{"slug": "ai-seems-to-do-everything-impossible-so-where-are-we-in-time-travel", "title": "AI Seems to do Everything Impossible. So Where Are We in Time Travel?", "summary": "AI may become extremely useful in the search for new physics involving time, gravity and causality, but it is nowhere close to engineering movie-style time travel, according to an analysis of 2025 and 2026 physics research. The analysis cites a 2025 Physical Review D paper by Achintya Sajeendran and Timothy Ralph constructing a controlled warp-drive spacetime in which a freely falling particle could follow a closed timelike curve, and a second 2025 Physical Review D study building a nonsingular wormhole spacetime containing CTCs through topological surgery that violated standard energy conditions. The central obstacle is that physicists do not yet know whether nature permits the required physical conditions at all, with Stephen Hawking's chronology protection conjecture still unsettled and 2026 work examining whether instabilities around chronology horizons could provide such a mechanism.", "body_md": "Every few days, we hear about an AI system discovering a new material, solving a mathematical problem, or finding a scientific pattern humans had missed. At the same time, physics has never completely killed the idea of time travel. Einstein’s equations contain strange solutions involving wormholes and closed timelike curves, while quantum mechanics keeps challenging our everyday intuition about cause and effect. So it is tempting to combine the two trends and ask a very 2026 question:\n\n**Could AI help us build a time machine?**\n\nThe short answer is that AI may become extremely useful in the search for new physics involving time, gravity and causality. But we are nowhere close to engineering the kind of time travel shown in movies. The biggest obstacle is not simply that the equations are too difficult for humans. We do not yet know whether nature permits the required physical conditions at all.\n\nStill, the state of the art is much more interesting than a simple “no.”\n\nBefore discussing wormholes, there is one misconception worth clearing up. Traveling into the future faster than somebody else is already allowed by established physics.\n\nSpecial relativity tells us that a clock moving very quickly relative to another clock accumulates less elapsed time. General relativity adds another effect: clocks in stronger gravitational fields run more slowly. (Who hasn’t seen interstellar movie, right?)\n\nThese are measurable engineering effects, not philosophical ideas. Modern atomic clocks have measured gravitational time dilation across millimeter-scale height differences, and GPS requires relativistic corrections to remain accurate. If I could travel sufficiently close to the speed of light and later return to Earth, less time would have passed for me than for the people who stayed behind. In that limited sense, I would have traveled into their future.\n\nThe problem is scale. Accelerating a human-scale spacecraft to relativistic speed requires extraordinary energy. AI might optimize propulsion, trajectories and materials, but it cannot negotiate with the speed of light.\n\nTravel into the past is a completely different problem. General relativity describes gravity as the geometry of spacetime. Surprisingly, some mathematically valid spacetime geometries contain what physicists call **closed timelike curves**, or CTCs.\n\nRoughly speaking, a CTC is a path that always moves locally forward through time but eventually returns to an earlier event. These are not just science-fiction inventions. CTCs arise in serious solutions of Einstein’s equations, including Gödel-type universes, idealized rotating spacetimes, wormhole constructions and some warp-drive geometries.\n\nAnd research on them is still active.\n\nIn 2025, Achintya Sajeendran and Timothy Ralph published a *Physical Review D* paper constructing a controlled warp-drive spacetime in which a freely falling particle could follow a closed timelike curve and return to flat spacetime. Importantly, the authors explicitly noted the questionable physical realizability of the metric. In other words, the mathematics provides a time-machine-like geometry. It does not tell us how to build one.\n\nAnother 2025 *Physical Review D* study showed how a nonsingular wormhole spacetime containing CTCs could be constructed through a form of topological surgery. But the resulting spacetime violated the standard energy conditions.\n\nThat somewhat innocent-sounding statement hides one of the biggest problems in the entire subject. Many wormhole or time-machine proposals depend on exotic distributions of energy that we do not know how to create or sustain on useful scales.\n\nThere is also Stephen Hawking’s famous **chronology protection conjecture**: perhaps quantum effects prevent time machines from forming in the first place. The question remains unsettled. Work published in 2026 is still examining whether instabilities around chronology horizons could provide such a protection mechanism.\n\nThen came a particularly interesting result in June 2026. Kaiyuan Ji, Seth Lloyd and Mark Wilde published a *Physical Review Letters* paper studying how much information could theoretically be sent backward through a noisy, postselected closed-timelike-curve channel. They derived information-theoretic limits for these hypothetical retrocausal channels.\n\nThat sounds spectacular, but the distinction is crucial. They did **not** send information into the past.\n\nThey studied what communication theory would look like *if such a channel existed*. That gap between **“mathematically describable”** and **“physically constructible”** is basically the entire time-travel problem.\n\nThis is where the question becomes genuinely interesting to me as someone coming from computer science and AI. AI does not need to magically invent a time machine to change this field. Its immediate value is as a **search and modeling engine**.\n\nEinstein’s field equations are nonlinear tensor differential equations, and interesting spacetime geometries occupy an enormous mathematical search space. Traditionally, physicists make simplifying assumptions, choose a form for the spacetime metric, solve the equations and then test whether the solution has sensible physical properties. AI gives us another possibility: searching and representing much larger spaces computationally.\n\nWe are already seeing early versions of this.\n\nA 2025 project called **Einstein Fields** used neural representations to encode four-dimensional numerical-relativity simulations as implicit neural networks representing the spacetime metric. The researchers reported up to roughly a 4,000-fold reduction in storage on their test cases while preserving high numerical accuracy.\n\nAnother 2025 project, appropriately named **AInstein**, used machine learning to numerically solve Euclidean vacuum Einstein equations with a cosmological constant without making the usual symmetry assumptions.\n\nNeither project is trying to construct a time machine. That is not really the important part. What matters is the direction of progress: neural methods are beginning to represent and solve complicated geometries that are expensive to explore using conventional numerical techniques.\n\nAI is also getting better at discovering equations themselves.\n\nSymbolic-regression systems search huge spaces of mathematical expressions and try to recover compact physical laws from data. A recent *Nature Computational Science* study on parallel symbolic enumeration demonstrated substantial improvements in equation-recovery accuracy and computational efficiency across more than 200 synthetic and experimental problem sets. Now imagine extending this idea to spacetime.\n\nInstead of manually proposing one strange metric at a time, researchers could give an AI system a collection of constraints: Find a geometry that satisfies Einstein’s equations, avoids singularities, minimizes violations of energy conditions, remains stable under perturbations and exhibits a desired causal structure.\n\nThe AI could search candidate geometries, eliminate inconsistent solutions, run numerical simulations and perhaps even derive symbolic descriptions of promising candidates.\n\nThat would not prove that time travel is possible. But it could search the theoretical design space far faster than humans can today.\n\nThis is one place where headlines have gotten somewhat ahead of the physics. In 2022, researchers used Google’s Sycamore quantum processor to study a quantum system whose behavior has a dual description resembling a traversable wormhole. The result, published in *Nature*, was important because it provided an experimental way to study ideas connecting quantum information and gravity.\n\nBut nobody created a physical tunnel through spacetime. Caltech explicitly clarified that the experiment did not produce a rupture in physical spacetime. Instead, the processor implemented a quantum system whose dynamics could be interpreted through the mathematics of a theoretical wormhole.\n\nThat distinction is important when discussing AI too.\n\n*Simulating a wormhole, finding an equation describing one and physically constructing one are three very different achievements.*\n\nThe experiment is nevertheless fascinating because the deepest questions surrounding wormholes and time may require something we still do not possess: a complete, experimentally verified theory combining quantum mechanics with gravity. That is another area where AI might matter enormously.\n\nPossibly — **if there is a trick to discover.**\n\nI think discussions about AI and time travel often frame it as an engineering problem. Give a sufficiently intelligent model enough scientific literature, compute and laboratory access, and eventually it will design the machine.\n\nBut there are two very different possibilities.\n\nIn one universe, backward time travel is physically possible but hidden behind mathematics or engineering far beyond our current abilities.\n\nIn that case, advanced AI could be transformative. It might discover completely new spacetime geometries, quantum states, materials or control strategies that humans would never think to test.\n\nBut consider the other possibility.\n\nPerhaps backward time travel is prohibited by a deeper law of nature. Maybe quantum gravity enforces chronology protection. Maybe physically realizable matter can never generate the required geometry. Perhaps causality itself emerges as an unavoidable consistency condition of the universe.\n\nIf that is true, no amount of intelligence changes the answer. An AI can find a solution that we overlooked, it cannot make an inconsistent universe consistent.\n\nThis reminds me somewhat of computation itself. Better algorithms can turn previously impractical problems into tractable ones, sometimes spectacularly so. But intelligence does not automatically erase fundamental limits. It helps us explore the space permitted by the rules.\n\nSo, are we close?\n\nAs of September 2026, **no — not in an engineering sense**.\n\nWe can measure relativistic time dilation with extraordinary precision.\n\nWe can write mathematically consistent models containing closed timelike curves.\n\nWe can simulate quantum systems related to wormhole physics.\n\nWe can calculate how information would behave inside hypothetical retrocausal channels.\n\nAnd AI is rapidly becoming better at representing complicated physical systems, solving equations and discovering new mathematical relationships. What we cannot do is send matter, information or a person into their own past.\n\nStill, I would not dismiss the question. Black holes were once strange mathematical consequences of general relativity. Today, we observe their environments and measure gravitational waves produced when black holes collide. That history should make us careful about declaring every exotic mathematical solution physically irrelevant.\n\nBut there is another lesson hidden in that history. Black holes did not become real because our computers became smarter. *They became real because nature contained them.* That is probably the most important distinction when thinking about AI and time travel.\n\nAI may eventually become the most powerful scientific search engine we have ever built. It could explore solutions to Einstein’s equations that humans would never think to try, uncover unexpected connections between quantum information and spacetime, and perhaps contribute to the theory that finally joins gravity with quantum mechanics.\n\nIf a route to time travel exists somewhere inside the laws of physics, AI may dramatically improve our chances of finding it.\n\nBut first, the universe has to have left the door open.\n\n[AI Seems to do Everything Impossible. So Where Are We in Time Travel?](https://pub.towardsai.net/ai-seems-to-do-everything-impossible-so-where-are-we-in-time-travel-fa740bdb8349) was originally published in [Towards AI](https://pub.towardsai.net) on Medium, where people are continuing the conversation by highlighting and responding to this story.", "url": "https://wpnews.pro/news/ai-seems-to-do-everything-impossible-so-where-are-we-in-time-travel", "canonical_source": "https://pub.towardsai.net/ai-seems-to-do-everything-impossible-so-where-are-we-in-time-travel-fa740bdb8349?source=rss----98111c9905da---4", "published_at": "2026-09-23 07:23:40+00:00", "updated_at": "2026-09-23 07:53:29.122050+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-research"], "entities": ["Achintya Sajeendran", "Timothy Ralph", "Physical Review D", "Stephen Hawking", "Einstein"], "alternates": {"html": "https://wpnews.pro/news/ai-seems-to-do-everything-impossible-so-where-are-we-in-time-travel", "markdown": "https://wpnews.pro/news/ai-seems-to-do-everything-impossible-so-where-are-we-in-time-travel.md", "text": "https://wpnews.pro/news/ai-seems-to-do-everything-impossible-so-where-are-we-in-time-travel.txt", "jsonld": "https://wpnews.pro/news/ai-seems-to-do-everything-impossible-so-where-are-we-in-time-travel.jsonld"}}