HackerTube English?
学习信号与模型行为 →

实验测到的,真是日常建议里说的那件事吗?

从一个早晨散步的例子开始,试着给观察写上准确的动词:移动眼睛、报出字母、测出一种酶。动词一换,我们可能已经在回答另一个问题。

本文的问题 · 16
  1. 研究中的眼动,就是早晨那段散步吗?
  2. 看见第一个字母后漏掉第二个,量到的是眨眼吗?
  3. 同一组数据,能回答几个问题?
  4. 两杯咖啡和几只汉堡,分别在比什么?
  5. 声音变大与抑制减弱怎样分开记?
  6. 测一天的暴露,还是找到排放源?
  7. 都能悬起来,还需要观察哪件事?
  8. 两只时钟、两处停顿
  9. 把床清出来以后,怎样知道改的是哪件事?
  10. 打乱组别以后,哪一个问题还可以回答?
  11. 标签补回了一种成分,实验也测回了原来的食物吗?
  12. 两个检验读到相近的东西,算独立确认吗?
  13. 前后两次观察,是同一群个体吗?
  14. 看见蜂群来取食,下一项实验要数什么?
  15. 能做实验的动物,和要兑现的承诺
  16. 值得研究,是结果里的哪一栏?

研究中的眼动,就是早晨那段散步吗?

一次散步同时改变太多东西:身体在前进、眼前的景物在移动,还可能有室外光线。Huberman 指出,他引用的一项研究考察的是眼动,没直接测向前行走。他也在睡眠节目里区分了咖啡带来的清醒感和光照对身体时钟的影响。把研究搬到日常生活前,写下两栏:研究改变了哪一项,日常情境又一起改变了哪些项。这里不能仅凭一次散步后的感受,把作用分给其中一个因素。

原文与时间戳 · 4
Now, those are eye movements. They didn't specifically look at forward ambulation
Maximizing Productivity, Physical & Mental Health with Daily Tools · ▶ 15:05–15:11
the walk and the optic flow that I talked about before.
Maximizing Productivity, Physical & Mental Health with Daily Tools · ▶ 21:13–21:17
it binds to the adenosine receptor.
Master Your Sleep & Be More Alert When Awake · ▶ 8:16–8:20
it will phase delay, it will delay your clock.
Master Your Sleep & Be More Alert When Awake · ▶ 57:27–57:29

看见第一个字母后漏掉第二个,量到的是眨眼吗?

设想一串数字里闪过 R 和 Z。被试报告看见 R,却没有报出 Z;任务记录的是少报了一个目标。把 attentional blink 译成「注意瞬脱」,正是为了防止把这个结果听成眨眼次数。Huberman 在另一期谈物理眨眼时,又反对拿它判断一个人是否撒谎。两段合在一起给出一个核查动作:把结果写成可以直接观察的动词——「报告」「合眼」——再看解释里有没有悄悄换词。他在神经系统入门中使用的注意力聚光灯,是帮助理解选择性的比喻,并非实验仪器。

原文与时间戳 · 4
attentional blinks, not actual eye blinks.
ADHD & How Anyone Can Improve Their Focus · ▶ 72:24–72:28
be on the lookout for the letters R and Z
ADHD & How Anyone Can Improve Their Focus · ▶ 74:19–74:27
There is absolutely no evidence that blink frequency correlates with anything except alertness.
Optimizing Workspace for Productivity, Focus, & Creativity · ▶ 72:27–72:32
it's like a spotlight, except it's not one spotlight.
How Your Brain Works & Changes · ▶ 18:33–18:37

同一组数据,能回答几个问题?

在饮食实验的解读里,高纤维组的纤维消化酶增加,是一个具体结果。把它改写成「多样性也增加」就换了终点。Palmer 的访谈把界线往前移了一步:他说当时正有随机试验进行;这说明有人准备测量,却连结果的方向都还没有给出。读到「有研究支持」时,试着补完整句话:哪一组、与谁比、测了什么、结果是否已经公布?这两段谈的是怎样读研究,并不提供个人治疗建议。

原文与时间戳 · 4
the group assigned to the high fiber condition maintained high fiber intake for six weeks
How to Enhance Your Gut Microbiome for Brain & Overall Health · ▶ 85:29–85:34
there was an increase in these fiber digesting enzymes.
How to Enhance Your Gut Microbiome for Brain & Overall Health · ▶ 92:47–92:51
documenting efficacy in large numbers of patients with these disorders, we don't have them.
Diet & Nutrition for Mental Health | Dr. Chris Palmer · ▶ 58:14–58:21
we've got five randomized controlled trials underway
Diet & Nutrition for Mental Health | Dr. Chris Palmer · ▶ 60:49–60:57

两杯咖啡和几只汉堡,分别在比什么?

咖啡机的试饮努力固定用水、粉量、滤纸等条件。另一部视频把一个人的汉堡口感判断交给更多人尝,再转去讨论饲料投入。增加试吃人数可以改变口感判断的证据,没办法替代资源投入的测量。为下一次对比只写一个句尾:「我们希望据此判断的是……」。要填两个句尾,就准备两种观测。

原文与时间戳 · 3
same water, same coffee, same amount
Drip Coffee Makers — super simple, super cheap · ▶ 22:16–22:22
I wanted a larger sample size
Feeding Bill Gates a Fake Burger (to save the world) · ▶ 7:32–7:35
you're building bone and respiration for that cow
Feeding Bill Gates a Fake Burger (to save the world) · ▶ 10:32–10:37

声音变大与抑制减弱怎样分开记?

Huberman 用聚会上说话变响来讲饮酒后自我监控的变化。录音可以比较音量;若想判断是背景噪声、周围人的行为还是神经抑制变化导致,还得换一种观察设计。把「听见了什么」与「讲者怎样解释」分开记录,回看时才知道哪句话需要补证据。

原文与时间戳 · 2
people stop modulating their level of speech quite as much
What Alcohol Does to Your Body, Brain & Health · ▶ 22:50–22:55
the prefrontal cortex and top-down inhibition is diminished
What Alcohol Does to Your Body, Brain & Health · ▶ 24:26–24:31

测一天的暴露,还是找到排放源?

柴静展示同一张滤膜在采样前后的样子。这给出了某段时间、某条生活路径的证据。她随后谈到空气流动怎样影响污染物扩散。即使排放量相同,局部测得的浓度也可能变化。若要定位哪个源头改变了,把地点、风和采样时段加回调查记录。

原文与时间戳 · 2
After 24 hours it looked like this
Chai Jing's review: Under the Dome – Investigating China’s Smog 柴静雾霾调查:穹顶之下 (full translation) · 字幕译文 · ▶ 6:48–6:51
Air movement has decreased but the same quantity of pollutants remains
Chai Jing's review: Under the Dome – Investigating China’s Smog 柴静雾霾调查:穹顶之下 (full translation) · 字幕译文 · ▶ 24:40–24:42

都能悬起来,还需要观察哪件事?

在交变磁场中悬起的盘子还会发热。影片又展示了无需依靠感生涡流的悬浮。两次都拍一张离开支架的照片,看上去很像;把温度变化、材料与场的条件加进记录,才开始有能区分解释的观测。

原文与时间戳 · 2
the eddy currents in that plate generate so much heat
World's Strongest Magnet! · ▶ 10:06–10:11
nothing to do with induced eddy currents
World's Strongest Magnet! · ▶ 15:14–15:18

两只时钟、两处停顿

Sapolsky 给神经递质信号和人的症状各设一只时钟。如果一个小时就能量到前者,两周后才观察到后者,那么快时钟的读数还不足以解释这段等待。Walker 描述一串按键时换了另一种分辨率:整串的总时间会下降,真正变化的却可能只是两个键之间那次停顿。重新读一张「随时间改善」的图时,分别写下起点、终点,以及两者之间究竟测过哪些步骤。

原文与时间戳 · 4
norepinephrine signaling is changing within like an hour.
Stanford's Sapolsky On Depression in U.S. (Full Lecture) · ▶ 22:02–22:06
they don't get better for a couple of weeks.
Stanford's Sapolsky On Depression in U.S. (Full Lecture) · ▶ 22:08–22:11
people learn it, but they have these problem points
Joe Rogan Experience #1109 - Matthew Walker · ▶ 41:42–41:47
that friction point in your motor skill sort of deficit and it smooths it out.
Joe Rogan Experience #1109 - Matthew Walker · ▶ 42:37–42:42

把床清出来以后,怎样知道改的是哪件事?

Grey 建议给睡眠区一条可遵守的边界,比如把刷消息移到别处。要检验这种改变,你会记录「在床上刷了多久」,还是「多久睡着」?前者能检验有没有执行分区,后者才接近这条建议希望改善的结果。同一周还可能有作息等其他变化。把两栏分别记录,比只写一句「这周睡得好」更容易看见哪一步已有证据。

原文与时间戳 · 2
the sleep station boundaries as where you go to sleep only.
Your Brain Thinks Your Whole Home Is One Room · ▶ 4:19–4:23
Make them feel like truly different spaces.
Your Brain Thinks Your Whole Home Is One Room · ▶ 9:59–10:02

打乱组别以后,哪一个问题还可以回答?

VanderPlas 保留观测值、打乱分组,构造在标签可交换时组间差会有多大的参照分布。接着他有放回地从同一批观测抽样,问统计量本身会怎样波动。拿 Haidt 的网上道德问卷试想:反复重采样可以描述这些答题者的变动,却无法把没来答题的人加回来。Zimbardo 的演讲又提醒我们,谁参与设计情境、谁有权结束观察,本身也是研究设计的一部分。

原文与时间戳 · 6
We first -- we shuffle the labels.
Jake Vanderplas - Statistics for Hackers - PyCon 2016.mp4 · ▶ 13:26–13:28
with replacement to generate new samples.
Jake Vanderplas - Statistics for Hackers - PyCon 2016.mp4 · ▶ 18:02–18:05
made a questionnaire, which we put up on the web
Jonathan Haidt: The moral roots of liberals and conservatives · ▶ 8:20–8:24
the slope is steeper on in-group, authority, purity
Jonathan Haidt: The moral roots of liberals and conservatives · ▶ 9:26–9:29
What does the situation bring out of them?
The psychology of evil | Philip Zimbardo · ▶ 8:46–8:48
And I ended the study the next day.
The psychology of evil | Philip Zimbardo · ▶ 20:11–20:13

标签补回了一种成分,实验也测回了原来的食物吗?

Lustig 在早期演讲里谈到为了货架寿命而去除纤维;后来又用整粒小麦的比例比较食品标签,并区分补回可溶性纤维和原来的纤维组合。这两份录制来自同一讲者,能帮助读者辨认他论证中的条件。若要判断实际健康结果,仍须说出具体比较的食品、受试者和测量终点。

原文与时间戳 · 4
The removal of fiber for shelf-life and freezing.
Fat Chance: Fructose 2.0 · ▶ 42:22–42:25
Anybody know what evaporated cane juice is?
Fat Chance: Fructose 2.0 · ▶ 77:05–77:09
the carbohydrate to fiber ratio of a wheat berry is three to one.
How Sugar & Processed Foods Impact Your Health | Dr. Robert Lustig · ▶ 190:50–190:56
If you only add the soluble fiber back
How Sugar & Processed Foods Impact Your Health | Dr. Robert Lustig · ▶ 197:03–197:05

两个检验读到相近的东西,算独立确认吗?

病例叙述中,一个异常生化读数先引起怀疑,针对溴离子的专门检验才回答「究竟是哪种离子」。PCR 历史里的对照和重复也可作参照:重复同一种可能误认对象的测量,主要告诉我们读数是否稳定;换成能区分竞争解释的测量,才检验对象识别。这里讨论的是推理链,而不是根据病例替另一位读者作诊断。

原文与时间戳 · 4
The lab blood test does it.
A man asked AI for health advice and it cooked every brain cell · ▶ 7:23–7:26
a blood test specifically for bromide arrives back from the lab
A man asked AI for health advice and it cooked every brain cell · ▶ 11:28–11:33
He didn't have any controls or do repetitions of his experiments.
The Man Who Took LSD and Changed The World · ▶ 16:56–17:00
survive the high heat of the PCR cycle
The Man Who Took LSD and Changed The World · ▶ 22:32–22:35

前后两次观察,是同一群个体吗?

Sapolsky 回忆狒狒群行为变化时,同时提到群内成员经过了筛选。取第二次观察的平均数减去第一次,差值既可能来自个体改变,也可能来自被统计的个体改变。超长跑者关于疲劳和休息的叙述则给出一个人的时间线。将两种材料并读,能区分「群体构成改变」与「同一个人不同时间报告不同感受」,它们需要不同的对照。

原文与时间戳 · 4
your social rank and your personality and your patterns of social affiliation
Joe Rogan Experience #965 - Robert Sapolsky · ▶ 34:06–34:07
the baboons who are left are the nice guys.
Joe Rogan Experience #965 - Robert Sapolsky · ▶ 39:27–39:31
I did the 20 minute nap
Joe Rogan Experience #1027 - Courtney Dauwalter · ▶ 21:02–21:07
There's aid stations out there
Joe Rogan Experience #1027 - Courtney Dauwalter · ▶ 76:40–76:45

看见蜂群来取食,下一项实验要数什么?

Stamets 回忆自己在花园里看见蜜蜂拨开木屑,接近菌丝上的液滴。后来他追问真菌提取物对蜜蜂会有什么影响。先给两句话分别配一个可观察的结果:记录访问次数能检验它们是否来取食;要判断某种提取物是否影响健康,则需对照、规定剂量和事先指定的蜂群指标。这一步设计是读者可提出的问题,花园里的那一幕本身只支持第一种观察。

原文与时间戳 · 2
bees moved the wood chips aside
Joe Rogan Experience #1035 - Paul Stamets · ▶ 82:00–82:05
what would they do with bees?
Joe Rogan Experience #1035 - Paul Stamets · ▶ 87:41–87:43

能做实验的动物,和要兑现的承诺

研究者能在特制小鼠体内控制何时清除某些细胞,这使一种因果问题变得可检验。Kurzgesagt 自己把人类试验列为下一步。Watson 介绍电子墨水设备时也作了一个朴素的区分:用 HDMI 显示图像是已展示的功能,使用它会不会改善自己的睡眠则是想追踪的结果。两段的证据强度差异很大,却都能用同一张纸做检查:把实际干预对象、实际测量结果、最终希望帮助谁分别写出来。

原文与时间戳 · 4
scientists genetically engineered mice
How to Cure Aging – During Your Lifetime? · ▶ 1:52–1:57
we need human trials.
How to Cure Aging – During Your Lifetime? · ▶ 6:24–6:29
as an external monitor over HDMI.
E-Ink on Android, and other new things! · ▶ 1:17–1:19
see if it helps with my sleep.
E-Ink on Android, and other new things! · ▶ 2:01–2:05

值得研究,是结果里的哪一栏?

Grey 反问,若因为衰老普遍存在就赋予它价值,同样的推理是否也适用于视力不佳?这是关于研究目标的判断。Kurzgesagt 则描述能干预小鼠体内衰老细胞的实验。先问什么样的结果值得追求,再问特定干预在哪个物种上取得何种结果;把两个问题分栏,就可以支持有关研究,同时检验每一种疗法的证据。

原文与时间戳 · 3
Happiness is meaning itself.
Don't Be the Last Person to Die · ▶ 1:14–1:16
focusing our attention and shaping our tools
Don't Be the Last Person to Die · ▶ 3:46–3:53
scientists genetically engineered mice
How to Cure Aging – During Your Lifetime? · ▶ 1:52–1:57

读完再想一步

把一条「研究发现」写成四格:谁参与、和谁比较、测了什么、何时测。遮住结论再读一遍:只靠这四格,还能推出原来的日常建议吗?