Podcast #9: AI mind reading – how close are we to AI reading our minds?
Today’s podcast episode On the Wave of Code is on the border between reality and science fiction. Is mind reading a reality or just a technological utopia? Michal Bystrický, an expert in data science, has once again accepted our invitation to join us in the studio and will tell us what is really possible today and what remains only in the world of fantasy. Listen to him or read his abridged transcript.
In Slovakia, this area has not yet been treated in detail, but at the suggestion of our listeners, we decided to open up this – partly controversial – topic, which is provoking increasingly intense discussions in the world.
How close are we to being able to decipher human thoughts? What technologies are being used today – EEG, MRI or others? Can AI mind reading help in medicine or education? Or can it degenerate into a tool of manipulation? And where is the limit of our privacy?
Michael, let’s look at mind reading from different angles. Nowadays, technology and research in neuroscience and artificial intelligence have reached a level where they are able to recognize certain patterns in brain waves that can indicate a person’s thoughts or intentions. Nevertheless, there are still many questions as to whether direct and complete access to thoughts can be achieved. Michael, is this really just a myth, or are we so far along that it is actually possible?
I think of mind reading as the ability of a system, such as a computer or artificial intelligence, to interpret the signals we measure. These signals correlate with the thought processes of individuals. They are time series that show when and through which neuron an electrical impulse is passing. Artificial intelligence can recognise these signals and assign meaning to them in a computer.
Just imagine you’re thinking about a car. And 300 times we record brain activity and 700 times we don’t. It creates a pattern. This pattern is then “car” and the artificial intelligence can learn it. However, it should be stressed that this is not a literal mind-reading. Rather, it is about identifying patterns – for example, reactions to stimuli or intentions, such as the movement of a limb. That’s when a particular part of the brain is activated.
So, to summarize – the computer monitors brain activity. If I repeatedly think about the car, the brain always reacts the same way. The computer remembers this pattern and can recognize it. But that doesn’t mean it can read what I’m thinking specifically, for example, that I want to drive a car. Rather, it knows that that thought is related to a particular object. That is, it is pattern recognition in the brain, not direct thought reading.
Yes, exactly.
On the subject of mind reading, not only is there a lot of different information being spread, but unfortunately also misinformation. I can think of a moment during the pandemic when there were heated discussions in Slovakia about the introduction of microchips into the brain using test sticks. I am sure you remember that, too – it was during the pandemic. Can we talk a little more about that? Would something like that even be technically possible?
Yes, there was once a hoax that the test sticks used to collect samples for COVID-19 diagnosis contained microchips. These microchips were supposedly to be inserted into the body, specifically the brain, through the nose, and then used – by who knows who – to control thoughts or monitor actions. It has to be said, however, that no mind control or Internet access via such a thing is technically possible yet.
Nowadays, we have various technologies such as electroencephalography (EEG) or fMRI – functional magnetic resonance imaging – that allow us to monitor brain activity. How do these technologies work? On what principle are they based? And which technologies do you consider the most important when trying to decode thoughts? What makes them unique?
If we are talking about mind reading, there are three basic methods. Measuring the electrical activity of the brain – for example, when we think, do an activity, it produces electricity. The second is measuring the oxygenation of the brain – for example, with a magnet or infrared light. So, when we think, we have different amounts of oxygen in different areas. And the third is the use of X-rays.
In electrical activity EEG, for example, which we measure using electrodes placed on the scalp. These pick up the electrical potentials, i.e. the voltage differences on the surface of the head, which are generated by the activity of neurons. The brain is still processing information, for example, regulating bodily functions and so on. Interestingly, artificial intelligence is being programmed in a similar way today. It actually works like a loop. An agent in artificial intelligence processes information and makes decisions within this loop. In the same way, the brain is continuously processing information. Electrodes on the skin can pick up the electrical signals that are generated when neurons communicate with each other. They pick up the electrical changes that occur when large groups of neurons are activated.
This electrical activity can be divided into waves – delta, alpha, beta, gamma. That is, when I’m in deep sleep, when I’m relaxed, when I’m in an active state of mind, or when I’m in a high state of activity.
The BCI, Brain-Computer Interface, is also based on electricity. An example is Neuralink, which was founded in 2016. It uses microscopic electrodes that are implanted directly into the brain. It’s then much more accurate. In this group there is also the MEG, which is a rather expensive device that measures the magnetic fields generated by the electrical activity of neurons.
Then we have another group – technologies based on brain oxygenation. This includes, for example, fMRI. This is a large magnet that uses hydrogen atoms in water (or therefore in the brain) that are exposed to a magnetic field. These atoms react and give off weak signals. The idea is that the active parts of the brain consume more oxygen, so we can see which areas are more active. Another method in this category is NIRS. This method uses infrared light to measure oxygenation. The light penetrates through the tissues and reflects differently according to the oxygen level.
The last method is by X-ray, i.e. CT scan or computed tomography. These are X-rays, with the computer producing cross-sectional images from multiple angles. The idea is that bones absorb more X-rays than soft tissue, so they appear whiter in the resulting image. By taking images from different angles, the image can be reconstructed into 2D sections.
I’ll summarize. EEG measures the electrical activity of the brain via electrodes on the head. Some people have already encountered this test in practice – it is the monitoring of brain activity, for example in the diagnosis of neurological conditions such as epilepsy. During the examination, we are given a special cap that is placed on the head and brain activity is picked up using electrodes. It’s quick, but doesn’t show details of the deeper parts of the brain. Do I understand this correctly?
Exactly.
Then you mentioned BCI, which allows direct communication between the brain and the computer. It uses EEG to decode signals from the brain. How can I imagine that? What is the difference between EEG and BCI?
Talking about the difference between EEG and BCI, under BCI I mentioned specifically Neuralink. In this project, the electrodes are directly implanted in the brain, which means that surgery has to take place. So the accuracy is much higher than when we just take signals from the skin. EEG captures overall electrical activity, whereas Neuralink captures extremely fine details, that is, individual neurons or small groups.
And then there’s MRI, which measures blood flow in the brain associated with brain activity. It provides high spatial resolution, but on the other hand, MRI is slow and delayed compared to EEG.
Yeah.
And then we have a CT scan. It uses X-rays to create images of the brain. It’s fast and gives a clear view of anatomical problems like fractures or bleeding. But on the other hand, it doesn’t show brain function. Hence, each technology has its pluses and minuses. And why did we stop there? It’s to get to the heart of the subject. Are some of these technologies being used today to map human thoughts?
We’ve already mentioned that we have technologies that can capture brain activity. So we have some signals. Imagine, for example, an EEG recording from one person. We have him do different activities, for example, when he sits still or when he moves his arm.
We will record and mark these sections. When the hand moves, some activity occurs and we mark it as movement. When a person does nothing, we mark the stretch as stillness. Using these labeled segments, we can train artificial intelligence, specifically neural networks.
The artificial intelligence then goes through the EEG recording and learns to discriminate these activities. It can identify what is hand movement and what is stillness. Later, it can identify sections associated with hand movement quite accurately on new, unlabelled recordings based on a pattern.
Should we be worried that mind-reading will one day invade our privacy?
The closest thing to what you mentioned is just Neuralink, founded in 2016. It uses microscopic electrodes that are inserted directly into the brain. This means the patient has to undergo surgery. The first human clinical trials are already underway as of 2023.
The first patient was a man with quadriplegia, a dysfunction or loss of motor and sensory function in the cervical spinal cord. He was implanted with a chip that enabled him to move a cursor on a screen, for example. This allowed him to control music, play games like chess.
The study also included another patient who could create 3D designs using CAT Fusion 360 software. So we’re already at the level where human clinical trials are underway.
So in this way it could help people with physical disabilities to communicate better.
Yes, especially in the medical field, this technology can help a lot.
Apart from medicine, where could mind-reading be of most benefit? The field of education comes to mind. Will we be able to learn languages more easily, for example? Or could it help in the area of security?
This topic is very broad. The biggest contribution I see is in medicine, as I have already mentioned. For example, the restoration of motor function – the patient who was able to move the cursor on the screen using only his thoughts. Next, prosthetic control – the brain interface may allow bionic limbs to be controlled with greater precision and speed. This is a more precise technology.
In the treatment of neurodegenerative diseases such as Parkinson’s disease, the implant can help control tremors or improve motor coordination. For Alzheimer’s disease, such technologies could help boost memory processes or recover lost memories. Perhaps in the future, Neuralink could store important information directly in the brain or assist in its recall.
In general, as you mentioned, mind control of devices could have a multifaceted impact on all aspects of our lives and society.
When we were explaining the EEG, I also mentioned epilepsy. I had read that in epilepsy there is excessive activation of neurons, which can lead to a seizure. In artificial intelligence, again, it can happen that a certain part of the system gets too much signal and the output is nonsensical or extreme. Do you see a similarity between this overload in the brain in epilepsy and in artificial intelligence systems?
Yes, I do see some similarities in behaviour in epilepsy and in AI. I will mention two examples. In AI, when you intentionally or accidentally modify inputs, it can lead to nonsensical or extreme outputs. In AI, this is called adversarial attacks. Such attacks can disorient the system. Similar to how abnormal electrical activity can disorient the brain during an epileptic seizure.
Second example: in AI, it may happen that a certain layer or neural pathway receives too much signal or activation. This can lead to the output being unnaturally extreme or beyond what we expect. This phenomenon is conceptually similar to excessive neural activity during a seizure.
Could mind reading be a tool for manipulation? How can we prevent this?
So far we have talked about mind reading. That is, we read what we measure and interpret information from that. But we haven’t yet talked about notation. Right off the bat, it should be said that so far this is only about stimulation. It is not possible to write down an actual thought or concrete information.
So what is stimulation? The aim is to affect neurons in a specific area, which can then be used to restore functions that are impaired due to, for example, neurological disorders. This is used to improve motor skills, to stop seizures or to aid memory and learning.
There are already technologies that are being used for this. For example, Deep Brain Stimulation (DBS), i.e. deep brain stimulators. These are medical devices that deliver electrical stimulation to certain parts of the brain.
As for stimulation in AI, there we can amplify scales that are weak. This means that a certain part of the neural network has a higher probability of being activated. A simple way to think of it is that if we stimulate a certain part of a branch of the neural network that is related to reinforcement, that part will be more active and “talk about it more often”.
It is interesting to mention how these deep brain stimulators are implemented. This uses techniques such as MRI or CT scans, which precisely identify the target area related to a particular disorder or symptoms.
Small electrodes are surgically implanted in specific areas of the brain, which are selected by MRI or CT. Then there is a generator that sends controlled electrical impulses through the electrodes to the targeted areas of the brain.
These impulses change or modulate abnormal electrical patterns, which can relieve symptoms such as tremors or stiffness. Specifically, these are the manifestations of Parkinson’s disease. Then the doctor and the patient work together to set the right parameters for the stimulation – intensity, frequency and duration.
I’ll still mention the AI parallel. We can measure which parts of virtual neurons are activated and amplify their weights, that is, stimulate them.
I’ll summarize. It is impossible to write down specific information because the brain is extremely complex and we don’t fully understand how it works. Transmitting specific data would require an accurate understanding of how neurons encode different types of information. And that’s where machine learning comes in to help decode at least the patterns.
Another problem is that if something were to be written down, it would have to be done in such a way that the neurons would correctly interpret the new signals without disrupting existing processes. So we don’t know how to write down specific information yet. The benefit, as has already been said, is more of a medical nature.
Do you think it would be possible to manipulate our thoughts? Because I already sometimes feel like someone is reading my mind – for example, when I want to buy a ski hat and suddenly all I see all winter is ski hat ads everywhere. So someone might be reading my mind, but no one has bought me one yet.
About that ski cap – sometimes it can seem like the current systems are already reading our minds. The way it works is that if you’re talking about ski caps with a friend, for example, and you’re also active on different devices such as a mobile or a computer, algorithms can connect you through information such as location, IP address or network. They’ll then show you ads for ski caps even if you weren’t looking for them before – it’s just that you’ve been talking about it.
If we were to speculate why mind reading might be possible in the future, perhaps it would be related to neuroplasticity. That is, the brain has the ability to adapt and reorganize its structures and functions in response to new experiences, learning, or damage. Perhaps one day the brain will be able to learn to receive new external information, such as from Neuralink.
A good example is playing a musical instrument. When we play when we are young, we see physical changes in the brain, i.e. more activity in specific areas, improved connectivity and lasting changes.
“Sometimes it can feel like the systems are reading our minds – for example, when we see ads everywhere after a conversation about a ski cap. It’s not magic, but a combination of data, devices and algorithms.”
Do you think we should use mind reading to expose lies? Wouldn’t that be too dangerous?
I will give my opinion. I think we should never allow anyone to read our particular thoughts. They could be misinterpreted or misused, which would violate privacy. When we talk about controlling a computer by the subconscious mind, that’s different. But when reading specific thoughts, there is the problem of mistranslating those patterns into reality.
With mind-reading technologies also come various ethical dilemmas, particularly with regard to privacy. If these technologies are used without consent or proper oversight, we could run into problems. What precautions would you recommend to be taken to prevent the possible misuse of these capabilities?
Again, this is a big topic, so I would like to mention here just some of the important aspects. It is mainly about privacy. We should ensure that our thoughts remain private, and that is very important. Any intervention in the brain should be voluntary and informed, that is to say, the user should give consent for such intervention.
Then you have to ask, what happens if someone changes your thoughts or memories? How will this affect your identity and personal autonomy? These are questions of identity, integrity and free will. Finally, there is the risk of misuse of these technologies, for example, for unethical purposes, manipulation or mind control.
Do you think these technologies will become a normal part of our lives or will they remain only for a small group of professionals?
Let me use an example from history. When cars began to appear in the early 20th century, many were sceptical. Cities were full of horse-drawn carriages leaving litter in their wake. Despite the initial mistrust, the practicality and efficiency of cars overcame this scepticism.
Similarly, today we have blockchain technology. Despite the initial difficulties, they are already changing the way we store value (e.g. bitcoin) or trust, as they enable transparent execution of contracts without intermediaries. The practical benefits may in time ensure that blockchain will be as much a part of our lives as cars.
When we talk about Neuralink and similar technologies, they are still at the very beginning of their development – just like cars in their early days. Neuralink is currently the subject of discussion and research, and it is possible that it will eventually become integrated into our daily lives. But that will take some time, in my opinion.
Michal, how would you sum up today’s topic in a nutshell?
I see the theme in two main levels – reading and writing information. In terms of reading, we currently have only very limited ability to measure and interpret patterns of what is going on in the brain. Writing is not currently possible – we can only stimulate certain areas. So these technologies do not allow for literal mind reading, let alone writing. The benefit so far is mainly medical.
Very close to this topic is artificial intelligence, specifically neural networks. There are still many unknowns there, but it is artificial neural networks that can help us better understand how the real brain works. The difference is that we can easily do experiments on an artificial neural network – for example, activate or deactivate certain areas – and observe how the system behaves. With a human, of course, it’s much harder. If anyone is interested in this topic, I recommend you start researching artificial neural networks.
Dear listeners, thank you for listening to today’s episode of On the Wave of Code, brought to you by msg life Slovakia. I say goodbye to my guest Michal Bystricky, who shares his dreams on the fascinating topic of AI and mind reading. My name is Ivana Hricova and I look forward to seeing you again together on new topics.
- The best IT podcasts: both interesting and entertaining. Which podcast is worth listening to?
- Java vs C# – comparison of programming languages (2024)
- Book Review of Head First Java (3rd Edition)
- The latest IT news from the tech world
- Is AI really intelligent? What is artificial intelligence, types, trends, risks of AI
- Generations X, Y, Z and Baby boomers in the workplace: how to work with them?
