January 27, 2020
This blog post is inspired by me reading Thinking Fast & Slow by Daniel Kahneman while setting up dual experiments with mobile EEG.
I have a friend who has an interesting tradition – each year he publishes a single facebook post and it goes like this:
Hey folks, thanks much for your warm (Bday) wishes! In this virtual world I am unfortunately not able to share a piece of delicious hazelnut cake I’m now having in my fridge, but I can share with you one of my greatest online discoveries in the year behind us
and then he proposes a number of books and blogposts that he has read since the same message last year. That’s how I bumped into Thinking Fast and Slow. The book is simply marvelous and I recommend it to everyone with even the slightest interest in psychology and human behavior. It mostly deals with the biases in thinking and judgment, but it starts with explaining the cognitive load.
This specific example is based on what Kahneman calls “competing for the same, limited mental budget” and I will refer to it here as Dual Tasks. Dual Task essentially means performing two tasks concurrently. We do that every day, not paying attention to it at all (we can walk and talk at the same time… most of us… well, most of the time :)). However, when we add another mental task in parallel, these routine operations get influenced, and we cannot do it fluently at our natural pace. Try the following — while walking with your friend, ask him to quickly name 10 friends from his elementary school. You will see him pace down significantly.
If you focus on all the small things, there is no enough available neurons to do the real important work.
We slow down because even these routine operation consume cognitive power and, having in mind that the total cognitive power is limited, we cannot achieve successful performance when the cognitive load by another task is imposed (in this case naming the friends). However, in research, we are interested in not degrading the performance of the primary task (in the case above, we examine walking). Try asking the friend above to name 10 friends, and insist on keeping the pace constant — he can do that as well. The performance of the routine operation (primary task — walking) will not change, but the cognitive load (naming 10 friends) will be significantly higher, and his ability to name 10 friends quickly, decreases. In this way, by measuring the performance of the cognitive load “naming 10 friends”, we can actually measure how difficult is walking!
To describe the cognitive load, let’s think of it as the following. Imagine a person that can lift 100kg weight when standing.
The same person can carry the load of at most 40kg over the course of 100m.
In this way, we figuratively measure that the load of walking is 60kg. If we further measure that he cannot run 100m with 40kg, but he can run with 20kg, this means that running is 80kg and that 100m running is more difficult than 100m walking. In this way we can measure how difficult are different types of walking (like walking and running) which is difficult to measure per se.
So, Dual Tasks always consist of a Routine Operation (primary task) and a Cognitive Load (secondary task) competing for the same Total Cognitive Power. The interference caused by a secondary task while walking (also called Dual Task Cost) has been examined in multiple studies in the elderly and persons with neurological impairments. The results indicate that a Dual Task paradigm is useful, e.g. for identifying gait anomalies not identifiable during normal walking and for predicting falls.
Today, Dual Tasks are very important and gaining popularity as they can, in some cases, substitute subjective measures and provide more scientifically valid objective results.
Dual Task experiments can be failry simply set up using electroencephalogram (EEG) to measure cognitive load of the secondary task, while a subject is fully focused on the primary task. The early study published in Science by Wickens et al at the university of Illinois describes that in such a setup the event related potentials (ERP) can tell us a lot. What they showed is that, as the resources required for the primary task increase, the resources available for the secondary task decrease, and that this is clearly visible in the P300 component of the ERP response. This component is strongest when the resources are available, and reciprocally diminishes as the available resources decrease.
P300 is the measure that we use to assess the cognitive power (or the amount of yet unemployed neurons) available to perform the secondary task. If there is no primary task, the cognitive power consumed is zero, and P300 is then the measure of total cognitive power.
You can think of the cognitive load figuratively as that you can only focus on a task as long as there are enough available neurons to evoke the P300 response. Otherwise, all of them are consumed with the primary task, and you simply have no cognitive availability to follow the secondary task. That is why you should be careful with multitasking — if you focus on all the small things, there are no enough available neurons to do the real important work.
The emergence of wireless EEG systems today allows also the freedom of movement during data collection. This is extremely important, so that the experiment can be performed in the natural environment of the primary task (clinical, operational, or real-world settings). A good example, apparently, is the gait analysis, where it is utterly important that unrestricted movement is possible. You cannot really imagine walking while wired to a computer (although early studies used treadmills to SIMULATE the environment).
To further make the environment (and walking) as natural as possible, it is a good practice that all the acquired data are saved to a pocket device, and a smartphone emerges as a good candidate. Simply because, in case of walking, walking naturally and walking with a backpack with a computer in it, is not equivalent task.
Some mobile EEG systems (consisted of a mobile EEG device, an EEG cap and a smartphone), are already available on the market (this was the part where I advertise our product, as I believe it’s an awesome device 🙂 ). With it, you are now well-equiped to start these experiments.
So, how hard is walking? Let’s put it on a test together with some other interesting Dual Task studies.
By measuring P300 while walking with different prosthesis alignments, we can find the most natural adjustment.
Christina Schöllig et al., from TU Berlin, studied the gait in lower limb amputees. By training, lower-limb amputees learn to walk remarkably good, and it is challenging to objectively assess how difficult the walking is for them. However, the alignment of the prosthesis is very important as it reduces effort during gait.
To test this, they collected biomechanical parameters of walking by the mobile gait analysis system developed by TU Berlin. They used mobile EEG to assess the changes in the amount of required cognitive resources by measuring P300 deflection.
Their study showed that the participants tend to require more attention for walking with translation of the knee axis as compared to their familiar alignment.
Comparing control algorithms for the prosthetic limb to choose the more natural one (Pattern Recognition Control versus Direct Control)
In another study by Sean Deeny et. al from Institute of Rehabilitation, Chicago, the authors examined EEG/ERPs as measures of cognitive workload to compare the relative cognitive load of two different hand prosthetic control approaches.
Their study was performed on intact-limb subjects. P300 again exhibited a strong cognitive workload effect and was virtually absent in the hard conditions.
Although this particular study was performed only on the intact-limb subjects, the results are promising and should be replicated on the amputation patients and should be adapted for the manipulation and mobility tasks using a prosthesis. Pattern Recognition Control (PRC) showed higher amplitude in difficult cognitive task compared to Direct Control (DC), implying lower cognitive workload in PRC condition.
They also conclude that the ERP approach and other EEG measures are adaptable to a variety of human-machine interface (HMI) tasks as objective outcome measures of cortical and attentional effort.
When setting up experiments outside the lab, scientists should be aware that the results may be affected as there is always another simultaneous task (being out in the environment)
Rob Zink et al., from KU Leuven, Belgium, studied the cognitive load of the three-class auditory oddball task in three different conditions:
1. The subjects sits on a bike
2. The subject pedals a bike on the fixed standard
3. The subject bikes freely on a 500m course on campus
They examined the P300 ERP response evoked by the target tone from the oddball task the participants performed while biking. They concluded that the P300 amplitude was the strongest in the sitting still condition, and was significantly decreased in the free biking conditions, due to the cognitive workload imposed by having to pay attention to the road while biking.
By monitoring assembling workers attention in a workplace, we can predict when they should be sent for a break, to prevent accidents.
Human Factors and Ergonomics (HF&E) is concerned with designing products, processes and systems and their interaction with humans. This field is important as with the advancement in technology, more than 80% of the accidents in factories are due to human error. For long, the methods for assessing the cognitive state of the worker have been dependent on the subjective methods (like questionnaires). As you may assume, those are unreliable.
This is the reason why psychological sensors were included in the ergonomics research with the aim to provide the objective and quantitive measures. Neuroergonomics emerged as a scientific discipline, investigating the human brain functions in related to performance at work, some of them relying on the Dual Task principle.
In the work of Pavle Mijovic et al, they followed P300 of workers in the manual assembly task to try to predict the decrease in attention and if they can increase the attention with instructed responses (e.g. which hand should be used to perform a task. The end goal would be to detect when the attention is decreased in order either to send the worker for a short break, or to influence him with a response, increase attention, and prevent an accident.
Dual Tasks using EEG features (such as ERP) showed to be good tools for discriminating among different levels of cognitive load of the primary experiment. Other EEG features may also be used in these growing trend of using Dual Tasks and we hope to see many more clever experimental setups in the following years.
We at mBrainTrain certainly continue our work with other teams on similar setups, and will be proud to share some of our own work on this blog.
Until that time, check the www.mbraintrain.com/smarting/#references
Stay tuned 😉
Some of the explanations in the text are descriptive (not scientifically fully correct) and serve to explain the basic idea behind complex concepts.