All posts by Danny Gallagher

Machine Learning in Neuroscience: A Primer

Machine learning has exploded onto the neuroscience community over the past few years. Everyone wants to talk about it if only to use the words “machine learning” or “artificial intelligence” (which is a far broader category, but is often incorrectly used interchangeably with machine learning). Machine learning is far more than a buzzword- it allows us to analyze rich datasets in exciting new ways. However, as with any new technology or methodology in our field, it is important to distinguish between capability and hype.

The basics- What is machine learning?

To help with this, I am here to give a brief primer on machine learning in neuroscience. My goal will be to introduce you to the basic concepts of machine learning- so no specific algorithms or hieroglyphic-looking formulas. Hopefully, by the end of this post, you will have a working knowledge of what machine learning is, how it is used in the neuroscience community, and what precautions need to be taken when using it. So, let’s make like our computers and get learning!

Traditionally, programmers would give their programs specific sets of rules which would determine how they acted. The programmer would have to anticipate various conditions which might occur and set the program to act accordingly. So, if your program trades stocks, you may tell it to buy a stock if it dropped to a certain percentage of its year-long average, and to sell if it rose in the same way. A real stock-trading program may include hundreds more parameters, with various combinations all leading the program to take specific, predetermined action. There are obvious difficulties with this. For one, it is time consuming for a programmer to try to predict every combination of parameters and determine how a program should act in each situation. Secondly, what if the programmer doesn’t know what to do in the first place? If the programmer does not know when to buy and sell stocks, the program will only make the same mistakes.

Enter machine learning. At its core, this is the idea that we can give computers data and let them learn from it, ultimately using it to make predictions. Though this sounds simple, it’s a huge advancement in computing. Now, instead of the programmer defining each way the program acts, he/she feeds it data and gives input on how it should learn. The stock-trading program, for instance, may be given a decade’s worth of data on hundreds of different stocks and try to classify which patterns lead to success and which lead to failure. Now the program can use those patterns to decide when and what to buy and sell.

Computers that know us better than we know ourselves

In neuroscience, machine learning is typically used to classify states, particularly when using rich datasets consisting of multiple different properties, like electroencephalogram (EEG) or combinations of peripheral nervous system measures (skin conductance (EDA/GSR), heart rate (ECG), etc.). The hope is that a sophisticated computer can uncover patterns which a human analyst otherwise could not. Scientists use this method frequently to examine states of cognitive load (how much stuff your brain is trying to process) (Kothe & Maekig, 2011; Stevens, Galloway, & Berka, 2006; Berka et. al, 2007), and many are trying to apply it to the detection of various emotions (Soroush, Maghooli, Setarehdan & Nasrabadi, 2017).

To understand how this works, it helps to break the process down into simple steps. For context, I will use a composite of a rather common experimental archetype in which investigators want to classify emotions (see Heraz, Razaki & Frasson, 2007; Sohaib, Hagelbäck, Hilborn & Jerčić, 2013; Wang, Nie & Lu, 2014). In our case, they want to use EEG to determine when somebody is happy or not.

The steps I describe will not capture the full complexity of practical machine learning processes (there are many tiny but important decisions a machine-learning practitioner has to make throughout the process), but will provide a basic framework which may help understand the pipeline.

A simplified version of a typical machine learning pipeline

Let’s break it down:

  1. Feed the program data. You have to give your program data from which to learn. This is referred to as training data. In our case, we want to classify happy states, so we need data from when our subjects are happy and when they are not. Many experiments try to do this by collecting physiological data (like EEG) while people look at happy or neutral stimuli (often pictures, videos, or audio). We can then tell the program which training data means “happy” and which means “not happy” so that it knows how to classify future data. Our definitions of “happy” and “not happy” are referred to as our ground truth, or what we tell the program is true. This will be discussed in more detail later.
  2. Create a model to fit the data. Next, the program tries to extract patterns from the data. The hope is that our model represents a general pattern of physiological “happy” and “not happy” which we can then try to fit to future data, allowing us to later determine one’s emotional states. There are many ways of going about this, but to describe them specifically is out of the context of this post (for a relatively easy read about common algorithms, try this. For a more adventurous review that pertains specifically to neuroscience, try this).
An example of what a simple, regression-based model might look like. The curve does its best to differentiate between the two datasets (represented by different colors). This model may then be used on new datasets, with points falling below being classified as orange and points above as blue. Note, due to outliers, the classification is not 100% accurate.
  1. Test and refine our model. In practice, our first attempt at building a model may not accurately fit our data. For instance, maybe it only correctly classifies data from our training set 50% of the time (this percentage is often referred to as a model’s “classification accuracy” or “prediction accuracy”). In this case, we will want to refine our model, either by tweaking what we have or using an entirely different algorithm. By doing this over and over, we can hopefully create a model that can give us an accurate classification.
  2. Validate our model. Perhaps we’ve reached the point where our model correctly classifies our training data a high percentage of the time. However, our goal is not to classify our training data, but to classify future data, when we don’t know the “answer.” Thus, we need to see if our model generalizes to other “unseen” data. To do this, practitioners often split the original data, training the model with some and validating it on another (for more detail on how this is done in modern processes, see here). If the model does not accurately predict the test set, we need to rework it or try a new one.

    A model that fits the training data very well but does not fit unseen data is referred to as “overfitting.” It is actually a very common machine learning problem. The model trains too much and becomes so optimized to the specifics of the training data that it now only represents that sample as opposed to the whole population. For instance, this model goes out of its way to work around specific data points, even though those points seem to be outliers and likely do not generalize well. For this reason, if someone tries to tell you that their model has a 98% prediction accuracy, make sure you ask about how it performed in validation tests!
  3. Profit! After iterating through these steps enough times to create a good model, we can use it on future data to try to determine when people are happy in other situations.

 In a field where different activity is often represented by incredibly complex patterns, the ability to have a machine learn these patterns for us is a massive step. While a human experimenter may be able to infer conclusions from a certain defined metric (like the amplitude of a waveform at a given time), a machine can infer from minute changes in many different metrics at once. Considering that some aspects of physiological responses may be shared across many different mental states, the ability to use multiple different aspects of your data at once is a huge advantage. Though the value of one parameter may potentially be indicative of a range of different conclusions, the combination of many may help us significantly narrow down that range.

If only it were this easy…
©momius [Adobe Stock]

So there you go. All we have to do is give our computers the data and let it decide what means what. Problem solved. Great article.Machine learning is also used frequently for real-time applications, such as brain-computer interfaces that use physiological feedback to adjust in the moment. An example of this is an interface that adapts to someone’s cognitive load so as not to overburden the user with information. Instead of needing an analyst to monitor the data and tell a program what to do, the program decides for itself.

As you could probably guess, it’s not quite that simple.

Glitches in the system- the problems with machine learning

As I mentioned, machine learning represents an incredible advancement in data science and can be a valuable tool in a neuroscientist’s kit. However, it is by no means perfect. Any academic will say that you can’t simply give the computer your data and expect an accurate result (contrary to what some marketers may suggest). The problem with artificial intelligence in this case is that its “intelligence” is artificial. It doesn’t know anything about what you are measuring; it’s just finding the best pattern it can to describe a series of numbers. A common phrase in the field is “garbage in, garbage out,” meaning that your output is only as good as the underlying data behind it. So, in neuroscience specifically, what adds “garbage” into our conclusions?

Noisy data. If you have ever worked with physiological data, you don’t need me to explain the problem of noise. “Noise” refers to the factors which you are not interested in measuring, but affect the data anyway. These factors distort the “signal” in which you are actually interested. In physiology, noise is commonplace in essentially every measurement. Whether it be random electrical noise or artifact (data arising from unwanted influences) caused by a participant moving too much during the experiment, the scientist will have to deal with a lot of “garbage” on top of the actual signal of interest. A great hope of machine learning is to cut through random noise to find a true underlying pattern, but again, a computer will not know which data is signal and which is noise. Thus, the noisier your input data, the noisier your output.

This is especially true when the noise is not simply random, but introduces bias. For instance, going back to our happiness experiment, if some of the “happy” stimuli cause participants to laugh, the electrical activity caused simply by the movement of laughter will influence the EEG data. If that same movement is not present when the participants view the neutral stimuli, we now have noise that is specifically biased to one condition. This bias means our machine learning program may be more likely to interpret the noise as part of the pattern of interest. When we then go to apply our model to future studies, it may be more likely to call someone “happy” simply because they are moving around.

Thus, as with any experiment, it is imperative to analyze the cleanest data possible. The easiest way to achieve this is through good experimental design and solid technical recording practices. Basically, record as little noise as possible in the first place. After that, you can apply filters and reject/correct “dirty” segments of data. Just know that these processes will, by nature, distort your data, and so must be used with finesse. It is also important to know that these procedures work better after the fact, as opposed to in real-time, which is why I always encourage people to analyze the results after all the data has been collected if real-time monitoring is not imperative.

At the end of the day, it is important to remember that machine learning is not a substitute for collecting good, clean data. Even if your model’s prediction accuracy is very high, it is not useful if it’s predicting a biased dataset. This brings me to my next point.

Ground “truth.” As mentioned, using machine learning to classify things requires the use of a “ground truth,” where we tell the machine what the training data represents. This is referred to as supervised learning, since we are supervising the way the program trains by telling it what certain datasets mean. You can also create an unsupervised program without a ground truth if you simply want to cluster data or make associations, but this is less utilized in neuroscience due to the interest in classifying mental states.

A perfect ground truth is objective. We know that dataset A represents B. However, this is rarely the case in the real word- especially when dealing with something as complex as the brain.

Let’s go back to our happiness indicator. As is common in these types of studies (Heraz, Razaki & Frasson, 2007; Sohaib, Hagelbäck, Hilborn & Jerčić, 2013), let’s say we determined “happiness” by showing participants “happy” pictures from a validated database.

Happy (left) and neutral (right) pictures, taken from the Open Affective Standardized Image Set (OASIS) database (Kurdi, Lozano, Banaji, 2017)

The first question we must ask is “did these pictures actually make the participant happy?” We can (and should) assess this by simply asking them to rate how positive or negative they feel after each picture. Of course, there is always the possibility that our respondent is lying, perhaps because they don’t want us to think of them as some monster who hates puppies in teacups.

Even if we could reasonably determine that the pictures made our respondents happy, we don’t know what other reactions those same pictures instilled. Perhaps the pictures were all arousing, biasing our algorithm toward other arousing stimuli. Or perhaps the pictures triggered nostalgia, leading to additional activity that had nothing to do with emotion, but instead had to do with the process of retrieving memories.

In addition, it is hard to be sure which physiological reactions are stimulus-specific. What if our program picks up on responses that have little to do with emotion, and much more to do with our reactions to certain colors or certain physical features? We can assuage this by “aiding” our program, deciding to include only features of the data for which we have some contextual reason to expect might be related to our variable of interest. However, given the complexity of our bodies, and the fact that certain signals are often correlated to multiple different states or activities, this may be easier said than done.

For similar reasons, it is hard to know whether the “happy” patterns generated by our pictures generalize to other stimuli. Does it also apply to video? Audio? What about real life? Neuroscientists traditionally attempt to keep stimuli between conditions as similar as possible because they know that even small changes can affect our brains’ responses.

The crux of the issue is that our ground truth is anything but objective. We are complex creatures and our responses to anything are going to be equally complex. Our machine learning model may predict both the training data and unseen data very well, but that doesn’t mean it predicts our variable of interest. This may seem hopeless (if you are feeling down, I refer you back to the puppy in the teacup), but it does not mean that our conclusions are useless. It simply means that we cannot simply take what our program says on blind faith. Just as in traditional experiments, we have to approach our conclusions with skepticism, and consider in the experimental design process how to minimize potential confounds.

Speaking of experimental design, I have one more point that is often overlooked…

We’re all basically the same, right? Of course not! We are individuals! Inside each of us surges a wild entanglement of biology and experience- millions of tiny differences forming the lattice that separates us from the rest of the world. It allows us to make our own unique marks on the world. It allows me to be me and you to be you. It also makes neuroscientists hate both of us.

People are variable, as are their physiological signals. This is why sample size matters; only by aggregating enough people can we try to smooth out this variance and start generalizing conclusions. And that’s for one metric. Now imagine a program that is trying to glean patterns from dozens of different metrics, all of which can vary across each participant. For this reason, it can be difficult for machine-learning models to accurately predict certain states.

Some scientists suggest that you can create more accurate models by training subjects individually, then using these individualized models to make predictions about the same person (Berka et. al, 2007). However, this requires you to lengthen your experiment to train these participants, which may be less feasible. In addition, this paradigm holds no guarantee that you will be able to achieve a high prediction accuracy, whereas that information would be available in a previously trained model. Finally, even when training individually, it is important to make sure that the subject is in the same mental state during training and during the actual study. For instance, if you show them pictures for an hour before even starting the experiment, they will likely be more fatigued when looking at new items than they were when collecting data to create our model. Before deciding to use machine learning, you should be cognizant that it may require an additional investment in your experiment.

Wrapping it up

Hopefully, you now know the basics of machine learning, how it is applied in neuroscience, and when we should be skeptical. The technology is truly amazing, and I assume will only become more prevalent in this as well as many other fields. However, it is not yet perfect. As scientists, we cannot shirk our duty to the truth. We must understand that a machine learning program knows nothing of neuroscience or emotion or whatever else we feed it- it only knows the data it’s given. It is therefore useful as another tool in our belts. We use it to dig ever deeper to the truth- but ultimately the conclusions must be our own.

Works Cited

Berka, C., Levendowski, D. J., Lumicao, M. N., Yau, A., Davis, G., Zivkovic, V. T., … & Craven, P. L. (2007). EEG correlates of task engagement and mental workload in vigilance, learning, and memory tasks. Aviation, space, and environmental medicine78(5), B231-B244.

Gupta, P. Cross-Validation in Machine Learning (2017, June). Retrieved from

Heraz, A., Razaki, R., & Frasson, C. (2007, July). Using machine learning to predict learner emotional state from brainwaves. In Advanced Learning Technologies, 2007. ICALT 2007. Seventh IEEE International Conference on (pp. 853-857). IEEE.

Kothe, C. A., & Makeig, S. (2011, August). Estimation of task workload from EEG data: new and current tools and perspectives. In Engineering in Medicine and Biology Society, EMBC, 2011 Annual International Conference of the IEEE (pp. 6547-6551). IEEE.

Kurdi, B., Lozano, S., & Banaji, M. R. (2017). Introducing the open affective standardized image set (OASIS). Behavior research methods49(2), 457-470.

Le, J. A Tour of The Top 10 Algorithms for Machine Learning Newbies (2018, January). Retrieved from

Lemm, S., Blankertz, B., Dickhaus, T., & Müller, K. R. (2011). Introduction to machine learning for brain imaging. Neuroimage, 56(2), 387-399.

Sohaib, A. T., Qureshi, S., Hagelbäck, J., Hilborn, O., & Jerčić, P. (2013, July). Evaluating classifiers for emotion recognition using EEG. In International conference on augmented cognition (pp. 492-501). Springer, Berlin, Heidelberg.

Soroush, M. Z., Maghooli, K., Setarehdan, S. K., & Nasrabadi, A. M (2017). A Review on EEG Signals Based Emotion Recognition. International Clinical Neuroscience Journal4(4), 118-129.

Stevens, R., Galloway, T., & Berka, C. (2006). Integrating EEG models of cognitive load with machine learning models of scientific problem solving. Augmented Cognition: Past, Present and Future2, 55-65.

Wang, X. W., Nie, D., & Lu, B. L. (2014). Emotional state classification from EEG data using machine learning approach. Neurocomputing, 129, 94-106.

Your Brain on Bitcoin

You buy in with some spare bucks you have lying around. You’ve heard the news of lucky young investors making millions, and you decide to get in on the action while you can. At first, you check every couple of days- whether it increases or decreases, it’s not the end of the world. But over time, something starts to click. The price has increased- you just made some spending money. Checking once in a while turns into once a day turns into once an hour. You start selling and buying more. Just when you think you have it figured out, the price plummets, and when you lose all hope, it spikes again. Regardless of how much money you’ve made or lost, one thing is certain: when that price goes up, it feels good.

© [Parilov] / Adobe Stock

Chances are if you have invested in cryptocurrency or even talked to someone who has, this story sounds familiar. Reaching the price of $10,000 on November 29, Bitcoin quickly became famous (and, to some, infamous), taking only two weeks to nearly double to its peak of over $19,0001 before crashing back down to a measly $8,000 or so. While it is difficult to predict if the cryptocurrency’s value will resurge, it is clear that, at the very least, Bitcoin and its altcoin counterparts have attracted a whole new market of investors2. Whether cryptocurrency is the future of everything or a doomsday bubble sent from the dark web to punish us for dreaming, its explosive growth provides a fascinating look at how our brains respond to rewards. This is because, unlike stocks, it is hard to see the utility of a Bitcoin3 (though the Blockchain, or the technology on which Bitcoin is based, is a different story). Much like a paper currency, the value is almost entirely in the minds of the people willing to use it.

Price of Bitcoin in US dollars, from 2015-today.

This begs the question, what are the psychological drivers behind diving into this new market? What caused the value of cryptocurrencies to rise and fall so dramatically in such little time? To answer this question, we must combine insights from social psychology, behavioral economics, and neuroscience.

Everyone’s doing it, so it must be right.

Representation of stimulus from Asch’s (1951) conformity experiment. The goal was to identify which line on the right was most equal in length to the line on the left. When surrounded by others choosing incorrectly, participants chose incorrectly on 32% of the total trials, as opposed to less than 1% of the time when the participant was alone.

Humans are naturally social creatures. The actions of others influence our own decisions constantly. This generally happens out of a desire to fit in (normative conformity), an idea that others know more about the subject than you (informational conformity), or both. Solomon Asch famously demonstrated4 how social pressure can cause people to make overtly wrong decisions; in this case, judging two clearly unequal lines to be the same length simply because a room full of people said so. While the effect of conformity is not this strong for everyone, it is still an important driver of our decision-making, even at a physiological level. In fact, when we conform, the brain recruits neural mechanisms like those involved in reinforcement learning5. For instance, when our decisions conflict with the group’s, activity in the nucleus accumbens, an area of the brain heavily associated with reward, decreases. Basically, our brains keep track of when our beliefs differ from those of others and may try to nudge us in the other direction.

Locations of several parts of the brain involved in reinforcement learning, including the nucleus accumbens. Brain illustration ©[nicolasprimola]/Adobe Stock
In the case of Bitcoin, it is easy to see the role of conformity. As the number of people championing Bitcoin increased, so did the power of social pressure, especially if you knew very little about cryptocurrency. We start to think that if all those people are doing it, they must be onto something, and we don’t want to be the chumps who missed out.

Of course, conformity works both ways. When the price drops and people start to panic, those more susceptible to social influence often follow the crowd once again and sell. This has been seen in the past few months. As adverse news spreads, often about Bitcoin’s bubble-like nature6 or about potential new regulations7, bearish investors panic and sell before the price can drop. As more people sell and the price drops, those with little knowledge of the market often have one source of recourse- to again follow the group. Thus, the price plummets further, and the panic is fueled until either the market collapses or a change convinces enough investors to start investing once again. This contributes, in large part, to the volatile, swinging nature of cryptocurrency. Importantly, whether conformity results in wise or unwise decision-making, it is an undeniably powerful influencer of our decisions. It is a mistake to view perception solely from the standpoint of the isolated individual, since humans largely do not act in isolation.

Working with what we have

Of course, while conformity can be influential, it is not the only cognitive process at play. Indeed, the way we respond to social influence can vary based on our own beliefs, even sometimes pushing us further away from the group consensus. Thus, a large piece of the puzzle is perception and how we internalize the cryptocurrency-related information around us.

As it turns out, the way we process this information can be influenced as much by our own minds as by the information itself. This is due to the brain’s nature to operate heuristically, or by using preconceptions to inform future decisions.  It does this for an important reason- if we had to fully process all available information every time we were presented with a choice, we would be paralyzed with each trip to the grocery store. Yet, as Behavioral Economics tells us, it can also cause biases that lead to illogical decision-making.

When examining cryptocurrency, a source of bias that immediately comes to mind is the availability heuristic. This refers to the tendency to make decisions based on information that is most readily available in memory. Often, such information will be a particularly salient, emotion-inducing, or a recent event. For instance, if I was asked my opinion on enforcing a mandatory curfew in my neighborhood, I would be much more accepting if I had just seen a news story about a string of nighttime muggings.

With Bitcoin, these events take form as stories of people whose lives were transformed by Bitcoin. As the price started to soar, stories emerged8 of “Bitcoin millionaires,” a lucky group often hailed as “crypto geniuses” who made a fortune from Bitcoin. Whether these newfound millionaires quit their jobs and founded their own companies or simply spent their days traveling around the world, their story sticks with us. It makes us think, what if that had been me. This deep emotional relevancy encodes the story strongly into our memories, so that when we evaluate whether to invest, it weighs heavily on our minds, pushing us to pull the trigger. Of course, for every Bitcoin millionaire, there are many others who lost money through cryptocurrency. Our brains are simply trying to process information as efficiently as possible, causing them to weigh more heavily the information that is easily available. In this case, that happens to be the multitude of striking, get-rich-quick stories. This effect has been utilized in many successful ad campaigns, such as “The Real Cost” series for anti-smoking9. By creating visuals that are particularly memorable and emotionally salient, their campaign sticks in our heads, and in turn, disproportionately drives our decision-making.

Importantly, these perceptions often act like snowballs, in that they are largely determined by initial belief formation. This is mainly thanks to two other cognitive biases, the anchoring effect and confirmation bias. The anchoring effect refers to the tendency to rely more heavily on the information we see first. This is a large driver of the effectiveness of discounts- I will likely attribute more value to $100 pair of pants discounted by 50% than the same pair of pants at a starting cost of $50, because the former situation anchors my perception of how much the pants should cost to $100. Future evaluations are then judged from that $100 anchor. Confirmation bias is the tendency to search for evidence that confirms our beliefs and ignore evidence that contradicts them. This bias often plays a role in customer loyalty. For instance, since my first phone, I have been an Android loyalist, originally due to some of the physical features of the phones when I had first purchased. Now, even though the specs of all the phones have changed drastically, I cannot deny that I evaluate Androids and iPhones differently. When I look at Androids, I usually focus my search on cool features and benefits, whereas when I look at iPhones, I often catch myself looking for flaws.

It is easy to see how these biases manifest in the crypto market. If the first information you see about Bitcoin is a story of a Bitcoin millionaire, the anchoring effect and availability heuristic work together to weigh your perception of Bitcoin toward it being positive. Confirmation bias then causes you to evaluate further information through the lens of “Bitcoin is good,” over-trusting data in its favor and over-criticizing skeptics. By playing off how our brains make sense of the world, these stories become an incredibly effective recruiting tool.

Of course, these biases work both ways, and some may conflict and outweigh others. If the first story you see about cryptocurrency is somebody who lost their savings, you will likely view crypto praise with a great deal of skepticism. Along the same vein, if the market starts to plummet and the stories of Bitcoin millionaires are largely replaced by these strongly negative ones, the availability heuristic can cause a shift in opinion, even if you were originally pro-crypto. The negative stories play an equally important role as the positive ones, spurring the panic and fear that has recently sent the market plummeting back down to Earth.

Winning makes me feel good

Molecular structure of dopamine, © [petarg]/ Adobe Stock
Our description of the crypto-obsessed brain still feels incomplete. While we have described some of the motivators of spikes and panics from an information-processing perspective and an emotional perspective, we are still missing something even more carnal. After all, even amidst panic, something about making money from cryptocurrency just feels good. Quite frankly, it’s fun. To understand this, we must dive into our neuroscience roots and examine everyone’s favorite neurotransmitter- dopamine. Dopamine is commonly associated with reward and addiction. When your brain thinks something positive has just happened, it releases dopamine to make you feel good, conditioning you into further doing whatever brought about the positive outcome.

Yet, as with most neuroscience, it’s not quite that simple. The release of dopamine is not a binary process (yes for positive and no for other). It is modulated by expectations; more dopamine is released for outcomes that are better than expected10. The encoded difference between the positivity of outcome vs our expectations, referred to as the reward-prediction error, is heavily related to our brains’ roles as pattern-seekers. Essentially, our brains are constantly looking for patterns in the environment- this is key to learning and adapting. And though brains are great at finding patterns, they are not as good at detecting randomness. Even when there is no pattern to be discerned, the brain will try to find one anyway, using dopamine as a signal that something might be working. This adds further clarity to why Bitcoin became so popular. The incredibly volatile nature of the coin means that discerning a pattern in the data is incredibly difficult. Thus, the crypto-holder’s brain is never able to adequately establish a reasonable expectation of reward, instead stuck in a state of uncertainty. This uncertainty causes the brain to respond to each price surge with maximal amounts of dopamine, since an expectation of reward can never be properly established. “Winning” with cryptocurrency then becomes proportionately more addicting than achieving low but steady yields on a predictable, blue-chip stock. The same process is reminiscent of that of habitual gamblers11. These effects may wear off when losses become more consistent, potentially compounding the effects of sustained price drops, but often all it takes is a single spike to hook us back. Perhaps the most interesting aspect of all of this is that the volatility that scares so many investors away may actually play a large part in keeping crypto-holders in the game.

Lessons to be learned

©[zeman88]/ Adobe Stock
Of course, the cryptocurrency market is far more nuanced than can be explained in 2,500 words. For one, investors are beginning to realize that not all cryptocurrencies are created equal- what is true for one currency may not be for another. Indeed, as the flash of the Bitcoin boom starts to settle, it would make sense for the specifics behind each coin to start to garner more focus. Thus, there are several deliberate thought processes that may start to play larger roles in buying a cryptocurrency. Perhaps you trust the potential of the technology upon which the coin is based, or the coin raises funds for a cause you support, or you just find it funny (you mean I’m supposed to let something called Dogecoin go un-bought?).

There are also many other psychological processes that likely factored into the rise of the crypto castle. Furthermore, the effect of these biases can vary among individuals, especially among those with different levels of knowledge of finance and the crypto market. Yet, the phenomena I discussed here share a common ground in that, in addition to affecting the crypto market, they are also essential to the field of neuromarketing.

On the marketing side, it is important to keep these biases in mind when attempting to influence consumer purchasing. Marketers already do this every day, whether it is paying social influencers to convince their peers to buy a product, using highly emotional ads to increase relevancy, or adding an element of randomness to reward deals to stimulate dopamine release. If you do not similarly take advantage of the consumer brain, you will be fighting an uphill battle against a competitor who does.

On the research side, biases are less tools and more plagues that we, as scientists, must do our best to combat. It is easy to conform to the beliefs of other researchers who you think may be more knowledgeable than you, even though the point of rigorous science is to ask questions and disprove hypotheses. It is also easy to fall into the traps of availability, anchoring, and confirmation bias, where we get so attached to our initial theory that we fail to accurately examine subsequent evidence. Finally, it is easy to get caught up in the rush of finding unexpected but positive results, drawing conclusions before first questioning or replicating them. This is particularly dangerous in neuroscience, where the sheer amount of data means that careless researchers can find “significant” differences simply by chance and not because of any underlying neural mechanisms.

At HCD, we constantly remind ourselves that, as with investing, it is important as researchers to recognize that we are flawed. We must keep conscious the limitations of not only our data but of ourselves as people, and actively work to overcome them. While the thought can be discouraging, it is essential to performing good science. This is a problem that we can overcome through rigorous questioning and understanding. But, as with most problems, the first step in solving it is to realize it exists.


  1. Bitcoin Price.
  2. Domm, P. (2017, December). Bitcoin fever is drawing investors into the stock market, Laszlo Birinyi says.
  3. Cheo, J. (2017, December). Why Blockchain Is Real And Bitcoin Is A Mirage.
  4. Asch, S. E. (1951). Effects of group pressure upon the modification and distortion of judgments. In H. Guetzkow (Ed.), Groups, leadership and men; research in human relations (pp. 177-190). Oxford, England: Carnegie Press.
  5. Stallen, M., & Sanfey, A. G. (2015). The neuroscience of social conformity: implications for fundamental and applied research. Frontiers in neuroscience, 9, 337.
  6. O’Brien, M. (2018, January). The bitcoin bubble is a joke, and you’re the punchline.
  7. Sundararajan, S. (2018, February). Mnuchin Calls for Crypto Regulatory Talk At G20 Summit.
  8. Robinson, M. (2018, February). This 24-year-old said he quit his job after making a fortune in bitcoin and other cryptocurrencies — here’s how he did it.
  9. The Real Cost Campaign.
  10. Bayer, H. M., & Glimcher, P. W. (2005). Midbrain dopamine neurons encode a quantitative reward prediction error signal. Neuron, 47(1), 129-141.
  11. Linnet, J. (2013). The Iowa Gambling Task and the three fallacies of dopamine in gambling disorder. Frontiers in psychology, 4, 709.

How One Choice Burned the New Period-Spacing Study

Every once in a while, a study is published that that seems to catch fire. Whether it adds to an existing debate or sparks a new one, it somehow seems to resonate with an issue that some may not have been aware even existed. One such article was recently published in Attention, Perception, & Psychophysics by a group of researchers from Skidmore College. Its premise- at the end of a sentence, two spaces after the period may be better than one.

While this may not be earth-shattering to you, it is for some. In fact, the period-space conundrum is quite vehemently debated among many journalists and typographers. The study, headed by Dr. Rebecca Johnson, recruited 60 college students to read several paragraphs, which used either the one-space or two-space convention. Johnson and her colleagues assessed the time it took students to read, as well as the reading comprehension of each paragraph. They also used an eye-tracker to examine how the students read the ends of each sentence (consisting of the punctuation mark as well as the word immediately preceding and following it), acquiring measures such as fixation time, probability of skipping the section, and likelihood of going back to it later. They found that participants who use two spaces when they type read the two-spaced paragraphs about 3% faster than one-spaced paragraphs. For all students, reading comprehension was the same for both styles of typing. Eye-tracking also revealed that all students fixated on the sentence-ends less for two-space paragraphs and skipped over the sections more. While this finding does not seem to be extraordinarily meaningful, given the available global speed measure, the researchers suggest that the eye-tacking may provide a sign that the stimuli needed to consist of more text for the reading-speed effects to become more apparent.

The article was quickly touted as a decisive blow to the one-space camp- science had finally made up its mind. Then, almost as quickly, it was torn to shreds by all sorts of bloggers and journalists. The criticisms are completely valid. The effect sizes were small and very specific. The student-based population made results difficult to generalize. In any subject, one study only proves so much. Yet, the most damning critique was perhaps that the researchers used a monospaced font for all of the stimulus paragraphs. That means the letters are all allotted the same width, as opposed to having a variable width, like essentially all modern fonts. Though this seems small, it is essential to the debate, since the argument in favor of one space after a period is largely based off the idea that modern fonts, with their variable widths, are designed to look best with one space after the period. Monospace fonts were only ever really used back when the typewriter was king, and that spacing was one of the main reasons why typists followed the two-space rule- otherwise, everything just seemed to blend together. Johnson defended her choice by explaining that most eye-tracking studies of this nature used a monospace font. Indeed, it is a boon when attempting to quantify how people read each sentence, since its consistency makes the study much more resistant to confounds brought about by the structure of different words. Yet, regardless, this does not overturn the fact that the choice left a glaring hole in the study’s applicability.

This leads into my main point. When we strip away the context, what we find is a classic research problem- does one emphasize internal or external validity? Internal validity can be defined as how well a study measures what it means to measure. Are there confounds that influence the data in ways that are not related to the topic of interest? Are the measuring tools accurate and used in the correct context? External validity refers to how well a study relates to the real world. Once you take the findings out of the lab setting and into the much more complicated environment of real life, do they still apply? Obviously, both are important, but any experienced researcher will know that one often has to make a choice that partially sacrifices one for the sake of the other. For the more you try to control a study, the less like real-life it often becomes. So, assuming (just for sake of argument) that we do not live in a perfect, fantast world where all of our dreams come true, researchers have to be able to discern when to emphasize each type of validity. Johnson emphasized internal validity over external validity, but because the subject of that choice was central to the one-space vs two-space argument, the article ended up getting burned. Arguably, if she had chosen a more modern font, the loss of internal validity would not have been nearly as detrimental.

This choice is often emphasized in market research. Marketers don’t want to know how people will react in a lab. Their business depends on how people respond in real life. Thus, many market research studies tend to skew toward being more externally valid. While this is extremely important, it is also imperative to remember that a certain amount of internal validity is needed to be able to conclude anything meaningful at all. It takes an experienced researcher to walk the validity tightrope, which is why any good researcher will let their client know if the protocol skews too far from either standard. At the end of the day, every choice counts, and even small details can throw this balance off. And when the study can potentially save a company millions of dollars, their researcher really has no room to space.

The article in question

Johnson, R. L., Bui, B., & Schmitt, L. L. (2018). Are two spaces better than one? The effect of spacing following periods and commas during reading. Attention, Perception, & Psychophysics, 1-8.