Artificial intelligence to understand natural intelligence?

The unprecedented pace of technological advancements over the last decades has transformed our understanding of the human brain. From being a “black box,” we now have delved into the workings of it. We understand how we remember things, how we make decisions, or how we solve problems, the so-called cognition. However, being one of the most complex systems in general and the most complex organ of the human body, there is no doubt that we still have a long way until we fully unravel its mechanics. We do not even know how long it will take and if it will ever be possible. However, progress is being made, and this journey is yet one of the most fascinating.

To understand how the human brain performs a wide variety of tasks, experts from a number of fields including neuroscience, psychology, mathematics, computer science and engineering, have joined efforts in order to develop powerful new tools. One of the most innovative “tools” that is currently widely used to answer many of the burning questions in neuroscience is Artificial Intelligence (AI).

What is AI?

Most probably, hearing about AI brings to mind applications such as Siri, Alexa, Google, all of the AI applications used to facilitate our lives. Even “google translate” that you have probably used at least once in your life or digital assistants are due to AI. However, AI application goes beyond that – it has been successfully used in a far wider range of applications including medical diagnosis and cognitive neuroscience.

Loosely defined, AI is for a machine what natural intelligence is for humans and (other) animals. We, as humans, use our minds to perceive information, analyze them, and take action. A machine can mimic natural intelligence if given an unambiguous set of instructions (a so-called algorithm), indicating how these functions should be performed. As defined in a recent paper by Kaplan and Haenlein:

“AI is a system’s ability to correctly interpret external data, to learn from such data, and to use those learnings to achieve specific goals and tasks through flexible adaptation.“

How has AI helped us to better understand human cognition? What advantages does this tool offer when compared to previous methods?

One of the main ways researchers try to understand how our brains perform different tasks is with the use of imaging techniques. In particular, a person is asked to perform a specific task, such as remembering a sequence of numbers or maintaining attention to a specific object. Meanwhile, a researcher images the brain or records its activity. These imaging tools have long been around; however, they produce a large amount of information for which the analysis and interpretation was very challenging, as there were no optimal computational and statistical methods.

The major advantage of AI is that it allows us to deal with extremely large datasets. One of the imaging techniques that has been traditionally used to measure brain activity is the electroencephalogram (EEG). This technique has a very high temporal resolution –we can measure brain activity on the order of milliseconds(!). This has been a great advancement in neuroscience, however, it led to very large datasets that were hard to handle with traditional tools. Thus, to draw conclusions from EEG, researchers used average techniques, due to which some information could be lost. A novel machine learning algorithm introduced by Anderson et al. overcomes this limitation – it allows us to model all of the data, from all the trials, without the need of averaging.

Another important feature of AI is its continuous improvement – you can improve your models by just providing them with more data. The idea behind this statement can be better understood when compared to human learning processes. Imagine that a five-year-old kid plays around a hot stove and suddenly gets burned. From that time on, the kid learns that this action leads to pain, and it will probably or avoid to do it in the future or be more careful when playing around the stove. Now imagine that a first-year elementary school student is doing her/his homework, which results in a high grade from the teacher. The kid learns that this action will maximize its chances to successfully achieve its goals. A model can be similarly improved by repeating strategies that worked well in the past and abandoning those that did not work.

Another important advantage of AI methods is the speed of processing. It means that it can quickly process a large amount of data, something that was not possible before.

Here, it should be mentioned that this list of AI applications in neuroscience is not exhaustive and it only refers to a type of AI, called machine learning. The field of AI is broad, and so are the applications of AI used to facilitate research of the human brain.

Another important application of AI, that we will not go into detail in this post, is modelling of the human brain. The main idea of this application is to mimic neurons, the working units of the brain, with artificial neural networks.

Challenges and limitations

AI is a technology that continually appears with new advancements. However, there are challenges and limitations associated with the application of this technology, as well as ethical considerations associated with its rapid growth. Among others, a major challenge is that one needs a massive volume of data to build an AI system. In medical research, this is usually not easy both time- and cost-wise. Another challenge is interpretation. Although it produces exciting results, sometimes it might be challenging to link them to a particular cognitive interpretation.

As expected, ethical considerations are also crucial. This field is still in its infancy. Currently, there is no legislation blocking how AI is being used and for which reasons. We should consider how the data are being used but also which outcome is being produced. Having an unbiased product, diversity-focused with an impact on humanity, is unbeatable!

Despite the challenges and limitations, this technology seems to be highly promising in addressing crucial questions on how our minds work. Importantly, developing the tools to understand the human brain is not only curiosity-driven science. In almost all major psychiatric disorders cognitive processing seems to be severely impaired. Understanding how the healthy and diseased human brain processes information will be a milestone for medicine. It will allow us to design effective treatments for several disabling disorders associated with neurodevelopment, neurodegeneration, and mental health.

I would like to thank Hermine Berberyan and Stella Tsoutsouri, both working in the field of AI, for their input in this post of SciFact.