We don’t have free will. Change my mind.

Be it fate, choice, or a downstream effect of our universe’s tendency to gradually decay into chaos, I ended up doing what I love this summer, and surrounded by some of the most interesting people I’ve met. One could not ask for more. But this begs the question, “Would I have made it into such an amazing situation without utilizing a tiny bit of free will?” Yes. We may not be as random as electron tunneling, but we are still physical beings governed by the same physical phenomena which characterize the rest of our universe. By some chance, our brains ended up being wired to draw useful associations between the impressions experientially left upon them while also becoming tuned enough to exact fine manipulation of their surroundings. Nothing more can be said.

Since I last posted, the volume of my data has grown considerably. I have now been able to locate the readiness potential in over 15 research participants, and under a variety of conditions. I have also fine-tuned the code which produces summary pages for entire recording sessions (the printouts are shown below). Throughout the summer, I have found the readiness potential within paradigms involving closed eyes, open eyes, spontaneous movement, random movement, and timed movement. Overarchingly, I have recognized a few trends in the data which could be exploited to allow for the real-time prediction of movements, which is my plan for the remainder of this fellowship. With additional differences in observed mu rhythm activity between spontaneous and planned movements, it is possible that small tweaks to the experimental procedure could prompt more predictable responses. Spontaneous movements tend to yield sharper readiness potentials (possibly indicative of planning) and a sharper reduction in mu rhythms closer to the onset of movement, which is why the standard experimental procedure I’m using (an adaptation of the classic Libet task described in my previous post), would theoretically yield the best results. The readiness potential and mu rhythm dynamics will be the first predictors explored in the post-hoc analysis of prediction efficacy.

The top chart is for a spontaneous Libet task, and for the second session shown, the task was to flex every third rotation of the clock on the 12. The third channel was faulty during this specific recording session.

In order to better reproduce Libet’s findings, each event (wrist flexion) is separated by a reiteration of the instructions and the subject confirming that they are ready for the next trial. When it comes to real-time prediction, it will be important to separate the conditions for classification into the discrete windows, such as the time after the trial has begun but before the event occurs. Additionally, the average wait time between trial onset and event can be considered to generate an exploitable probability distribution for the prediction algorithm. Wait time could end up also factoring into the readiness potential dynamics.

I am presently constructing the code which will train a classifier on previously collected data to test its ability to accurately detect the movements that occurred during those sessions. The magnitude of the readiness potential still falls below the magnitude of typical EEG signals, so other methods for adjustment of the voltage-time series are also being explored, such as monitoring the average over a set bin size through use of convolution with a pulse, or rectangle function. The results of these analyses will be reported in the final blog post. The best post-hoc methods will be used in the real-time interface.

I saw a beautiful rainbow over Ann Arbor the other day. Hopefully it’s a sign that my research will yield publishable results.