一部关于一位科学家的纪录片,他认为时间循环是真实存在的。
当前推动人工智能数据中心扩展的共识基于两个假设:
人脑是一种不受量子效应影响的生物计算机。
人脑并不解决计算上“困难”的问题(NP困难/ PSPACE);它仅仅运行有效的启发式近似。
在这些假设下,扩展经典计算架构最终将匹配生物认知能力。
考虑到这一点,我制作了一部纪录片。去年夏天,我跟随了丹麦生物医学工程的教学教授亨里克·凯尔德森,他认为上述假设是错误的。他相信大脑正在解决困难的问题,并且通过利用量子效应来实现这一点。
亨里克在领先的基础物理和哲学会议上展示了他的观点,我记录下了这一切。他与知名物理学家互动,试图弄清他的观点是否有价值,还是一种妄想。视频的早期评论者表示,这让他们想起了现实世界中的《引导者》。
亨里克论点的核心在于翻转关于两篇著名计算机科学论文的共识:
亚伦森/沃特罗斯反转
在2009年,亚伦森和沃特罗斯展示了一个连接到闭合时曲线(CTC)的经典计算机可以实现P-SPACE计算能力。这被用作反对CTC的不可行性证明,因为它们会违反扩展的丘奇-图灵论题(ECTT)。亨里克得出了相反的结论:如果时间循环赋予经典机器比我们认为当前量子计算机形式能实现的更多的能力,那么时间循环不仅真实,而且是基本的。这意味着量子力学不过是伪装的时间循环。
https://arxiv.org/abs/0808.2669
漏洞
托德·布伦在2003年的一篇论文中展示了经典计算机如何通过算法与CTC交互以解决困难问题,但它留下了一个不物理的违反(将信息发送到过去)。亨里克发现,如果你将布伦伪代码中“FINAL”目标标签向下移动一行,你可以在理论上测试时间循环而不产生悖论。(我们采访的物理学家在数学上验证了这一修改,但在实践中否定了它,因为他们认为CTC并不真实)。
https://arxiv.org/abs/gr-qc/0209061
生物硬件
亨里克获得了一个声望卓著的博士学位,他在其中使用声学全息术对大脑进行了建模,反对标准的准静态近似。他展示了在庞廷矢量中,可以将辐射能量与共振能量分开,并且在慢波δ睡眠(当突触重标定发生时)期间,虚值会增加。他的解释是,量子隧穿在大脑中发生,作为时空中时间循环的表现,以解决经典后进先出模型无法处理的全球计算困难问题(修剪)。
https://theses.ncl.ac.uk/jspui/handle/10443/2454
含义
如果这是真的,这意味着我们正在建立注定无法交付的庞大数据中心。
完整的2小时纪录片在这里:https://youtu.be/yTQa3O7Qlm4。
我很乐意回答有关制作的问题。我没有资格判断亨里克所提的观点是妄想还是革命性。因此,我将亨里克和他的想法呈现给世界顶尖科学家,并开始录制。
查看原文
The current consensus driving the scale-up of AI data centers rests on two assumptions:<p>The human brain is a biological computer operating without quantum effects.<p>The human brain does not solve computationally "hard" problems (NP-hard / PSPACE); it merely runs efficient heuristic approximations.<p>Under these assumptions scaling classical compute architectures will eventually match biological cognitive capacity.<p>With this in mind I made a documentary. Last summer I followed Henrik Kjeldsen, a Teaching Professor of Biomedical Engineering in Denmark, who argues the assumptions above are backwards. He believes the brain is solving hard problems, and it does so by utilizing quantum effects.<p>Henrik presented his ideas at leading fundamental physics and philosophy conferences, and I recorded it all. He interacted with well known physicists as they attempted to grapple with determining if his ideas have merit or if they are a delusion. It's not a hand holder. Early commenters on the video say it reminds them of a real-world Primer.
The core of Henrik's argument rests on flipping the consensus around on two famous computer science papers:<p>The Aaronson/Watrous Reversal<p>In 2009, Aaronson and Watrous showed that a classical computer connected to a Closed Timelike Curve (CTC) could achieve P-SPACE computing power. This is used as an impossibility proof against CTCs, since they would violate the Extended Church-Turing Thesis (ECTT). Henrik came to the opposite conclusion: If a Time Loop gives a classical machine more power than what we think a current form of a Quantum Computer can achieve, then Time Loops are not only real, they are fundamental. Meaning Quantum Mechanics is just Time Loops in disguise.<p>https://arxiv.org/abs/0808.2669<p>The Loophole<p>A 2003 paper by Todd Brun showed how a classical computer can algorithmically interact with a CTC to solve hard problems, but it leaves an unphysical violation (sending information to the past). Henrik found that if you move the target label of 'FINAL' one line down in Brun's pseudo-code, you can theoretically test for Time Loops without paradox. (The physicists we interviewed validated this modification mathematically, but dismiss it practically because they believe CTCs aren't real).<p>https://arxiv.org/abs/gr-qc/0209061<p>The Biological Hardware<p>Henrik earned a prestigious PhD, in it he modeled the brain using acoustical holography, arguing against the standard quasi-static approximation. He showed that within the Poynting vector, you can separate radiated energy from resonant energy, and that the imaginary value rises during slow-wave delta sleep (when synaptic rescaling occurs). His interpretation is that quantum tunneling is happening in the brain, acting as an expression of Time Loops in spacetime to solve a global, computationally hard problem (pruning) that classical LIFO models fail at.<p>https://theses.ncl.ac.uk/jspui/handle/10443/2454<p>The Implication<p>If true, it means we are building massive data centers that are destined to not deliver.
The full 2-hour doc is here: https://youtu.be/yTQa3O7Qlm4.<p>Happy to answer any questions about the production. I am not qualified to know if what Henrik is proposing is delusional or revolutionary. So I put Henrik and his idea in front of the world's top scientists, and hit record.