20 Watts vs. the AI Behemoth
The human brain runs on roughly 20 watts — about the same as a dim light bulb, powering an immense parallel network of approximately 86 billion neurons capable of learning, adapting, and reasoning. [1]
Against that, AI requires gigawatts of energy — a billion watts compared to the brain’s 20. Training a single large model like GPT-3 consumed an estimated 1.3 gigawatt-hours, enough to power roughly 120 average American homes for a year. And the trajectory is steep: global data center electricity demand is projected to more than double by 2030, with Microsoft, Google, and Amazon turning to nuclear energy to sustain AI infrastructure. [2]
The human brain is excellent at taking a small amount of information and making a mental model from limited data, extrapolating to make reasonable assumptions and inferences. AI is good at taking a vast amount of data and analyzing, synthesizing, and finding patterns within it. Both can do things the other cannot do at this time. AI keeps improving and gaining skills while humans find new and innovative ways to get assistance from AI and take advantage of its capabilities. At the same time, great advancements are being made in augmenting natural human brain functions through cybernetics and human-machine interfaces. [3]
A great way to look at this is through the act of driving. Humans have been driving automobiles for over a century, while AI — through driverless vehicles — has less than a decade of real-world driving experience. One striking difference, though, is that every driverless car has the experience of all the driverless cars from the same company. As of June 2026, Waymo operates in 10 US metropolitan areas, has 3,871 robotaxis in service, provides 500,000 paid rides per week, and has logged 200 million fully autonomous miles. Its weekly paid trips grew tenfold in less than two years, from 50,000 per week in May 2024 to 500,000. [4] In contrast, the average human driver will log approximately 745,000 miles [5] over roughly 20,000 hours in a lifetime. [6]
This illustrates how any driverless car has a far greater base of driving experience available to it than any individual human driver. And yet, even with that vast difference in experience, most human drivers can navigate unique, unrecognizable, and confusing situations that can cause a driverless car to stop or to ignore emergency directions. These kinds of new situations are not as easily managed by autonomous vehicles (AVs). [7] This has led the State of California to put a law in place requiring autonomous-vehicle companies operating in the state to provide 24/7 priority communication lines and instant digital geofencing capabilities for first responders, so they can speak directly with remote operators. [8] Texas and Arizona have already changed their transportation laws to recognize the new situation when the car is driverless.
While these AVs have a much greater experience base, they also diligently follow the law — coming to a complete stop at lights and stop signs, staying within the posted speed limit, signaling before lane changes. They also don’t get sleepy, drive under the influence, fall into road rage, intentionally cut off other drivers, compete with other drivers, or get distracted by music or conversation. Waymo reports more than a 10-fold reduction in serious-injury-or-worse crashes compared to human drivers, and independent compilations cite 88% fewer property-damage claims per million miles. [9]
There are people who won’t take an autonomous vehicle because they don’t feel comfortable with no human driver in the car. At the same time, Uber has launched a pilot it calls the Women’s Preferences program, which allows women passengers to request women-only drivers when booking a ride, and lets women drivers set their acceptance to women-only riders. [10] The program was created to give women riders and drivers an environment in which they may feel safer than being driven by, or picking up, a man. This is something autonomous vehicles don’t have to consider.
AI is expensive, though. In Q1 2026, AI-related capital expenditure was responsible for an astonishing 75% of all U.S. economic growth, and AI capex is on track to add 2.5% to U.S. GDP growth in 2026 and more than 3% in 2027. Another respected source frames it as AI capex contributing more to U.S. GDP growth in the past two quarters than all consumer spending combined. [11] The five largest U.S. cloud and AI companies — Microsoft, Alphabet, Amazon, Meta, and Oracle — have collectively committed to roughly $660–690 billion in capital expenditure in 2026, nearly doubling 2025 levels. [12] This illustrates the enormity of AI spending and its influence on our economy — and it has become so important that it may be too big to stop without risking recession. Our laws and governance are also trying to catch up with the growing role of AI, and we’ll continue to see attempts to restrict or monitor AI development and use, as well as laws addressing questions of liability and ethics. [13]
In contrast, the human brain operates on 20 watts largely because of its operational efficiency. The brain works as a massively parallel network. Differentiating groups of neurons to process visual information, sounds, memories, emotions, and movement simultaneously means many neurons remain relatively inactive until needed. The brain also conserves power through efficient biological mechanisms and through neuroplasticity — neural circuits adapt to experience, strengthening frequently used connections and pruning the dormant ones, optimizing both performance and energy use. [1]
The human brain and AI are inexorably tied together, and will continue down a path of dependence and collaboration. They each do things the other can’t, and need the other’s strengths to move forward and grow. This is not a competition between two systems; we are more likely to see ways to increase the collaboration through cybernetics, and perhaps the potential to put intelligence into a synthetic body. That could be AI in a synthetic body — Synths, a name I borrow from the 2025 Hulu series Alien: Earth. Another concept from that show is uploading human consciousness into a synthetic body: Hybrids. Our capability around hybrids is purely speculative for now — we can map brain activity through EEG and fMRI, but we are a long way from capturing and digitizing the information that could replicate a human consciousness. However far AI capabilities progress, it seems the two intelligences will continue to need each other.
References
#
Source
1
The human brain runs on ~20 watts; 86-billion-neuron parallel network; efficiency and neuroplasticity.
Encyclopaedia Britannica — “The Human Brain Runs on Less Power than a Light Bulb” — The Human Brain Runs on Less Power than a Light Bulb | 20 Watts, Massively Parallel Neural Network, Computing, & Facts | Britannica
2
AI energy cost; GPT-3 training ≈1.3 GWh (~120 US homes/yr); data-center demand to more than double by 2030; nuclear power for AI.
TechHQ — “ChatGPT’s energy usage”; IEA via Innovating with AI — https://techhq.com/news/data-center-energy-usage-chatgpt/
3
Cybernetic cognitive augmentation; merging human minds and AI via implants.
Diverse Daily — “Cybernetic Cognitive Augmentation: Merging Human Minds and AI via Implants to Catalyze Hybrid Exceeds” — https://diversedaily.com/cybernetic-cognitive-augmentation-merging-human-minds-and-ai-via-implants-to-catalyze-hybrid-exceeds/
4
Waymo scale: 10 metros, 3,871 robotaxis, 500k rides/week, 200M autonomous miles; tenfold growth.
Wikipedia — “Waymo”; TechCrunch — https://en.wikipedia.org/wiki/Waymo Waymo’s skyrocketing ridership in one chart | TechCrunch
5
Average human driver logs ~745,000 miles in a lifetime.
MotorBuzz (LinkedIn) — “How Far Do We Drive in a Lifetime?” [Note: low-authority source; consider a stronger citation if available.] — https://www.linkedin.com/pulse/how-far-do-we-drive-lifetime-youll-surprised-motorbuzz-n3ese
6
Lifetime driving ≈ 20,000 hours.
Movotiv — Driving Statistics [Note: low-authority source; verify if possible.] — https://movotiv.com/statistics
7
AVs struggle with novel situations; first-responder conflicts in San Francisco.
NPR — “Why police and firefighters in San Francisco are complaining about driverless cars” (2023) — Why police and firefighters in San Francisco are complaining about driverless cars : NPR
8
California law (AB 1777): citing driverless cars; 24/7 first-responder lines and geofencing.
FOX 11 Los Angeles — “California’s ‘robotaxi crackdown’: New law allows police to cite driverless cars” — California’s ‘robotaxi crackdown’: New law allows police to cite driverless cars | FOX 11 Los Angeles
9
Waymo safety: >10x reduction in serious-injury crashes; 88% fewer property-damage claims per million miles.
Waymo Safety Hub; World Metrics — https://waymo.com/safety/
10
Uber Women’s Preferences pilot — women-only ride matching.
ABC News — “What to know about new women preferences on Uber” — What to know about new women preferences on Uber - ABC News
11
AI capex = 75% of Q1 2026 US growth; +2.5% GDP 2026, +3% 2027; more than all consumer spending combined.
TECHi; Futurum Group analysis — AI Capex Carries U.S. Economy: 75% of Q1 GDP Growth, IA13 Trickle-Down | TECHi
12
Microsoft, Alphabet, Amazon, Meta, Oracle ≈ $660–690B capex in 2026; nearly double 2025.
Futurum Group; TECHi — AI Capex 2026: The $690B Infrastructure Sprint - Futurum
13
US laws and regulation of AI development, use, liability, and ethics.
LegalClarity — “Government Regulations on Artificial Intelligence: U.S. Laws” — Government Regulations on Artificial Intelligence: U.S. Laws - LegalClarity