引力不会有终极理论There Can Be No Final Theory of Gravity
Brett：几乎在所有情况下，你手上都只有一种理论可供选择。
就拿引力来说，眼下可供选择的理论实际上只有一种，那就是广义相对论。过去我们确实有过两种理论：牛顿引力理论和广义相对论——但我们做了一个关键实验。
"关键实验"这一概念是科学的点睛之笔。你有两种相互竞争的理论；如果实验结果是某种方向，其中一种理论被排除，而另一种则得以保留。在这种情况下，只要不出现问题，你就一直保留那种理论。
这种对知识的看法，让我们得以在追求进步的道路上永无止境。
这与任何其他关于知识的观点都截然不同。绝大多数物理学家仍然是贝叶斯主义者。他们之所以还是贝叶斯主义者，是因为大学里通常教授的就是这套东西，而这套东西也被视为一种在智识上严谨的理解世界的方式。
但说到底，它不过是我所说的科学主义的一种。因为他们背后有一道公式——贝叶斯定理——那是一道完全站得住脚的统计学公式。人们一直在以完全合理的方式使用它。问题在于，它并不是一种认识论。它无法保证、甚至无法让你确信你的理论确实为真。
我最喜欢的例子是：在 1919 年之前，每一个关于引力的实验都表明它与牛顿的引力理论相符。贝叶斯主义者在这种情况下会怎么说？他们会说，你对牛顿理论的信心越来越强了。
这怎么能说得通？在它被证明是错误的头一天，恰恰是你对它最有信心的那一天，这两者如何调和？
而波普尔主义者就没有这个问题。波普尔主义者会说："牛顿的理论从来就不是真正正确的。它包含了一些真理，但那些真理并不是我们能衡量的东西。"
我说它包含了一些真理，是因为比起某个普通人关于引力本质的胡乱猜测，它与现实有着更直接的联系。引力确实近似地遵循平方反比定律，但并非精确如此。我们需要广义相对论来纠正牛顿引力理论中的错误。
即便广义相对论目前是我们最好的猜想，它最终也不可能成为引力的终极理论。引力不可能有终极理论。我们所能拥有的，只是对现实越来越好的近似。
Naval：我觉得我们之所以如此轻易地陷入贝叶斯主义，很可能与为什么我们如此轻易地陷入悲观主义有关。我们在进化上天生就倾向于贝叶斯式思维。
地球上所有能形成良好解释的动物都是贝叶斯主义者。它们只是观察重复发生的事件，然后说："太阳昨天升起了，太阳明天还会升起。"或者："我摸到的那样东西是烫的，将来大概也是烫的。"我们的大多数生物系统和我们的大部分进化遗产，都是这样运作的。
只不过现在我们有了新皮层，能够用看不见的东西来解释看得见的东西。这赋予了我们更高层次的推理能力，但那种更高层次的推理对我们来说并非本能。它需要付出努力，需要深入思考。
我们默认采取贝叶斯主义，因为至少在纯粹的生物学层面，我们周围的大自然似乎就是这么运作的。
Brett: In almost all cases, you only ever have one theory on offer.
In the case of gravity, there literally is only one theory on offer at the moment. There’s general relativity. Previously we did have two theories. We had Newtonian gravity and we had general relativity—but we did a crucial experiment.
This idea of a crucial experiment is the cherry on top of science. You’ve got these two competing theories. If the experiment goes one way, one theory is ruled out but the other theory is not. In which case, you keep that theory for so long as no problems arise.
This vision of knowledge enables us to have an open-ended quest for progress.
This is completely unlike any other idea about knowledge. The overwhelming majority of physicists are still Bayesian. And the reason they’re still Bayesian is that this is typically what’s taught in universities and this is what passes for an intellectually rigorous way of understanding the world.
But all it is is what I would call species of scientism. It’s because they have a formula behind them, the Bayes’ theorem, which is a perfectly acceptable statistical formula. People use it all the time in perfectly legitimate ways. It’s just that it’s not an epistemology. It’s not a way of guaranteeing, or even being confident, that your theory is actually true.
My favorite example of this: Prior to 1919, every single experiment that was done on gravity showed that it was consistent with Newton’s theory of gravity. What do Bayesians say in that situation? They say you’re getting more and more confident in Newton’s theory.
How does that make sense? How do you square that the day before it was shown to be false was the day when you’re most confident in it?
Now, a Popperian doesn’t have this problem. A Popperian says, “At no point was Newton’s theory actually true. It contained some truth, but that truth isn’t a thing that we can measure.”
I say it contained some truth because it’s certainly got a more direct connection to reality than some other random person’s guess about the nature of gravity. Gravity does indeed approximately vary as the inverse square law, but not exactly. We needed general relativity to correct the errors in Newton’s theory of gravity.
And even though general relativity is our best guess right now, it can’t ultimately be the final theory of gravity. There can be no final theory of gravity. All we have is better and better approximations to reality.
Naval: I think the reason we fall into Bayesianism so easily is probably related to why we fall into pessimism so easily. We’re evolutionarily hardwired for Bayesianism.
Every other animal on the planet that can form good explanations is a Bayesian. They’re just looking at repeated events and saying, “The sun rose yesterday. The sun will rise tomorrow.” Or, “That thing I touched is hot. It’s probably going to be hot in the future.” That is how most of our biological systems and most of our evolutionary heritage worked.
It’s just now we have this neocortex that can form good explanations about the seen in terms of the unseen. That gives us a higher level of reasoning, but that higher level of reasoning is not instinctual to us. It requires effort. It requires deep thinking.
We default to Bayesianism because that is how a lot of the natural world around us seems to work at least at the purely biological level.