Sleep and Impulse Control
It’s 7 pm. A freshly baked sugar cookie stares back at me. The doughy middle revealing itself in seductive allure. The sugar formed in what looks like a smile, taunting me. All day I have prepared myself for this moment by promising that I will not cave to the temptation. But my track record is poor. Likelihood of failure: >95%.
Q: what aspect of my sleep habits most determines whether I will have sufficient willpower to resist?
A: the amount of sleep, in particular Deep sleep, that I got the previous night.
This is the most significant lesson I drew from a Kernel pilot study [1] which sought insights into how sleep affects aspects of cognition. Over the course of six weeks, eleven of my co-workers and I tracked our natural sleep with a wrist-worn Whoop device. In parallel, while wearing a Kernel brain interface, we completed up to eighteen, sixty minute sessions measuring our reaction times, short-term memory and impulse control.
When I was well rested, with a lot of deep sleep compared to my baseline, Kernel scientists found that I performed better [2] on an impulse control cognitive task based on a classic psychophysics study design.
When they looked at my brain’s activation to that task across days, some areas of my brain reflected the amount of sleep that I had gotten the previous night [3]. Importantly, the correlation between brain activity and sleep was not driven by my performance on the task. My brain activation was correlated to the previous night’s sleep independently; thus my brain yielded information about my ability to withhold keypresses (or, in the grander scheme of things, resist the cookie) above and beyond what my behavior could reveal.
Said differently, when I was well rested, my brain was more engaged in willpower control. What’s especially interesting to me is that Kernel brain data revealed a relationship of cognitive functioning with my Whoop data that behavior alone could not.
A similar connection between sleep and inhibitory task performance has been shown in the literature (Demos 2016) but to the best of our knowledge this is the first time this effect has been shown to have a neural origin on a single-participant level [4].
Oops, I Did It Again
These results made me contemplative. My mind traveled back to years ago, when the combination of severe sleep deprivation from three young children, the demands of building a startup, and an existential crisis of faith dropped me into a deep depressive state. These years were marked by struggles with my own behaviors and impulse control. The results of this study invited me to imagine a future where quantification of cognition and sleep becomes the norm, where these kinds of contemplations can empower all of us.
Everything that plagued me during those darker years seemed to culminate in a single moment every day: overeating at night, right before a late bedtime. Soon, almost certainly because of this, there were 50lbs more of me in the world. No matter what I tried or how hard, I was powerless to stop myself in this vulnerable day-end moment of exhaustion and stress. Misery and shame dominated my mind, identity and emotional states. Surely all of the poor decisions were ravaging inside my body, too?
Hundreds of millions of Americans face similar struggles daily, with sleep deprivation and overeating temptations driving their health into a worsening spiral. Food companies making it oh-so-easy to splurge and get addicted to the worst kinds of food.
My first instinct — most people’s first instinct — is that overeating has to do with diet. Makes sense, right? Who could have guessed it was problems of 1) deciding when to eat and 2) getting consistent high quality sleep. In other words, what if my dietary problems were, in fact, cognitive problems?
I’d never thought of sleep as something that could be engineered. The schools I attended didn’t offer sleep classes. No one pulled me aside and gave me tips. So, like most, I just crossed my fingers each day and hoped for the best until I read Matthew Walker’s book Why We Sleep, which introduced me to the science of sleep. My favorite passage:
“The best bridge between despair and hope is a good night’s sleep.”
Sleep is the human reset button.
Through trial (lots of trial) and error (so much error), here is what I’ve learned about myself, my habits, my willpower, and the relationship between my diet and sleep:
- Anecdotally, my cookie-snubbing track record went from a sad <5% chance success to a reliable >99% after I restructured my life to get predictable, high quality sleep.
- Being able to quantify my sleep with a sleep tracker, versus rely upon my subjective self assessments, was the critical enabler.
- Fasting 10 hours before sleeping produced strong sleep enhancement (dinner at 10 am).
- Being in bed at exactly 8 pm, versus somewhere between 9pm and midnight, enabled consistency of sleep outcome.
- This new lifestyle was cemented when I began building my Autonomous Self (explained in a forthcoming article).
Years into my efforts, today I weigh 155 lbs; 51 lbs less than before. My body fat is 10.1%. Emotionally, my personal and professional relationships are positive, constructive and without drama. My relationship with myself is the same.
The Future of Our Cognition
Which brings me back to the Kernel study, quantifying the subjective and impulse control.
Even though this study was primarily focused on building the infrastructure and tools for future customers —and less about generating a peer-reviewed level of scientific insight — it has me unable to stop thinking about the future of my mind.
A few years ago, I playfully made an attempt to become “Cognitively Perfect”, trying to rid my cognition of the 188 chronicled human mental and decision making biases. For example, with an opportunity to form an opinion on a given topic, I practiced resisting my default proclivities to preferring information that confirms what I already believe i.e. confirmation bias. And then there were the 187 other biases to consider.
My effort was modeled after Benjamin Franklin’s 1726 endeavor to become “morally perfect”. Like Franklin, mine failed but it made clear that absent technology to quantify my subjective self awareness and autonomous systems to scaffold the gains, lessening the amount of bias error in my thinking is perhaps more than I can reasonably aspire to at this moment. If I was never able to track my deep sleep, I would never be able to find, let alone even study, the insights linking it to impulsivity.
What if I could begin doing for my mental well-being and performance what I did with my sleep? That is, acquire data, hypothesize, test, analyze and repeat. We’ve never been able to consistently collect any data about my brain, until now. What if what was once felt can now be measured?
Might I discover the:
- conditions under which I optimally learn?
- circumstances to reach new creative abilities?
- mindfulness practice that is most suitable for emotion regulation?
- communication methods to connect more deeply with my partner, children and friends?
- dozens of reasons it’s a good idea to quit social media?
Access to this data will re-scaffold our decision making and, eventually, all of society. We already build aspects of the world around the few cognitive features we can approximate in the brain today, albeit crudely.
We all understand the unit of measurement called a calorie. Calories in, calories out. Sleep has its numbers, too: REM, Deep, HRV, etc. But when it comes to the function of our brains, why don’t we have any units of measure that allows us to converse about it objectively?
With non-invasive brain interfaces ubiquitous, maybe we’ll use something like attebytes, a concept I wrote about a while ago. As in: “I read Peyton’s social media post right before bed last night and got so upset that I burned 117 attebytes and got 72% less deep sleep than normal. Feeling awful this morning, I ate two donuts and had two coffees for breakfast.”
Whatever the units of measure are, one thing is for certain. The Neuro-Quantified Era (NQE) is here.
This will enable me and you to begin running daily experiments to build intuitions and systems around our individual and shared mental well-being and performance. Being able to collect the data which exists — but is outside my conscious or sensory awareness — was the primary enabler to kick start the cycle of systematic hypothesis generation and improvements in sleep. It was a treasure trove of insight about nearly everything in my life! I just got my first taste of it and I can’t stop thinking about the cookie-snubbing possibilities.
Measuring What’s Hidden
Figuring out my “sleep algorithm”, as though I was building a startup of Bryan Johnson Sleep, required an enormous amount of time, attention and dedication to build — but an investment unquestionably worth the effort. Where did this all lead? For me, to profound and everyday changes in my life.
To give you an idea of how extensive my experimentation was, last year The Economist reported on the outlandish measures someone was taking to achieve high quality sleep. This person seems ridiculously extreme!
Oh wait, that’s me.
First, close the blackout blinds in your bedroom. Eat dinner at 4pm, and do not eat or drink anything after 6pm. Put on your blue-light blocking glasses at 8pm. Set your bedroom temperature to 67ºF (19.4ºC) and your electric blanket to 69.8ºF (21ºC). At 8.45pm, meditate for five to ten minutes. Switch on your deep-wave sound machine. Put on your sleep-tracking [device]. You are now, finally, ready for slumber. This may all sound a bit over the top. But this is the “sleep hygiene” routine described in a recent blog post by Bryan Johnson, who sold his previous company to eBay for $800m and is now chief executive of Kernel, a startup developing brain-computer interfaces. He admits that his sleep routine has “decimated my social life”, and that his partner sleeps in a different room, but says all this trouble is worth it, because it has boosted his level of “deep sleep” by as much as 157%. He has bought [wearable sleep trackers] for all his employees.
Such measures may seem extreme, but that’s what it takes to understand and break the connection between sleep and cookies. So what did I learn?
I can predict with high accuracy what my sleep metrics will be if I eat a 6pm dinner instead of sticking to my normal 5am-10am* window (*my last meal of the day comes even earlier than described in the Economist’s article, nowadays!). I know the different impacts of various foods. I know the likely outcomes if I experience negative emotional arousal before bed. I know with high accuracy that if I eat that 6pm dinner, my sleep will be severely compromised which will obliterate my impulse control the following day and put me at high risk of kickstarting a vicious cycle.
The key to the NeuroQuantified Era is simple: numbers. Science begins with counting. Engineering thrives on them. We need the numbers.
Not everything that can be counted counts. Not everything that counts can be counted — William Bruce Cameron
An interesting thing happens when we are given a technological tool to measure something that was previously hidden or not measured (i.e. followers/likes/Twitter ratios/etc): our attention turns toward it as though we’re driving by an accident or movement in our peripheral vision. It’s impossible to not look. Evolution designed us to pay attention to some things against our will.
Measuring under the hood is the only way to get past subjective hunches. Wearable sleep tracking devices can now quantify the granular details of our sleep performance: REM and Deep, respiratory rate, HR, HRV, sleep latency, and body temperature. Even the limited accuracy of today’s consumer devices allows for astounding peeks behind the veil, past the limits of unaided introspection. Human performance measurement today is a world of low-hanging fruit.
Today, consumer-grade devices like a Whoop, Garmin or Apple Watch allow users to get most of the benefits of high-cost, in-clinic sleep studies for a fraction of the cost, at home. In my case, this enabled me to conduct hundreds of experiments in a quest to figure out: what lifestyle produces a perfect night’s sleep for me? (One day, I want the same data-derived insights for my mind — i.e. What lifestyle produces the most original thinking?)
The first thing sleep data showed was that my evening eating just an hour or two before bed would decimate my ability to get deep sleep. Less than 20 minutes most of the time. No wonder I had near non-existent will-power when in a staring contest with the sugar cookie!!
And so I kept going. I commenced systematic experimentation to assess the optimal time to eat my last meal of the day. 10am produced the best results. At bedtime my resting heart rate will be around 47 bpm which is the strongest predictor of sleep quality. If it’s 65 bpm because I ate later for some reason or ate the wrong thing I know I’m in for a rough night and following day.
Knowing sleep is affected by many variables, I also performed rigorous trialing of caffeine, light exposure, supplements, blood glucose levels, bedtime routines (movie versus book), HRV training, meditation and more to determine that my optimal bedtime was exactly 8pm. Not a few minutes before or after. 8pm. On the dot. This created no shortage of awkward situations having to explain why I needed to leave a social event early, just when things were getting started. But it worked, enabling me to score a near consistent 100% on sleep performance as measured by Whoop.
This process of experimentation and precision of insight would have been impossible if my only measurement tool was my subjective self-awareness. Yes, we know that circadian rhythms are genetically encoded and enabled by a neural pacemaker. We all have hunches about whether we are morning or night people or whether we slept a bit too little or too much the night before but these are hunches. The mind effortlessly performs approximation but is not to be trusted with the details.
The most important thing I learned was that my ability to improve myself was tightly associated with the quality of measurements at my disposal. In addition to on-body sleep trackers, I explored my mental and overall health with a wider lens and will also measure levels of Cortisol (4x/day), DHEA-s, Serotonin, GABA, Dopamine, Norepinephrine, Epinephrine, Glutamate and PEA.
Doing all of this wasn’t cheap and not everyone can do it. Hopefully the value is my being a guinea pig for the future where such monitoring is ubiquitous. As more consumer-grade personal health technology makes its way to the market, we are headed for a world where the only thing that will limit your ability to investigate yourself will be your curiosity, commitment, and willingness to make lifestyle changes.
Hey Sleep, What About This?
Granting sleep independent authority meant that no matter where the data pointed, I’d head in that direction. This is how I have arrived at eating dinner at 10 am and going to bed at 8 pm, even though I experienced significant disruption in my social life. After extensive experimentation, this is the protocol that produces the highest quality sleep for me.
While some of my practices may have some universal applicability, you will need to discover what works for you through experimentation.
To do that, you need the tools.
And the numbers.
I can’t wait. Until then, I will keep sharing the lessons I learn, good and bad. Stay tuned and good luck on your cookie-snubbing.
About Kernel
Kernel is a L.A.-based neurotechnology startup that has created a new high-quality, consumer-friendly brain activity imaging and recording system. Kernel Flow, being released in Q1 2021, is a bicycle-helmet sized cap which uses near-infrared light to capture fMRI-like neural activity across the entire cortical surface. Similar in concept to the PPG technology behind Whoop, Flow non-invasively shines light through the skin and skull and measures its absorption and scattering as the light bounces back to the sensors, which are placed around the head. Flow’s underlying technology, known as functional near-infrared spectroscopy, or fNIRS, is clinically proven safe and is a crucial step toward a world of consumer, high-quality brain interfaces in every home.
[1] Study Design
The six week study included myself and eleven other Kernel team members. We each did three, sixty minute sessions per week performing four tasks:
- Resting State: We began each session by doing a resting state, opening and closing our eyes every sixty seconds. This is to test levels of drowsiness (Karolinska drowsiness test).
- Memory Task: We then did a memory task (N-Back), where a sequence of uniquely colored turtles were presented one by one, every two seconds. We were asked to answer if the turtle we saw n-times ago (i.e. 3) is the same or different. The game requires you to keep in short term memory the sequence of unique turtles and decide match or no-match.
3. Impulsiveness Measurement: Go/No-go was next and is great for measuring impulsiveness; the capacity to not respond (i.e. to binge or not to binge!). Every second either a green leaf or a red flower appeared on the screen. Press the spacebar in response to the green leaf; do not press the spacebar in response to a red flower.
4. Reaction Time Measurement: Finally, my reaction time was measured by how fast I could press the spacebar from when a plus sign changed into a timer at random intervals.
Sleep Data Collection
Each of us wore a Whoop device to quantify our sleep and answered daily surveys about perceived quality of sleep, caffeine intake, prescription medicines, and mood assessments (i.e. rating 1–10 levels of happiness, alertness, anxiousness, irritation, etc.).
Cognitive Performance Measurement using a Kernel Brain interface
We used Kernel’s benchtop, fiber coupled Beta TD-fNIRS system for the study (the version before Kernel Flow). It has four emitters, 48 optical detectors (positioned on the prefrontal cortex) and 4 EEG electrodes (down the centerline).
Kernel Flow as of 11/2020
[2] Total, deep and sleep latency were correlated with my impulse control
The performance on the Go/No-Go task (described above [1]) can be quantified with a metric from signal detection theory known as the sensitivity index, a.k.a d’ (“d-prime”). It is a metric that takes into account both commission (when I pressed the key on trials when I shouldn’t have) and omission (when I failed to press the key on trials when I should have) errors. Higher d’ means better performance. This task performance metric was correlated to sleep metrics from the previous night as reported by my Whoop band, including total sleep, deep sleep, and time to fall asleep (see figure below; p-values are not corrected for multiple comparisons). The more sleep (and deep sleep) I got the previous night, the better my performance. The longer it took me to fall asleep the previous night (it never takes me very long!), the better my performance.
[3] Total (and Deep) sleep duration were correlated with my neural activations during the impulse control task
Kernel scientists extracted my brain activations while I was performing the task — attempting to resist pressing the key when red targets appeared. They specifically looked at how the concentration of oxygenated hemoglobin changed when I was performing the task, with respect to periods when I was resting.
On average, they found that my brain showed a bilateral de-activation (blue) during the impulse control task, compared to rest (see figure below; arbitrary units).
Yet there was a lot of variability in the activation pattern across days — while on most days my bilateral prefrontal cortex was deactivated, it was activated with respect to baseline on other days! (see scatter plot below, where the brain activation is shown on the y-axis)
Interestingly, the variability in my brain activation pattern across days was highly correlated with how long I had slept (and how much deep sleep I had gotten) on the night preceding the experimental session (see figure above; thresholded at p<0.01 uncorrected). An example of the relationship between total sleep duration and brain activation is shown in the scatter plot below, picking a representative location in the large yellow cluster in the right hemisphere.
[4] Notes on results
Cautionary note: correlations with few samples (in my case, eleven) lead to inflated effect sizes and it is possible that the relationships discovered in these preliminary analyses would not hold if I kept collecting more data. However, this is a statistical issue that has been plaguing all of human neuroscience research. Collecting 10 sessions of data for a single subject on separate days is actually quite a feat for a standard neuroscience lab, in particular due to the constraints and costs of collecting high quality neural data. Kernel Flow will make such experimentation much easier. It will remove the data paucity hurdle that stands in the way of a better understanding of our brains and how they relate to our everyday functioning.
