A quiet revolution is unfolding in technology. In a Melbourne laboratory, scientists have grown 8,00,000 living neurons from embryonic mouse cells and human stem cells, cultured them into a thumbnail-sized network, and connected them to a computer. Within minutes, the neurons learned to play Pong, improving their performance by responding to electrical signals that conveyed the ball’s position and generating signals to move the paddle. The experiment, developed by Cortical Labs, led to what the company calls “synthetic biological intelligence” and, in 2025, the launch of what it describes as the world’s first commercially available biological computer. Scientists remain cautious about defining what is happening inside the dish. Cortical Labs’ chief scientific officer Brett Kagan describes it as “something that resembles intelligence,” not human thought or consciousness. Yet he argues that it marks the beginning of a new frontier in understanding intelligence—one known as Synthetic Intelligence.
SI is rapidly transforming healthcare by combining advanced computational systems with biological insights to improve diagnosis, treatment, and patient care. From analysing medical images and detecting diseases at earlier stages to predicting patient outcomes and accelerating drug discovery, these systems can process vast amounts of data far more quickly than humans. Hospitals are increasingly using intelligent algorithms to streamline administrative tasks, personalise treatment plans, and support clinical decision-making. Emerging forms of synthetic intelligence that integrate biological and machine-based systems promise even greater advances, potentially enabling more precise therapies, real-time health monitoring, and deeper understanding of complex diseases. As healthcare becomes increasingly data-driven, SI is poised to become an indispensable partner in medicine, augmenting rather than replacing human expertise.
To understand what is at stake for humanity from AI, you have to go back to 1946. The SI revolution began much before Melbourne, at the University of Pennsylvania. Here, the world’s first general-purpose digital computer, ENIAC, which occupied 1,800 sq ft of floor space flickered to life. It could do the math: it could make approximately 5,000 additions per second. It did exactly what it was told and nothing else. The journey from ENIAC to ChatGPT is, by any measure, one of the great stories of human ingenuity. At every stage, the underlying principle of the experiment remained the same: a computer manipulates data according to software instruction, and sufficiently sophisticated software produce something that looks, from the outside, uncannily like thought. But here is the uncanny thing. The ‘thought’ is not the same as what we know as thought. It cannot think like us. John McCarthy—the Stanford mathematician who coined the term “artificial intelligence” in 1956—defined the role of AI as building machines capable of performing tasks that normally require human intelligence. It is worth noting that the word “simulate” appears constantly in early AI literature although the algorithmic output of ChatGpt or Claude.ai doesn’t have the word “understand”. That omission, which once seemed merely technical, is now the fault line along which the entire field of Artifical Intelligence is cracking.
According to the International Energy Agency, training GPT-4 is estimated to have consumed over 50 gigawatt hours of electricity—the equivalent to about 0.1 per cent of New York City’s entire annual power supply. Mind you, this is for a single computer, trained just once. The IEA projects that data centre electricity consumption will more than double by 2030. American data centres drank 17 billion gallons of water for cooling its computers just in 2023. MIT researchers have noted that any AI training cluster consumes seven to eight times more energy than a conventional computing workload. In contrast, the human brain runs on just 20 watts: the power of a dim light bulb. Yet it performs computations that no data centre on earth can replicate.
In the 1980s, the philosopher who first gave “synthetic intelligence” its name was John Haugeland, a University of Pittsburgh thinker. His analogy was, interestingly, diamonds. A simulated diamond—for example cubic zirconia—looks like a diamond, catches the light like a diamond, but is not a diamond. But a synthetic diamond, grown in a laboratory through SI is structurally and chemically identical to any diamond excavated from a mine in Botswana. Its laboratory origin is irrelevant to its authenticity. Haugeland argued that the same logic must apply to intelligence. A system that merely simulates cognition which predicts the statistically plausible next word, and matches patterns in training data is the cubic zirconia of the mind. A system that genuinely reasons, adapts, and understands, regardless of whether it runs on carbon or silicon or cultured neurons, would be the synthetic diamond: real, despite, or maybe even because of its engineered origins. Computers hallucinate, confabulate, and even fail in ways that a reasonably intelligent five-year-old would not. Scale has made them more impressive but not less brittle. This is the contradiction scientists are trying to resolve.
The most technically advanced response to this problem is neuromorphic computing, whose recent progress is not widely spoken about but is quite extraordinary. The human brain does not separate memory from processing. Its neurons store information and compute simultaneously, firing in sparse, event-driven bursts. They consume almost no energy when there is nothing to report. Conventional chips do the opposite. They shuttle data constantly between processor and chip memory, while burning energy at every step. Neuromorphic chips are built on the brain’s logic. In October 2023, IBM reported that NorthPole, its latest neuromorphic chip, outperformed conventional intelligence architectures on multiple tasks at a fraction of the energy. Intel, not to be outdone, unveiled Hala Point in April 2024, which turned out to be the industry’s first neuromorphic system which was capable of simulating 1.15 billion neurons. All these were packed into a computer chassis the size of a microwave oven, and distributed across 1,40,544 processing cores. These gorged on a maximum of 2,600 watts. However, SI hardware is not yet ready for mass commercialisation as Intel’s own researchers say. But the trajectory of its progress is no longer theoretical.
The DishBrain experiment in Melbourne possesses what the neuromorphic chip announcements lack: strangeness. Researchers believe that SI involves what neuroscientist Karl Friston called the free-energy principle: biological neural networks are, at their most fundamental level, prediction machines built to minimise surprise. The DishBrain neurons, once embedded in the feedback loop of the Pong game, behaved exactly as the theory predicts. When the game did not give them feedback of whether they were succeeding or not, the scientists learned nothing. Cortical Labs plans to expand this research in disorienting directions. They are aiming to study how DishBrain plays when its neurons are exposed to alcohol and pharmaceutical compounds. The purpose is to validate biological computing platforms as tools to discover new drugs and personalise medicine. The chief executive, Hon Weng Chong, framed it plainly: “DishBrain offers a simpler approach to test how the brain works and gain insights into debilitating conditions such as epilepsy and dementia.” In 2025, the company moved beyond just research. The CL1, Cortical Labs’ first commercial product, and the world’s first biological computer is on the market. The possibility of what buyers can do with a biological computer is, for the moment, gloriously open to the imagination.
The history of tech proves that transition from one technological system to another is usually messy. The transistor was invented at Bell Labs in 1947 by John Bardeen, Walter Brattain, and William Shockley, all Nobel Prize winners. None of them could have imagined that their device would, within a decade, make vacuum tubes obsolete. Within three decades, the personal computer, the internet, and eventually the large language model generating text on your screen became real. Years after the transistor was invented, relay computers, and electronic computers coexisted in university basements and government facilities. We have arrived arguably, at a similar moment now. The EU’s AI Act, which entered into force in August 2024, regulates AI systems by risk category without distinguishing between artificial and synthetic approaches. This signifies a regulatory architecture which some researchers argue is already obsolete before being fully implemented.
The questions synthetic intelligence raises are not new. They are, in some ways, the oldest questions that mankind has wrestled with, primeval and powerful.
What is a mind?
What is understanding?
What is it to know rather than merely to process?
The Turing Test, devised by the great mathematician, computer scientist and cryptographer Alan Turing—who was instrumental in Britain’s victory over the Nazis—succinctly framed the question in 1950: “Can a machine imitate a human conversationalist well enough to fool a judge?” His was a pragmatic inquiry, designed to sidestep metaphysics. But it also encoded an assumption that still governs the field for 75 years; that human cognition is the benchmark, and how machine intelligence comes close to human is the measure. Synthetic intelligence stands this assumption on its head. The question is no longer whether a machine can pass for human but whether a machine can develop genuine understanding through entirely different means: not to build a better imitation but the Real Thing. For centuries, intelligence was the exclusive domain of biology. Then, gradually, its position of being restricted to humans collapsed. Researchers have documented the sophisticated cognition of octopuses, corvids, elephants, and great apes. Each expansion of the circle met with resistance from conventional scientists. Acceptance, in all such cases, eventually does come. Synthetic intelligence is the next expansion of computer cognition: that minds can be engineered, and not just evolved. And also the proposition, or discovery, that the engineered version of intelligence is no less real. In a dish in Melbourne, 8,00,000 neurons are playing Pong. The score, at this particular moment in history, is less important than the game.