New methods of artificial intelligence science are helping researchers to understand more about how the brain works – and in some cases, these scientists can actually intervene and push the brain to work differently.
If it sounds complicated, that's because it is. A Wired story introducing a University of Pennsylvania research project starts out by pointing out that the human brain is largely an unknown “black box” to scientists, and that there are significant barriers to affecting brain activity.
However, UPenn psychologist Michael Kahana and a team of scientists were able to utilize electrodes going into the brains of 25 epilepsy patients in order to start learning about how the brain works during memory.
It's significant that the team was able to do this by “piggybacking” on already existing infrastructure. (From the wording, it’s assumed that the group was able to use subjects who were already hooked up for more prosaic medical reasons.) As the article points out, it's pretty difficult to get buy-in from research subjects to put invasive technology into the brain.
The researchers started out by simply reading brain activity – specifically, in precise calculation of electrical activity inside the brain while people were in the process of learning and memorizing words.
After doing this for a while, and building up a substantial training set, the researchers were able to predict certain kinds of learning.
After the foundational research, scientists were eventually able to send electrical stimulation to the brain to assist in the process of memory.
When you talk about the use of electrical stimulation to help with memory, it sounds simple – but when you look more closely, everything is predicated on very high-tech methodologies and quite a lot of guesswork.
Without the initial machine learning that identified memory activity, the scientists wouldn't have had as good of an idea of how to electrically stimulate brains to promote good memory function.
Also, it's clear from reading about the study that the team doesn't know how the electrical stimulation is working – they just know that it is. In other words, the scientists are using the results of machine learning to fine-tune the system, without actually understanding the ins and outs of the brain function itself.
This intriguing example is perhaps one of the best examples of “hands-on” machine learning – here, the data is not just put into training sets to model more data. Here, the training set actually acts as a catalyst for specific experiments in bioinformatics, and the results are based on the calculations that machine learning programs made. It's a very interesting look into the synergy between artificial intelligence and our own human biological brains, and how the two are intersecting as we make rapid progress toward the “singularity” of Ray Kurzweil and other future outcomes.