Living Human Neurons as Processors: FinalSpark and the Future of Biocomputing

概览

This episode of Data Science with Sam discusses biological computing with Dr. Lena Curtis, a neuroscientist and strategic advisor at FinalSpark. The central idea is that living human neurons grown in a lab could act as computing processors, not as a power source, and potentially offer a radically more energy-efficient substrate for AI.

The conversation moves from Curtis’s career path in brain imaging, medical imaging startups, and AI into FinalSpark’s work on a remotely accessible “neuro platform.” The episode emphasizes that the field is still early: researchers can stimulate and read neural activity, but they do not yet fully understand how neurons encode information.

Key conclusions are that biocomputing is not positioned as a full replacement for silicon. Curtis presents it as a complementary future hardware layer, especially relevant for energy-intensive AI workloads, while also highlighting major open questions around learning in vitro, cost, regulation, ethics, and social acceptance.

分段落总结

[00:07] Introducing Biocomputing and FinalSpark

[事实] The host opens by asking what computing would look like if the next generation of computers used living human neurons instead of silicon.

[事实] FinalSpark is introduced as a Swiss startup connecting lab-grown brain cells to electrodes and using them to process information.

[事实] Dr. Lena Curtis is introduced as a neuroscientist with a background in brain imaging, medical imaging, AI, and biocomputing.

[推测] The episode frames biocomputing as a frontier technology that sits between neuroscience, AI hardware, and deep tech entrepreneurship.

[01:38] Lena Curtis’s Path into AI and Biocomputing

[事实] Curtis did her PhD in medical brain imaging and remained focused on the brain throughout her academic work.

[事实] She moved into industry through medical imaging startups in London, where large imaging datasets exposed her to automation, deep learning, and AI.

[事实] She met FinalSpark’s founders at an AI Summit in London and became interested in the combination of neuroscience and artificial intelligence.

[事实] She learned some Python and worked on commercial AI applications before joining FinalSpark’s work on the future of AI.

[04:49] Why Energy Efficiency Matters

[事实] Curtis says FinalSpark was founded in 2014 by Fred Jordan and Martin Kutter, two Swiss entrepreneurs with PhDs in signal processing.

[事实] The founders initially wanted to build a thinking machine, but they concluded that AI progress depends heavily on budget, servers, training, and usage costs.

[事实] Curtis says living neurons are around one million times more energy efficient and could address AI scalability problems.

[推测] The motivation for biocomputing is not only scientific curiosity, but also the strategic need for smaller players to pursue a hardware path that large AI budgets cannot simply dominate.

[08:02] Neurons as Processors, Not Power Sources

[事实] The host initially describes neurons as an alternate power resource, and Curtis corrects this point.

[事实] Curtis explains that neurons do not provide energy to computers; they are alternative processors.

[事实] The intended comparison is closer to a new kind of CPU or GPU, not a biological battery.

[推测] This correction is important because the value proposition depends on lower energy use during computation, not on neurons generating power.

[08:57] Programming and Learning in Neural Cultures

[事实] Curtis says neurons already process information continuously in the human brain.

[事实] The core challenge is how to control the process when neurons are grown in a dish, because they are dynamic, plastic systems that change over time.

[事实] She identifies learning in vitro and understanding how neurons encode information as the biggest current challenges.

[事实] Researchers can measure neural activity and see signals being processed, but they do not yet know exactly what that activity means.

[推测] The bottleneck is less about building an interface and more about discovering the “language” of biological neural activity.

[11:15] Digital-Biological Input and Output

[事实] Curtis explains that neurons are placed on electrodes, and their analog signals are converted into digital signals through analog-to-digital converters.

[事实] The measured neural activity consists of electrical spikes, which can be sent to a computer for analysis.

[事实] Researchers can write Python code to run experiments, analyze data, and define stimulation procedures.

[事实] Signals sent back to neurons are converted from digital to analog form and delivered as electrical impulses through the same electrodes.

[13:25] Remote Access to Living Neural Systems

[事实] The host describes a researcher in Tokyo or Bristol logging into a browser, writing Python code, and sending electrical signals to neurons located in Switzerland.

[事实] Curtis confirms the broader technical principle and connects it to electrophysiology methods that have been used for a long time to measure electrical activity in tissues.

[事实] FinalSpark applies these existing biological measurement methods to the different goal of biocomputing.

[推测] The platform model makes living neural tissue function more like remotely accessible research infrastructure than a traditional local lab experiment.

[14:22] How Neurons Differ from Digital Hardware

[事实] Curtis says neurons process information in a completely different way from digital systems, without logic gates or a zero-one structure.

[事实] Neural information includes both timing and spatial location of activity.

[事实] She says neurons learn continuously, while AI systems are usually updated globally as models.

[事实] She also notes that biological brains filter information selectively, contain many recurrent connections, and combine memory and processing in one place.

[推测] These properties help explain why neurons may be energy efficient, but they also make them harder to engineer predictably.

[16:37] Cost, Data Centers, and Commercial Potential

[事实] The host asks how human neuron processing might compare economically with NVIDIA or other AI hardware.

[事实] Curtis says energy efficiency only matters if it has practical implications, and she argues that neurons should also be much cheaper.

[事实] She says silicon systems require substantial energy and water for cooling, while future biocomputers are expected to be cheaper even after facility maintenance costs.

[事实] Curtis claims that if biocomputing can replace some AI tasks, especially expensive generative AI workloads, it could support very large profits.

[推测] The commercial thesis depends on future biocomputers reaching usable scale, which Curtis presents as an expectation rather than a proven current capability.

[18:23] Applications, Collaborations, and Current Stage

[事实] Curtis says FinalSpark believes generative AI is the strongest potential application, but she clearly labels this as an assumption because biocomputers do not yet fully exist.

[事实] She says the company is still doing fundamental research into information processing and has stored one bit of information in neurons.

[事实] FinalSpark invited nine universities from around the world after selecting projects from 34 candidates.

[事实] The company rents access to its neuro platform to companies, individuals, and universities interested in researching how neurons process information.

[事实] Curtis says external collaborators need to bring their own ideas, while FinalSpark also has internal R&D focused on patents.

[20:55] Ethics, Regulation, and Human-Machine Boundaries

[事实] Curtis says there is currently no legislation specific to biocomputing, which she considers understandable because the field is still in active research.

[事实] FinalSpark has reached out to philosophers to encourage work on the ethics and philosophy of biocomputing.

[事实] Curtis discusses the long history of humans interacting with tools and technology, including smartphones, prosthetics, pacemakers, and robotic systems that may include biological components.

[事实] She says living neurons in machines raise questions about the boundary between human and machine.

[推测] Her position is that public acceptance and ethical framing may become as important as technical progress for the adoption of biocomputing.

[25:38] A Future of Mixed Computing Substrates

[事实] Curtis says she expects silicon, quantum, biological computing, and possibly other substrates to coexist.

[事实] She says future hardware will likely be more diverse, with chips optimized for different purposes such as AI and large language models.

[事实] She does not expect biocomputing to replace all digital computing.

[事实] She says neurons are slow, so applications requiring high speed are unlikely to use biocomputing.

[推测] Biocomputing is presented as a specialized complement for suitable workloads rather than a universal computing replacement.

[27:12] What Could Change for Users

[事实] Curtis says the biggest change could be the cost of AI.

[事实] She argues that users do not fully feel AI’s cost today because many models are still in an adoption phase.

[事实] She expects AI may become more expensive as usage grows, which could make it less accessible.

[事实] She says the biggest expected benefit of biocomputing is accessible, cheap artificial intelligence.

[28:10] Where Listeners Can Learn More

[事实] Curtis says listeners can reach her on LinkedIn.

[事实] She points listeners to finalspark.com for information about FinalSpark’s projects, importance, and technical documentation for engineers.

[事实] She says people can use the website contact form and that FinalSpark answers emails.

[事实] The host says he will share Curtis’s LinkedIn profile, personal blog, and the FinalSpark website in the episode description.

播客点评/总结

This episode is valuable because it makes a speculative-sounding technology concrete. The strongest parts are Curtis’s clarifications: neurons are processors rather than power sources, the interface uses electrodes and signal converters, and the field is still at a fundamental research stage rather than already delivering full AI systems.

The discussion is also useful for AI listeners because it connects biocomputing to practical concerns around energy, compute cost, data centers, and hardware diversity. Curtis repeatedly avoids overstating current capability, especially when discussing generative AI as a future application.

The main limitation is that several claims remain forward-looking, including cost reductions, future profits, and the role of neurons in generative AI. These are presented as expectations or assumptions rather than demonstrated outcomes.

[推测] This episode is best suited for listeners interested in AI hardware, neuroscience, deep tech startups, and the ethics of emerging computing systems, rather than listeners looking for a mature product walkthrough or near-term deployment guide.