In vitro neurons learn and exhibit sentience when embodied in a simulated game-world

Area of Science:

• Neuroscience
• Artificial Intelligence
• Computational Biology

Background:

• Integrating biological neurons with digital systems offers potential for enhanced computational performance beyond silicon-based limits.
• The theory of active inference and the free energy principle provide a framework for understanding goal-directed behavior in biological and artificial systems.

Purpose of the Study:

• To develop a system, DishBrain, that integrates in vitro neural networks with digital computing to harness neuronal adaptive computation.
• To investigate the learning capabilities of biological neural networks when embedded in a simulated game environment.

Main Methods:

• Human or rodent neural cultures were integrated with in silico computing using a high-density multielectrode array.
• Neural networks were stimulated and recorded from while playing a simulated version of the arcade game "Pong."
• The study applied principles of active inference and the free energy principle to analyze neural activity and learning.

Main Results:

• Apparent learning was observed in neural cultures within five minutes of real-time gameplay, a phenomenon not seen in control conditions.
• Closed-loop structured feedback was crucial for eliciting sustained learning over time.
• Neural cultures demonstrated self-organization of activity in a goal-directed manner in response to sparse sensory feedback, termed synthetic biological intelligence.

Conclusions:

• DishBrain successfully demonstrates the rapid learning capacity of biological neural networks integrated with digital systems.
• The findings highlight the importance of structured feedback for goal-directed behavior and learning in biological neural networks.
• This research opens avenues for exploring synthetic biological intelligence and its potential applications in understanding intelligence.