The article by Luís Rita explores the central paradox of the artificial intelligence revolution: human biological neural networks, over billions of years of evolution, have created artificial neural networks whose results are now transforming the environment that will shape human evolution itself. The human brain has approximately 86 billion neurons, while models such as DeepSeek-V3 have 671 billion parameters, although an artificial parameter does not correspond directly to a biological neuron or synapse, which change over time and depend on electrical activity, chemical signals and neuromodulators.
In the energy comparison, the brain operates on just about 20 watts, simultaneously maintaining perception, memory, language, motor control and consciousness. Training Llama 3.1 with 405 billion parameters is estimated to have consumed approximately 21.6 gigawatt-hours of energy, equivalent to the operation of a 20-watt brain for more than 120,000 years. After training, each query to a frontier model consumes approximately 0.34 watt-hours, roughly the energy the brain uses during one minute, although the brain simultaneously processes balance, temperature, emotions and signals from all organs.
Limited cycles of self-improvement already exist in artificial networks, such as AlphaEvolve and SIMA 2, which generate programs, automatically evaluate them and use the experience to train subsequent versions. UNESCO adopted in November 2025 the first global regulatory framework for neurotechnology ethics, and Portugal approved in January 2026 a National Artificial Intelligence Agenda with 32 initiatives and more than 400 million euros by 2030. Brain-computer interfaces have already allowed tetraplegic patients to walk again, and Neuralink has given public visibility to the field, although the most consolidated clinical evidence belongs to a broader scientific field.
The article concludes that the decisive question may no longer be whether machines will become human, but whether humans will remain exclusively biological. Luís Rita argues that in the long term, gradual integration between biological and synthetic intelligence could be the only strategy to avoid permanent separation, but warns that the brain cannot become a subscription-based platform, requiring open norms, interoperability, portability of neuronal data and prohibition of commercial lock-in.
