How AI can teach the brain to remember more

Artificial intelligence has seen dramatic advances in recent years, advancing from a reliably clever tool for observation to a keen observer in its own right. We’ve seen AIs compete with and learn from human beings and, in many cases, surpass them. We’ve even seen examples of an AI with a neural network capable of teaching itself and developing beyond its initial capabilities. Now, researchers at the University of Pennsylvania have found a way that an AI may be able to stimulate the human brain and enhance the ability to learn and remember.

According to Wired, Michael Kahana, a professor of Psychology at the University of Pennsylvania, and his team have demonstrated that machine learning algorithms can not only understand, but actually enhance the brain’s ability to create new memories. Because the human brain is incredibly complex, and also because we can’t directly observe its functions, it’s hard for researchers to determine the exact processes which determine the difference between remembering and forgetting certain pieces of information. However, an AI programmed to monitor and interpret brain signals during certain tasks can recognise the difference between learning and forgetting.

The researchers worked with 25 epilepsy patients who had already been fitted with over 100 electrodes fitted to their brains. These electrodes were originally implanted for the purpose of monitoring the electrical signals associated with seizures, but they were also well suited for Kahana’s needs. The patients were asked to memorise lists of words and the electrical signals around their brains’ memory centres were measured by the electrodes. Later, the patients were asked to repeat the words and the different signals related to which words they forgot and which ones they could remember were established.

After repeating the process three times with each patient, it was possible to create individual profiles for each one. A different algorithm was produced for each patient based on their electrical activity to predict what information they would be able to remember and which information they were likely to forget. Once this was established, the algorithm could be adapted to actually provide a stimulating charge to the brain when it seemed likely that the patient was going to forget something. On average, the patients were able to recall information 15 percent better with the memory-stimulating algorithm in effect.

The algorithm was able to recognise when something was being remembered or forgotten and provide appropriate stimulation to prevent the brain from forgetting.

With machine learning, the algorithm could refine itself and improve its own ability to recognise and prevent loss of information in the human brain. In effect, the electrodes could see when a person is about to forget something and give them the stimulation needed to recall it. These charges are apparently more beneficial when the brain is in different states. For example, if the brain is focused on trying to remember something, then the external stimulation can actually have a negative effect on the process. It is apparently more helpful if the memory centres are in a restful state when the electrodes are activated in this way. As such, it might be possible for people to memorise something they studied earlier while their brains are later distracted by another activity. A neural network could potentially identify the best times to stimulate the memory centres. It could also develop more understanding the more that it works and, if provided with a greater range of precise electrodes, its effectiveness could be dramatic.

People may not be in much of a hurry to hand their memories over to an artificial intelligence. It’s certainly easy to imagine some of the potential downsides to letting a computer programme decide what you remember or forget and it’s certainly easy to understand anyone’s fear in letting someone tamper with their minds. However, for people suffering from existing conditions which affect memory, this could be a chance like no other, to be able to hold on to new memories which could otherwise be forgotten.

If this technology is further developed, it’s not out of the question to think that other parts of the brain could similarly be trained or even enhanced similarly. Perhaps a neural network could be created to enhance someone’s cognitive reasoning or their physical coordination. The brain controls everything the body does and this may be an early step towards reprogramming the brain itself.

Ronan Daly

Ronan Daly is a staff writer for My Good Planet who specialises in Technology and Science. With a Masters Degree in English, and over a decade's experience as a teacher and writer, Ronan has brought a breezy, learned style to My Good Planet, making occasionally complex material accessible and understandable to all.