Researchers have found that non-living hydrogels can play the video game Pong and improve their gameplay as they gain experience, as reported on August 23 in the journal Cell Reports Physical Science. The team connected hydrogels to a virtual gaming setup and created a feedback loop that linked the hydrogel’s paddle — defined by the arrangement of charged particles in the hydrogel — to the ball’s position, which was determined through electrical stimulation. After practicing, the accuracy of the hydrogel improved by as much as 10%, leading to longer game rallies. The researchers indicate that this shows non-living materials can exhibit some form of “memory” to update their understanding of the environment; however, more studies are necessary before concluding that hydrogels can actually “learn.”
Vincent Strong, a robotics engineer from the University of Reading and the first author, states, “Ionic hydrogels can utilize memory mechanics similar to those found in more sophisticated neural networks. We demonstrated that not only can hydrogels play Pong, but they can also improve their performance over time.”
The motivation for this study came from earlier research indicating that brain cells in a laboratory setting can learn to play Pong when they receive electrical feedback on their performance.
Yoshikatsu Hayashi, a biomedical engineer and corresponding author from the University of Reading, mentions, “Our research explores whether simple artificial systems can execute closed-loop computations like the feedback loops that allow our brains to function. Both neurons and hydrogels operate on the principle that ionic movement and distribution can serve as a memory function, connected to sensory-motor loops in the Pong scenario. In neurons, ions move within cells, whereas in the gel, they move externally.”
Hydrogels are intricate polymers that become jelly-like upon hydration, with gelatin and agar being natural examples. In this study, the researchers employed an “electro-active polymer,” which means that the hydrogel reacts to electrical stimulation thanks to the presence of ions (charged particles) in its surrounding medium. When the hydrogel is stimulated, ions shift position, pulling water molecules with them, which causes the hydrogel to temporarily change shape.
Strong explains, “The speed at which the hydrogel loses water is significantly slower than how quickly it absorbs water, implying that the next movement of ions is influenced by their previous movement, akin to memory formation. The ongoing rearrangement of ions within the hydrogel builds from earlier arrangements, all the way back to when it was initially created with a uniform ion distribution.”
To determine if the hydrogel’s physical “memory” could allow it to play Pong, the researchers used electrodes to link the hydrogel to a virtual gaming environment and initiated the game with the ball launched in a random direction. They applied electrical stimulation to relay the ball’s position to the hydrogel and monitored the movement of ions within the hydrogel to gauge the location of its paddle.
As they played Pong, the researchers tracked the gel’s hit rate and analyzed whether its accuracy increased. They found that with increased practice, the hydrogel was able to hit the ball more consistently, resulting in longer rallies. While the neuron-based systems reached their peak ball-handling skills within around 10 minutes, the hydrogel required approximately 20 minutes to achieve its highest Pong performance.
Strong notes, “As playtime increases, the gel amasses a memory of all movements, allowing the paddle to adjust to the ball’s trajectory within the simulated environment. The ion movements encapsulate a chronological memory of actions, enhancing overall performance.”
The researchers suggest that since most current AI algorithms are based on neural networks, hydrogels offer a unique form of “intelligence” that could lead to the creation of new, simpler algorithms. Looking ahead, they aim to delve deeper into the hydrogel’s “memory” by investigating its underlying mechanisms and testing its capability to perform additional tasks.
Co-author William Holderbaum from the University of Reading adds, “We are considering how to extract the algorithm from the hydrogels that enables memory formation.”
Strong remarks, “We’ve established that memory emerges within the hydrogels, but the next phase is to verify whether we can demonstrate that learning is taking place.”
This research received support from Process Vision Ltd.
