Hon Weng Chong, MD (Founder & CEO) and Brett Kagan, PhD (Chief Scientific Officer) from Cortical Labs share their insights on power-efficient biological computing using lab-grown neurons and the future of synthetic cyborg brains.
What is biological computing?
Traditional computers utilize logic gates which are deterministic and great for accurate numerical calculations. Biological computing uses neurons and as such, would be as inefficient as humans at large mathematical calculations. However, neurons have been selected for effective operation in a noisy and often unpredictable world by natural selection. Biological systems need the ability to predict what will happen to them in a dynamically changing environment and modify their behaviour to adapt in the fastest and most calorie-efficient way. In other words, the result of this ability to model the world and react to its changes is an observable form of intelligence. This is a shift forward in how we think about computational processing. The future of computation may integrate a BPU alongside a CPU and GPU.
Where do you extract the cells to grow neurons?
The neurons we use are from volunteer donor skin or blood cells. Using stem cell technology, we induce the formation of stem cells to grow neurons. When one of these cell lines are made, they are infinitely renewable meaning the cells can keep on dividing to provide us with plenty of stem cells to work on.
Would you say these neurons are conscious?
Currently, there is no evidence showing that these neurons are conscious. We describe them as sentient because they respond to the electrical impulses. Research has shown that neurologically intelligent action can be achieved without a conscious experience of it.
How do you interpret neuron behaviour in the virtual game environment?
The neurons are embedded on a computer chip with 26400 electrodes. Some parts of the chip are assigned sensory regions while others are motor regions. Information about the virtual game environment is fed into the sensory region such as where the ball is in relation to the paddle. The neurons take an egocentric view of the world, essentially receiving information from the perspective of the paddle. The location of the ball whether it is to the left, centre or right of the paddle and how far it is at is delivered to this sensory region as tiny jolts of electricity.
Electrical impulses produced by neurons that are in the motor regions are monitored after the sensory region information is delivered. If more activity is present in one motor function such as moving the paddle up in comparison to moving down, then the paddle is moved up in the virtual game environment.
You can see this on the visualiser where every time a neuron produces an electric impulse, we can decode the activity in real-time.
How do you teach neurons goal-oriented tasks?
Cortical Labs is exploring the physical underpinnings of innate intelligence using electrical information. The Free Energy Principle suggests that neurons try to predict the external world and they try to minimize the error between what they see and what they think. This means that neurons can either change their behaviour so that the results are predictable or get better at predicting the world around them. By providing an unpredictable stimulation pattern when neurons do not behave as we desire, this removes the ability to get better at predicting, which means the neurons need to get better at self-organizing to adapt to the environment.
During training, the neurons are sent a noisy and unpredictable electrical impulse for incorrect behaviour such as when missing the pong ball in the game. There is also a certain degree of randomness and unpredictability when the ball restarts. After some time, the neurons adapt to move the paddle to deflect the ball demonstrating the system wide co-ordination and intelligent network ability of the system.
Do the neurons from different donors behave differently?
This is something that we are looking into for applications such as personalized medicine to probe whether donor cells have prior knowledge. There is compelling early evidence to suggest that there will be traits inherited from the donor and it is certainly interesting to explore how much of who we are and what we think are already encoded in our neurons.
Do the neurons remember how to play the game?
We do not have built-in memory centres in the cells and cortical cells are not optimized for memory. However, we are exploring multiple cell types. Theoretically it is possible to use specialised cells known to be critical in memory to allow the cultures to have better memory of past experiences.
It is the Year 3000, what is Cortical Labs working on?
Imaging what could be in the Year 3000, Cortical Labs is the foundational technology that is responsible for millions of intelligent metaverse-native lifeforms. These lifeforms can interact with regular people, both in the physical world and the metaverse, performing tasks that these systems can be specialised for; even innovating and evolving in their abilities. The development of these lifeforms can provide numerous ethical and environmental benefits, allowing menial tasks that are too dynamic to be entrusted to machine learning but too dangerous or trivial to continue to be done by humans to be done by Synthetic Biological Intelligence (SBI) which also have lower power consumption than traditional computing. SBIs have allowed the capacity to better treat diseases in a personalised medicine manner, streamlining neurological healthcare, along with providing us a better understanding of intelligence, ultimately leading to a healthier and smarter society. SBIs have led to a synergistic advancement with the human race, allowing people to live up to their full potential.