One of the projects I have been working on is to measure one’s irrelevant memory. Wait, what? Yes, irrelevant memory. Let’s say there is an object with color and orientation, like an ellipse with a color. I ask you to memorize orientation. Now, orientation is relevant and color is irrelevant. If I want to test whether color information has been automatically registered (or encoded) in your memory, what shall I do? There is one way to test this. Let’s say we have 100 trials of an experiment. For the first 99 trials, I ask you to only memorize orientation from a display. At the end of each trial, I test your orientation memory. But at the 100th, last trial, I ask you to recall color. And yes, you did not see this coming. That is the most important part : you should not know about the last trial!
However, once you experience this type of trial, you now tend to memorize color intentionally because you might think that this type of trial will come again. So to speak, you are not naive any more. Thus, I can only collect single trial data from a participant. This is quite problematic in terms of data collection. What shall I do now?
Amazon Mechanical Turk
Aha! We can crowdsource our data at Amazon Mechanical Turk (mturk). For those who are not familiar with mturk, it is a website where people (“requesters”) can upload online surveys or experiments to collect data from a large group of anonymous people (“workers”) by providing monetary reward. It is fast and easy way to collect a huge amount of data in a short time. So I decided to upload online experiments to solve my problem.
I used an experimental paradigm called delayed estimation. Here, I briefly show an oriented ellipse with a color to a participant, and ask him/her to recreate its orientation (or color) after a delay. To test irrelevant color memory, I had a following design:
Access to your irrelevant memory
First, we tested whether participants were doing their job well. In the figure below, the top row shows the errors from the first 30 trials. This is an error, the difference from participant’s response and actual stimulus value. So, zero means they are identical. In both orientation and color, we have a very high peak around zero, which means online participants’ performance is pretty decent.
What is interesting is in the bottom row. If a participant did not have an access to his/her irrelevant memory, at the last (“surprise”) trial, he/she must have made random guesses. In this case, the error distribution will be flat (or uniform). For the irrelevant orientation memory (bottom left), the distribution was significantly different from uniform. It means participants were able to access their irrelevant orientation memory. For the irrelevant color memory (bottom right), the results were better. Conclusively, this shows that we are able to access irrelevant memory, meaning that irrelevant information (from orientation-color objects) can be automatically encoded in our short-term memory.
There’s just something about color
Why is the irrelevant color memory much better than the orientation one? What is it that makes color memory much easier to recall compared to orientation? One speculation is that color information in nature might be highly related to the identity of an object such as a predator or a poisonous prey. Also, it is possible that neurons processing color information have a slower decay rate than orientation neurons.