Revolutionary Brain Cells on a Chip Master Doom in Just One Week

A cluster of human brain cells has been demonstrated to play the classic game Doom. While the performance doesn’t yet match human ability, experts believe this breakthrough gets us closer to practical applications for biological computers, such as controlling robotic arms.

In 2021, researchers from the Cortical Research Institute employed a computer chip featuring neurons known as Pon. The chip, comprising over 800,000 living brain cells on a microelectrode array, was capable of both sending and receiving electrical signals. The researchers meticulously trained the chip to manipulate the paddles on the screen’s edges.

<p>Recently, Cortical Labs introduced an easier interface to program these chips using the widely-used programming language Python. Independent developer <a href="https://www.linkedin.com/in/sean-cole-8985a4207/">Sean Cole</a> utilized this interface to teach the chip how to play <em>Doom</em> in just about a week.</p>
<p>“Unlike the <em>Pon</em> project that involved years of rigorous scientific labor, this new demonstration was achieved in mere days by individuals with limited prior experience in biology,” said <a href="https://scholar.google.com/citations?user=bvWRHNcAAAAJ&amp;hl=en">Brett Kagan</a> from the Cortical Institute. “This accessibility and flexibility is incredibly exciting.”</p>
<p>The neuron-based computer chips utilized approximately a quarter of the neurons found in traditional chips. While the <em>Pon</em> demonstration yielded better results in <em>Doom</em> than random input from players, its performance still lagged behind that of top human gamers. However, it can learn significantly faster than conventional silicon-based machine learning systems, and new learning algorithms are expected to enhance its performance, according to Kagan.</p>
<section>
</section>
<span class="js-content-prompt-opportunity" />
<p>Comparing these biological chips to the human brain can be misleading, he suggests. "While it is indeed living tissue, the mechanisms it employs for information processing are dissimilar to those of silicon," he explains.</p>
<p><em>Doom</em> poses a substantial challenge compared to prior example games, and the ability to successfully engage with it marks a significant advancement in controlling and training living neural systems, states <a href="https://people.uwe.ac.uk/Person/AndrewAdamatzky">Andrew Adamatzky</a> from the University of the West of England, Bristol, UK.</p>
<p>Researchers like <a href="https://scholar.google.com/citations?user=jLnsiBEAAAAJ&amp;hl=en">Steve Farber</a> from the University of Manchester concur, noting that the ability to play <em>Doom</em> represents significant progress. He also pointed out that many unanswered questions remain regarding how neurons comprehend gameplay expectations and how they interface with a screen without visual organs.</p>

<p>Regardless, the leap in capabilities is promising. <a href="https://www.reading.ac.uk/biomedical-engineering/staff/yoshikatsu-hayashi">Yoshikatsu Hayashi</a> from the University of Reading is working towards practical applications like using biological computers to control robotic arms. His team is experimenting with a similar computer made of jelly-like hydrogel. “[Playing <em>Doom</em>] serves as a simpler analogy for controlling an entire arm,” Hayashi articulates.</p>
<p>“The significance here goes beyond just biological systems playing <em>Doom</em>,” adds Adamatzky. “It demonstrates the potential to navigate complexities, uncertainties, and real-time decision-making—skills essential for future biological or hybrid computing solutions.”</p>

<section class="ArticleTopics" data-component-name="article-topics">
    <p class="ArticleTopics__Heading">Topics:</p>
</section>