level mathematical reasoning, with IMO graders confirming a gold-medal-standard score for an advanced version in 2025. AlphaEvolve also has new algorithms and mathematical constructions, using automated evaluators to check proposed outputs.
Rosner: Interesting.
Jacobsen: There’s research on belief-like representations in AI and self-monitoring. A 2026 preprint offers evidence interpreted as internal representations of “beliefs,” or even metacognition, but those interpretations require further scrutiny.
Those are some major developments. This is aside from work involving things like the Navier–Stokes equations.
Rosner: This has prompted some people to say this is the end of human mathematicians, that AI will consistently kick our asses at shit like this. That’s not entirely true.
Some of this belief-like shit means we have to consider what thought is and what the consequences are seriously. If AI consciousness existed at all, it wouldn’t necessarily resemble our consciousness.
AI can already be multimodal, and some systems receive sensory input and control robots moving through the world. But that doesn’t establish human-like experience or evolved biological drives.
We have to examine what it does have and what that means in terms of its… What do you call it? It’s not the sensorium. It’s the mental arena.
You can make philosophical arguments. If you slowed down the world tenfold, you’d still be conscious, presumably. If you presented the world to yourself in slow motion and somehow slowed down your brain so you thought only 10% as fast, you’d still be conscious, just working more slowly. A hundredfold, a thousandfold, you might still be conscious, right? That’s the thought experiment.
AI processes information fast when it’s asked to. When it’s not being asked, you can imagine its processing being paused.
It’s only occasionally processing input from the world. If you’ve got an AI engine and it’s idling, it may not be doing anything. Its parameters are stored on servers until you ask it a question. Then it runs the question through its fucking neural network and presents an answer in conversational form.
You can ask yourself: this stop-start thing, where you can imagine it’s inactive most of the time and only starts processing when it’s asked a question, could that be conscious?
You could argue that interruptions wouldn’t necessarily rule out consciousness. Our conscious experience changes or is interrupted during sleep, although the brain doesn’t simply turn off. Interrupted consciousness could still be consciousness. That doesn’t establish that AI has it.
A lot of people are going to be thinking about shit like this. We’re better equipped to think about it than any civilization in the past, but we’re still not great at it. We need to get better as we confront a technology that could fuck us up. Cars were the devil wagons of 1905.
Jacobsen: Yes, that’s a good point. It gives us a moment to drop the supernatural nonsense and think more clearly about what ethics actually is. We’ve been asking the wrong questions, period.
Rosner: There’s a movie coming out called Klara and the Sun, which deals with that standard situation in science fiction: once you’re able to create artificial consciousness, how are we going to treat the beings that have it?
Jacobsen: Probably not well at first.
Rosner: No shit. Every science fiction story… My wife, who doesn’t know science fiction at all, wrote her version of this story. She spontaneously came up with it.
When does Klara and the Sun come out? Next month. It’s about an artificial companion and her experiences, directed by Taika Waititi, who did Jojo Rabbit. He’s a funny guy. I like his stuff. Jenna Ortega plays Klara. She’ll be living with a family.
You can see any number of shows that examine this shit. There’s one called Robots, or no, Humans, with Gemma Chan. It started in 2015, so about 11 years ago now.
Hold on. It’s about what happens when you bring a super-hot robot into your house, and everything you think is going to happen. Yes, here it is. Humans.
It ran for three seasons, and someone does have sex with the robot in season one.
We’re not going to be great to artificial consciousness, especially if you can make it for five bucks because computing keeps getting cheaper. Moore’s law describes transistor counts roughly doubling every two years with minimal cost increases. It doesn’t mean the cost of the same processing power drops a billionfold in 10 or 20 years, but computing has become dramatically cheaper over time.
If, at some point in the future, we can create human-level consciousness for five bucks and you can buy something at the conscious dollar store, that’s frightening. You’ll have people throwing away human-level consciousness. It ends up in a landfill. That’s a standard theme of science fiction: how we’re going to be assholes.
Jacobsen: The end?
Rosner: The end. To wrap it up, AI is a threat to us. If AI becomes conscious, we’re a threat to it.
Jacobsen: The end.
Last updated May 3, 2025. These terms govern all In-Sight Publishing content—past, present, and future—and supersede any prior notices. In-Sight Publishing by Scott Douglas Jacobsen is licensed under a Creative Commons BY‑NC‑ND 4.0; © In-Sight Publishing by Scott Douglas Jacobsen 2012–Present. All trademarks, performances, databases & branding are owned by their rights holders; no use without permission. Unauthorized copying, modification, framing or public communication is prohibited. External links are not endorsed. Cookies & tracking require consent, and data processing complies with PIPEDA & GDPR; no data from children < 13 (COPPA). Content meets WCAG 2.1 AA under the Accessible Canada Act & is preserved in open archival formats with backups. Excerpts & links require full credit & hyperlink; limited quoting under fair-dealing & fair-use. All content is informational; no liability for errors or omissions: Feedback welcome, and verified errors corrected promptly. For permissions or DMCA notices, email: scott.jacobsen2025@gmail.com. Site use is governed by BC laws; content is “as‑is,” liability limited, users indemnify us; moral, performers’ & database sui generis rights reserved.
