Book Review: Creativity in Science

Book Review: Creativity in Science

This book proposes viewing scientific creativity as a combinatorial process. Each scientific domain contains a population of ideas. Finding new permutations of these to create original ideas is what we think of as creativity. The more permutations a scientist can create, the more likely they are to produce creative work.

This idea has some interesting implications. It might explain why scientific creativity may diminish over time. The more ideas in your head, the more permutations you can find. When you're young, you can devote your entire life to acquiring new knowledge and giving yourself more ideas to permutate. As you get older and have other obligations (family, etc) your rate of acquiring new knowledge goes down and so you have a reduced supply of new ideas to permutate. Also you tend to become more focused on acquiring new knowledge within a narrower domain (i.e. you start specializing) which again reduces your potential permutations. Finally you become less receptive to new ideas as you age (Planck's Principle). 

There were some interesting observations from studies about the process of scientific research itself. Most studies measure creative output as the number of highly citied publications. This lets us measure scientific output at higher resolution than just relying on works like Principa Mathematica. The data shows that the most impactful scientists are also the most prolific and it's a very skewed distribution. An elite group of scientists are responsible for producing both the highest quality and quantity of research. Interestingly, the more scientists there are within a particular domain — the more elite it becomes. The combinatorial view of creativity would explain this as the larger population of scientists can create a larger population of ideas and therefore a large set of possible permutations for all scientists to find. The most creative scientists can take better advantage of this expanded population of ideas to increase their creative output and grow their market share of all creative ideas. This view fits with my experiences of extremely smart people in the startup world. The gap between them and everyone else seems to only grow over time as they continue spending weekends and free time learning new things and starting new projects.

There also seems to be a strong correlation between the most creative scientists and the diversity of projects they're working on at one time. Darwin worked on multiple projects while developing the Origin of Species and when he felt stuck on one, he'd moved onto another. It seems the optimal balance is to have a set of concurrent projects with a core set of themes but with a diversity of difficulty, type and how much of your attention they currently command.

There is also a look at some of the psychological attributes of highly creative scientists. It seems that IQ is important but it's more of a minimum threshold than a differentiator amongst the top scientists. You need an IQ higher than 120 but after that, other factors determine your scientific creativity. 

One of these factors might be "associative richness" — the ability to connect seemingly unrelated ideas to form new, logically sound ideas. Ernest Mach, the physicist, described this as the distinction between a mechanical memory that's good for recalling facts (which is necessary to avoid mistakes and work on hard problems) and the ability to connect these facts in interesting ways to invent new things. This distinction fits my personal experiences. I've worked with smart people who I wouldn't describe as creative. They're suited to solving well defined problems that require algorithmic problem solving but you wouldn't expect them to propose new or interesting ideas. They'll usually argue that "creativity" is a meaningless term and can be reduced to rigorous, logical thought. The author opposes this view and note that even computer programs best mimicking human creativity, have to introduce some random element or mechanism.

You can fit the associative model of ideas into the combinatorial model of scientific creativity. If you think of all the ideas in your head and graph the number of associations between them against the strength of each association. A steep curve would indicate you had a small number of associations but very strong links between them. A flat curve would mean a large number of associations with weaker links between them. To be more creative you want the latter. The more associations you have between ideas, the more permutations you can find and the greater your odds of finding a revelatory new one.

This model also explains why the most creative people also have a lot of bad ideas. If the associations between your ideas are high in quantity but lower in average quality, you'll draw a lot of connections that lead to dead ends i.e. you'll link together ideas that just create new and bad ideas. Still, the alternative is to generate fewer, more accurate ideas but be less likely to generate truly creative ones.

Overall I thought this book gave me a more detailed way to express my existing views on creativity and science. I do think it overemphasizes the importance of chance relative to how special the individuals who product scientific breakthroughs are. Having the energy to be the most prolific scientist within a domain and maintain a diversity of projects and interests, requires a special type of person.

Conversation and Ideas

Conversations about ideas should help you understand what you think. They need more tension than a regular conversation but not so much that you're afraid to share nascent thoughts. The best conversations make your ideas better and feel magical. The worst conversations make all thoughts blurry and leave you pessimistic. 

The people you regularly have these conversations with will have a long lasting effect on you. You'll keep simulating conversations with them in your head as you mull over ideas alone. These simulated conversations then become part of your internal monologue and shape how you think. Making bad conversations your default could permanently wreck your ability to have good ideas. That makes it important to know who the right people to talk about ideas with are.

I think the way to do this is by looking past the ideas someone has and figuring out their motivation to talk about them. For a good conversation about ideas both sides should share a single motive — finding the truth. I'd like to believe this is the only motive anyone has when discussing ideas but it's not the case. I have been guilty myself of believing my motive was finding the truth when I was really pulling myself in another direction.  There are a few such motives I've repeatedly come across. 

The first is one we have all felt — protecting our relationship with people we care about. This makes it hard to have a real conversation about ideas with family or close friends. Disagreeing with them might create conflict which risks the relationship. Just thinking about the trade off is enough to stop you fully engaging with the ideas. 

Another motive looks similar but is different — making the other person feel good. Some people care more about the experience of having a conversation with them than its content. An earnest example is a meek person who agrees with everyone so they’ll be liked. A more sinister example would be a sycophant. Continue along this line and you end at a politician. A skilled politician knows they have to disagree with something you say. Their skill is waiting for the right moment to counter with a truism and using their charisma to make it sound profound.

The opposite experience of being charmed by a politician is fighting with the chronic debater. Their motive is winning the argument at all costs. The reason might be the same as the politician, they care what you think about them, but their identity is wrapped up in seeming smart rather than being liked. Mostly though I think it's because they get a buzz from debating and are hooked on it. They win either by always speaking first and fast to keep you on the defensive or responding to your ideas by knocking them down instead of building on them. 

The motive I fear most is someone checking which tribe I belong to. Some people care only about figuring out if you believe the same things they do. The earnest reason for this is comfort. It's easier to be around people who think like you do. The more insidious reason is punishing you for not being in the right tribe. This is where you encounter the nastiest personalities, like the agents of Mao tricking teachers into revealing their beliefs so they could be fed to students.

It is hard to suss out someone's motives for discussing ideas without talking with them for a while. I think the best you can do to speed up the process is start paying more attention to the types of conversations you're having. Recognize when you're having a conversation about ideas and reflect on the motives of who you're having them with. To improve the quality of your idea conversations be picky about who you discuss ideas with. If you only want to talk about ideas and follow this rule strictly, your life will start fitting the shape of a reclusive genius. Ramanujan pretty much only talked about mathematics and had two conversation partners — Hardy and God. If you prefer a more regular shape of life but still care about having good ideas, adjust accordingly.

I appreciate Tatiane Souza Taggar, Garry Tan, Paul Buchheit, Kulveer Taggar, Aaron Iba and Paul Graham for helping me better understand my thoughts on this.


Recently I’ve been getting great sleep (without any major changes in general lifestyle or stress levels). I made a bunch of changes at once. One, all or some combination of them are having an effect. Here’s the list:
  • Comfortable mattress: This is obvious. What’s not obvious is how much effort people have already put into testing every available mattress and writing up their findings in detail. It’s worth spending time on Sleep Like the Dead and Sleepjunkie before buying a mattress to find one that suits you. (It’s possible spending a lot of time researching a mattress makes it feel more comfortable through placebo effect, which is fine by me).
  • Eat dinner early: I’d often eat dinner within an hour of going to sleep. Now I’ve been eating at least 4.5 hours before going to bed (related to starting a Intermittent Fasting protocol).
  • Oil diffuser: I turn on an oil diffuser with lavender scents about 45 minutes before going to sleep.
  • Reading fiction: I read an hour of fiction. I typically don’t read much fiction and would use any spare reading time for non-fiction/learning. That makes my mind too active though and it’s also tough to focus on dense material after a long day.
  • 10 min meditation: After reading I do a 10 minute guided sleep meditation on Simple Habit or Headspace. I think taking long deep breaths is 90% of the benefit vs the actual meditation.
  • Sleep stories: Calm has a feature called Sleep Stories, which are exactly what they sound like — bedtime stories for adults. I was skeptical at first but I’ve found them amazing for keeping you getting wrapped up in your thoughts and staying awake. I’m usually asleep before the end of one.
  • No phone in bedroom: The only electronics allowed are my Kindle and an old iPhone with just the Spotify, Calm, Headspace and Simple Habit apps.

Two other things I’ve used for a while are:

  • Blackout curtains: Make your room as dark as possible (I don’t like using a sleep mask)
  • Sunlight alarm clock: Wakes you up gradually.

Finally, the best advice on getting good sleep I’ve heard was from Naval — “don’t run a company” :)

(Edit: As many of my friends have noted, all of these suggestions will be utterly useless if you have kids).

My Experiments to Avoid Heart Disease

I wrote previously how learning that I had a family history of early heart disease prompted me to take a more proactive approach to managing my health. After getting tested for a number of markers associated with heart disease, and scoring in the high range for some of them, I began forming an action plan.

Of course I knew the basic plan would be to "eat well and exercise", as per the advice most doctors would offer. But I wanted more specifics. What's *my* optimal diet for preventing heart disease? Which has the greater effect, diet or exercise? Is focusing on one alone sufficient? How much should I be exercising?


I began by focusing on just diet. Generally I eat pretty well, avoiding sugar and refined carbs. This helps keep my weight under control (I have an endomorphic body type) but I do eat a fair amount of meat (minimum twice a day) and especially red meat (minimum twice a week). There seemed to be much literature advocating a plant-based diet as a means of not only preventing heart disease but reversing it. The physician behind much of this work, Dr Dean Ornish, worked with Bill Cinton to help him recover from his quadruple bypass in 2004.

I decided to try a modified version of the strict plant-based diet. I cut out all meat, fish, dairy and restricted eggs to a maximum of twice a week.  I found this change much tougher than I'd expected, I felt gassy and low-energy after the second week.  After one month I made a modification and switched to occasionally having lean, white meat or fish for dinner but still no red meat at all. I found this more workable and would still often go days without any meat at all.

After two months I ran another set of blood tests to see if my heart disease markers had changed:

Marker Change Target
Total Cholesterol 206 mg/dL (+/- 0%) < 200 mg/dL
LDL (bad cholesterol) 127 -> 119 mg/dL (- 6%) < 100 mg/dL
HDL (good cholesterol) 70 -> 77 mg/dL (+ 10%) > 40 mg/dL

I was happy to see improvement in the ratio of my LDL and HDL.  Even though it'd only been two months,  I was somewhat surprised that my total cholesterol hadn't changed at all though.  This made me curious about the exact link between diet and cholesterol.  I began looking into the research connecting them. I discovered something that stunned me. For 70% of people, diet supposedly has minimal effect on their cholesterol levels (more detailed explanation below).

More concerning to me was that my dietary change had negligible effect on two of the more advanced heart disease markers, in fact they showed a tiny increase:

Marker Change Target
Apo(B) 98 -> 102 mg/dL (+ 4%) < 80 mg/dL
Lp(a) 148 -> 151 nmol/L (+ 2%) < 75 mg/dL

I found there was considerable evidence that Apo(B) may actually be a more important predictor of heart disease risk than LDL cholesterol. The Canadian Cardiovascular Society has included Apo(B) in its heart disease management guideline since 2009.  A quick explanation of the theory is that the number of cholesterol particles you have, is more important than your total amount of cholesterol.  Apo(B) count is a proxy for the number of cholesterol particles because each particle has exactly one Apo(B) molecule. 

There were some more notable changes in two of my other makers:

Marker Change Target
Triglycerides (fat in your blood) 80 -> 49 mg/dL (- 39%) < 150 mg/dL
hs-CRP (Inflammation) 3.1 -> 0.8 mg/L (- 77%) <1 mg/L

I was happy to get my triglycerides down but they were already in the healthy range, unlike my other markers. It was also hard to know what to make of the inflammation decrease, since it's a notoriously variable reading (intense exercise or a recovering from a cold could both raise it). Frankly, it was a surprise to me that I'd scored so highly in the first place and I suspect it could have just been a funky reading (human involvement in the testing process means there's an inevitable margin for error on these tests).


During these two months I'd not been exercising at all. Now I kept the same diet and signed up for a crossfit gym, working out there 3x a week. Two months later I ran the tests again. This time I found even fewer changes in my cholesterol levels:

Marker Change Target
Total Cholesterol 206 md/dL (+/- 0%) < 200 mg/dL
LDL (bad cholesterol) 119 mg/dL (+/- 0%) < 100 mg/dL
HDL (good cholesterol) 77 -> 73 md/dL (- 5%) > 40 mg/dL

However I saw some significant movement in my other markers:

Marker Change Target
Apo(B) 102 - 90 mg/dL (- 12%) < 80 mg/dL
Lp(a) 151 -> 88 nmol/L (- 42%) < 75 mg/dL

Given the importance of Apo(B) in particular, as I mentioned above, this was really encouraging. My remaining markers showed some mixed results:

Marker Change Target
Triglycerides 49 -> 61 mg/dL (+ 25%) < 150 mg/dL
hs-CRP (Inflammation) 0.8 -> 0.5 mg/L (- 37.5%) < 1 mg/L

I'm not concerned by the rise in my triglycerides though, as I'm still well within the healthy range. I plan to continue monitoring it and if it keeps continues moving upwards, I'll have plenty of time to figure out a strategy to course correct.  


Given the importance of Apo(B) as a predictor of heart disease and its non-response to my diet-only modification, regular exercise is clearly an essential component of an effective heart disease prevention plan for me.

Looking at my LDL trend, the lack of change in the past two months is interesting. I can think of two explanations; 1) my body is sensitive to dietary cholesterol and I'd have to adopt a strict vegetarian/vegan diet over a longer period of time to bring it down into what's considered the healthy range, 2) my exercise regime increased my body's production of LDL and cancelled out any decrease from my diet modification (there is some precedent that increased muscle mass elevates LDL levels, and since starting crossfit I've gained just over 3lbs of muscle mass).

I plan to continue with my regime of limiting meat intake (I actually quite enjoy it now as it gives me a reason to explore new places and foods for lunch) and regular exercising. Hopefully in another couple of months my Apo(B) in particular will continue moving towards the healthy range.

I've always known that eating well and exercising are things I should be doing but tying them to specific data that affects how long I'm going to live for, gives me a level of motivation I've not felt before.

Thanks to Dr Mager for reading a draft of this.

HN discussion here.



For the full explanation of how cholesterol works, I'd really recommend reading The Straight Dope on Cholesterol. It's an incredibly detailed set of articles though. This is a fairly good summary of the main points, though still a decent read itself. I'll attempt to give a super simple nutshell explanation here:

  • Cholesterol comes both from the food you eat (only animal food products) and is also produced by your liver (it's present in every cell in your body).
  • The type of cholesterol from food is typically too large in size for your cells to absorb it, so it just passes straight through.

Avoiding Heart Disease

This summer my uncle went to see a doctor after complaining about experiencing shortness of breath. He's an officer in the Indian army and has a regular medical checkup every six months so he wasn't expecting to hear anything too shocking. This time, after some prodding from a cardiologist family friend, the doctors did a more extensive checkup. They told him that four of his coronary arteries were severely blocked. It was amazing he hadn't already had a heart attack. He underwent a quadruple bypass and thankfully is recovering well now.

I happened to be traveling in India at the time and learnt from my family that my grandfather had a fatal heart attack around the same age my uncle was now (mid 50's). He was also an officer in the army and seemingly in good health. I began wondering if a similar fate lies in store for me. The more I considered it, the more ridiculous it seemed. How could someone spend years thinking they're in good health with no idea that they're actually a ticking time bomb?

I decided to be proactive about managing this risk. I wanted to know what data I should be looking at and tracking over time so I could avoid being completely blindsided one day. I went to see a doctor and explained everything above. Her response, "you're young and healthy, come in for a physical once a year to check your cholesterol numbers and don't worry about it - just eat well and exercise regularly". I found this quite frustrating for two reasons:

  1. My cholesterol numbers came back fine (just like they were last year) but I knew cholesterol couldn't be a sufficient indicator of a healthy heart. My uncle had cholesterol numbers that were fine too.
  2. The advice to "eat well and exercise" didn't seem particularly insightful. To me it felt a lot like saying "don't worry about things for now, come back in 20 years when you have some actual symptoms".

So I started doing my own research into predictors of heart disease risk beyond monitoring cholesterol numbers. I came across three blood tests that seemed important; Apolipoprotein_B (ApoB), High Sensitivity C-Reactive Protein (hs-CRP) and Lipoprotein(a) (Lp(a))[1]. There appeared to be evidence[2] that all three taken together could provide a more accurate overview of heart disease risk than cholesterol alone.

I ordered these tests[3] and found that my levels for all three were elevated and fell into the "at risk" category.  There were a few conclusions I drew from my results:

  • While it's not exactly good news discovering that you have an elevated risk of heart disease, it made me feel better that I wasn't flying totally blind anymore. My mind immediately switched to researching ways to improve these numbers.
  • There seem to be fairly clear guidelines on how to go about reducing Apo(B) and hs-CRP numbers e.g. exercising reduces Apo(B) (though interestingly not LDL ("bad") cholesterol) and eating omega-3 fatty acids reduces inflammation.
  • Lp(a) seems trickier to manage. It appears to be mostly hereditary and I can't find a clear consensus on how to reduce it. It occurs to me that while I thought I wanted to gather data on every aspect of my heart, perhaps what I actually wanted was just the data about the things I can control.
  • My cholesterol numbers were checked again during these tests, in just two weeks my total cholesterol had dropped 14%. It seems incredible that such an important number can fluctuate so much over a couple of weeks and yet most people only measure it once a year, at best.

I've been thinking about whether taking this approach to managing my health is a good idea or not. It's certainly a slippery slope. There's always a new study to read indicating some additional marker or factor that should be taken into account. After a while it can seem somewhat overwhelming. I've been wondering though why didn't I have this level of discussion with my doctor? There are of course the obvious practical reasons. Doctors are busy and explaining heart disease risk to a healthy 28 year old isn't  viewed as a priority. Also running blood tests costs money and it's hard to see the insurance companies being willing to foot the bill for tests that aren't deemed necessary.

Ignoring these for a moment though, I've been thinking about the possible philosophical objections to treating managing your health as a data project. Here's what I've come up with so far:

Heart disease is complex and there's still debate about the validity of these additional markers
There's also debate about how useful linking cholesterol with heart disease risk is and yet that's still a standard test. There certainly appear to be enough studies and trials arguing that hs-CRP, ApoB and LP(a) are important so surely these numbers can't be completely irrelevant. If I want to take a proactive and preventative approach to managing my health, surely gathering the most information possible to form a plan of action seems reasonable?

The advice i.e. manage diet/exercise/stress is still the same. You should be doing it anyway.
True but I find that tying actions to specific numbers/metrics is better e.g. it's easier to lose weight if you weigh yourself regularly. Generic advice to eat well and exercise is not as effective as having a specific number you're trying to improve on e.g. lowering your ApoB count. Being specific also allows you to create a more personalized plan of action.

Optimizing for specific factors before fully understanding them can have unforeseen consequences
An example would be the conclusion by Ancel Keys in the 50's that animal fats caused heart disease, resulting in the American Heart Association pushing the idea of low fat diets to the public. By promoting fats as evil, the real danger of sugar and refined carbohydrates were ignored[4].

I'm not advocating pushing any particular theory or medical opinion on people. What I want is a system where I can choose to have access to my data with the explicit acknowledgement that the behaviour of the human body is inherently complex and uncertain. It's then my decision what course of action I want to take, taking the informed opinion of my doctor into account.

Numbers are constantly fluctuating, monitor them too frequently and you'll get stressed
This is actually a direct quote from my doctor. In my opinion it's still not a reason to ignore the numbers. I can opt out of knowing them if I wish but if I want access to that data, why should anyone but me have the right to make the decision? The role of a doctor should include telling me what fluctuations are normal and which ones aren't, rather than just checking in with me once a year.

I'd really like to hear more about what people think about this. Am I missing other major downsides to taking a more proactive approach to managing my own health?  Are there other things I should consider looking at to get a complete picture of my health?


[1] ApoB: This is a protein found on the surface of LDL ("bad cholesterol") particles. A regular cholesterol test tells you how much cholesterol is contained within the LDL particles but doesn't tell you either the number or size of these particles themselves. Each particle contains one molecule of ApoB i.e. ApoB count == the number of LDL particles you have. This means you can also make an inference as to their size. If you have a high ApoB value, your LDL particles are likely to be small and dense which are believed to be a stronger indicator of heart attack risk. There seems to be some debate both for and against the importance of ApoB as a heart disease risk indicator that's more valuable than just measuring LDL. The existence of the debate alone seemed reason enough for me to add it to my list of data to track.
hs-CRP: This is a protein found in the blood and its presence is a sign of inflammation in the body, which is a risk factor for heart disease.  While testing for hs-CRP alone isn't more predictive of heart disease than a regular cholesterol test, it does provide valuable additional information. If your hs-CRP level is higher than 3.0mg/L you're at high risk for cardiovascular disease, according to the American Heart Association.
Lp(a):This is a different form of LDL (Low Density Lipoprotein - also known as "bad" cholesterol) which attaches to a protein called Apo A. It's apparently unclear what Lp(a) actually does but if your level is greater than 30mg/dL it's deemed an increased risk factor for heat attack.


[3] I looked into how I'd get these tests ordered myself and found these options:

  • WellnessFX: A nice looking website that has various packages of tests you can run. The cheapest package that included hs-CRP, ApoB and LP(a) was their Baseline (, priced at $149. It also includes a variety of other tests
  • Directlabs: A dated looking website but offering the option of a la carte tests. Total cost for these three tests: $205
  • Health Tests Direct: Another dated looking website, again with the option of a la carte tests. Total cost: $115.50
I decided to go with WellnessFX since for marginally more money I could get a variety of other tests too.

[4] The opening chapters of Good Calories, Bad Calories contain a detailed explanation of the history involved.