By Javier Surasky
(Estimated reading
time: about 14 minutes)
Since
artificial intelligence began to integrate into everyday life, an almost
automatic comparison has emerged: Does AI resemble the human brain?
The
question arises intuitively because both "learn" and "generate responses", but it
also leads to confusion if not scrutinized. Understanding the
similarities, how they diverge, and the ethical and governance
implications of this relationship is essential, especially from the
perspectives of international relations and the social sciences.
The human
brain is a biological organ—the product of millions of years of evolution. It
is estimated to contain approximately 86 billion neurons, interconnected by
chemical and electrical synapses, forming a hierarchical, modular
architecture of enormous complexity. As Suzana Herculano-Houzel (2009:2) notes, "the human brain is not exceptional in its cellular composition", which
places its organization in continuity with that of other mammals, though with an exceptionally high number of neurons.
There are
multiple types of neurons in the brain, each with specific electrophysiological
properties that are influenced by neurotransmitters such as dopamine, serotonin,
norepinephrine, and acetylcholine. Brain functioning is also immersed in a
chemical–hormonal environment that regulates its activity, modulates its
plasticity, and shapes its learning capacity.
Modern AI,
by contrast, runs on specialized digital hardware such as GPUs, TPUs, and other
accelerators, and is based on mathematical models that optimize specific
functions. A central component of this AI is the so-called "artificial
neurons," a concept that dates back to the mid-1950s and has gained relevance
with the rise of machine learning and deep neural networks.
Each
artificial neuron in a neural network "computes a weighted sum of their
inputs and apply a nonlinear function to this sum" (Goodfellow, Bengio
& Courville, 2016:167) to produce an output.
Their similarity to a biological neuron is, however, overstated, and the distance between the two types of neurons is vast.
To begin
with, consider energy efficiency. The Human Brain Project (2023) estimates that
"a human brain uses roughly 20 Watts to work", less than a household
light bulb, while training large-scale AI models requires vastly greater
energy and infrastructure resources.
In the
domain of learning, the brain relies on synaptic plasticity—that is, the
ability to strengthen or weaken connections between neurons depending on joint
activation patterns, modifying synaptic efficacy and the shape of dendritic
spines. As Citri and Malenka (2008:18) summarize, "Synaptic plasticity is
the major mechanism by which experience modifies brain function." It is a
continuous, multimodal, and deeply contextual form of learning that shapes the configuration of our brains in response to our experiences and needs.
AI, by
contrast, learns through statistical optimization: loss functions, gradients,
and rules for adjusting the weights of its artificial neurons, along with other
statistical–mathematical techniques. Its goal is to process large volumes of
data. In the words of LeCun, Bengio, and Hinton (2015:43), "deep learning
allows computational models that are composed of multiple processing layers to
learn representations of data with multiple levels of abstraction."
Although
both systems modify internal connections, AI requires enormous amounts of data,
learns in discrete stages, depends on predefined error functions, and lacks the
kind of innate inductive biases that guide human learning.
Another
superficial similarity is distributed representation: neither the brain nor AI
stores concepts in a single unit. Both generate complex activation patterns that enable AI models to serve as working hypotheses for studying vision,
language, and semantic categorization. However, this functional convergence does
not imply cognitive equivalence. Humans incorporate social context, ethical
intuitions, autobiographical memory, and embodied experience into our cognitive
processes. AI, in contrast, generates outputs based on statistical correlations
with no semantic understanding or subjective experience. As Bender, Gebru,
McMillan-Major, and Mitchell (2021:1) observed, "an LM is a system for
haphazardly stitching together sequences of linguistic forms (…) without any
reference to meaning."
The deepest
difference arises when considering consciousness. In the film Transcendence
(2014), the AI scientist played by Morgan Freeman poses a persistent question
to the advanced AI systems he encounters: "Do you have self-awareness?" This is the
critical boundary: the human brain not only processes information but also
generates emotions, intentionality, and subjective experience. No current AI
model is capable of that. "Conscious access allows us to extract meaning and
reason about it" (Dehaene, 2014:105).
If
biological neurons are so superior to artificial ones, can a person run an AI
algorithm on their neurons? The answer is categorical: no. There are
structural, biological, and computational reasons behind this conclusion.
An
artificial algorithm operates by applying explicit mathematical operations. The brain has neither "mathematical working memory nor an operation register that would allow this, which is why our brain can perform mathematical
operations but with far less precision than AI. We have not develooed "tensors"
(a structured way of organizing numbers into multiple dimensions so that AI
models can process them efficiently) "or "matrices" that can be manipulated
with exactness during our biological development, and without them the
computation carried out by AI models is impossible.
If this
were not enough, algorithms operate through ordered steps (basically forward
pass, loss computation, backward pass, and parameter adjustment), whereas brain
dynamics are neither sequential nor deterministic, but massive, parallel, and
continuous, since "the brain is never silent; neuronal activity is
continuous even in the absence of external input" (Buzsáki, 2006:15). Its
activity cannot be paused or arranged into stages. It is not mathematics
but local biochemical rules modulated by neurotransmitters that give order to
neural processing. It is a noisy system (we will return to noise shortly) since
"noise is present at all stages of neural processing and fundamentally
shapes neural function" (Faisal, Selen & Wolpert, 2008:292). In part,
this noise arises from the fact that the human brain "lacks a single central
controlling structure; control is distributed across many interacting regions"
(Sporns, 2011:89), meaning it is a nonlinear system lacking a defined
computational flow, a fixed architecture, or an externally imposed execution
order—all of which are present in AI models.
More
reasons? AI models store information in vectors and matrices, whereas the brain
encodes information in distributed neuronal ensembles, using representations that include
affective behaviors and biologically predetermined responses (e.g., fear) whose weights cannot be explicitly adjusted to achieve predefined
goals.
By
contrast, the brain can learn to simulate cognitive strategies—such as learning
rules, identifying patterns, reinforcing behaviors through trial and error, or
generalizing from examples—which allows it to apply heuristics and behave "as if" it were running an algorithm.
Another
fundamental element is that the basic unit of the brain is an action potential,
which depends on sodium, potassium, calcium, neurotransmitters, temperature,
fatigue, prior history, and multiple interacting variables. In the words of
Kandel, Koester, Mack, and Siegelbaum (2021:14), "the action potential is
the fundamental signal that carries information through the nervous system."
Everything there is, by essence, imprecise, noisy, and contextual. In contrast,
the fundamental unit of any AI system is an exact matrix operation, free of
biological noise and controlled with floating-point precision. In short, AI
works through precision, whereas the brain works through the biological
noise it generates.
For this
reason, human memory—our archive—is semantic, "knowledge about the world
that is not tied to specific experiences" (Tulving, 1983:386). It "saves" meanings, personal histories, social context, emotions, together with logical
rules. It is a distributed (not centralized) memory integrated into historical
and bodily experience. AI stores statistical correlations encoded as numerical
weights—millions of them—but nothing more. This enables the brain to retrieve why
things happen, while AI focuses on retaining how to produce outputs over
time. AI "remembers" as vectors; the brain has multimodal memories
(auditory, visual, emotional, motor) because those have been its channels for
sensory experience.
All this is
entirely logical, since the brain learns what a person needs to live, while AI
learns what it is given to optimize a process. The brain understands; AI
predicts. Understanding leads to wisdom; prediction to accuracy. The brain is
part of each individual’s subjectivity, whereas AI is an intangible object—or,
if preferred, a process. The brain has a “self,” and therefore its own goals;
AI has neither.
In general
terms, we can summarize our negative answer by claiming that the brain is a
self-organized system developed by genetic evolution. In contrast, AI is a system
specified by design. This establishes fundamentally different bases for their
organization, functioning, and processing modes.
If we
combine this with the “astonishment” generated by conversations with an LLM
that does not understand what it says, we reach a crucial point: the brain and
AI may converge in apparent behaviors, but their architectures and dynamics are
irreducibly different.
These
differences should not be understood as a call to neglect the study of the
relationship between AI and the human brain. Both fields nourish each other: AI
algorithms have been inspired by biological principles such as plasticity, and
neuroscience relies on AI models to explore hypotheses and analyze large
volumes of neural data.
Moreover,
in these differences, we find the origin of ethical and governance issues that
must be urgently addressed.
The
temptation to anthropomorphize AI can shift responsibility from people toward
technology, diverting attention toward science-fiction scenarios—utopian or
dystopian—and obscuring real risks such as data bias, lack of transparency,
concentration of power, and the deepening of inequalities. As the Chinese
government states in its Global AI Governance Initiative (2023), “we
must adhere to the principle of developing AI for good, respect the relevant
international laws, and align AI development with humanity's common values.”
Understanding
that AI is not an artificial brain but a statistical artifact with significant
consequences for social life compels the formulation of policies that address
current risks: audits, traceability, transparency, capacity-building, and
strengthening digital infrastructure in lagging countries, in line with the
call of the UN General Assembly in resolution A/RES/78/265 (2024) to “close
the digital divides and to promote equitable access to the benefits of safe,
secure and trustworthy artificial intelligence systems.” Institutions
created to mitigate systemic risks must be understood as contemporary
mechanisms for protecting human rights.
In that
direction, UNESCO’s Recommendation on the Ethics of Artificial Intelligence
affirms that “the protection of human rights and dignity is the cornerstone
of the Recommendation” (UNESCO, 2021: para. 23). Similarly, the European AI
Act states its commitment to “improve the functioning of the internal market
and promote the uptake of human-centric and trustworthy artificial intelligence
while ensuring a high level of protection of public interests such as health,
safety and fundamental rights.”
For its
part, the United States government has pursued, with particular emphasis since
early 2025, an AI policy that prioritizes innovation, digital infrastructure
development, and investment over citizen rights. The Executive Order Removing
Barriers to American Leadership in Artificial Intelligence, of January 23
of that year, declares that the U.S. “must develop AI systems that are free
from ideological bias or engineered social agendas” and establishes as
national policy “to sustain and enhance America’s global AI dominance.”
This does
not imply neglecting the study of the relationship between AI and the human
brain. Both fields nourish each other: AI algorithms have been inspired by
biological principles such as plasticity, and neuroscience relies on AI models
to explore hypotheses and analyze large volumes of neural data.
In summary,
the human brain and AI share certain minimal, very abstract principles, but differ profoundly in structure, purpose, and capabilities. AI is not—and
cannot be—a “digital super-mind,” but a set of algorithms designed for specific
tasks. Understanding this distinction is essential for a serious debate on
regulation and governance that allows society to benefit from AI without
undermining fundamental rights or eroding autonomy and democracy.
References
Bender, E.
M., Gebru, T., McMillan-Major, A., & Mitchell, M. (2021). On the dangers
of stochastic parrots: Can language models be too big? In Proceedings of
the 2021 ACM Conference on Fairness, Accountability, and Transparency (pp.
610–623). Association for Computing Machinery. https://doi.org/10.1145/3442188.3445922
Buzsáki, G.
(2006). Rhythms of the brain. Oxford University Press.
Citri, A.,
& Malenka, R. C. (2008). Synaptic plasticity: Multiple forms, functions,
and mechanisms. Neuropsychopharmacology, 33(1), 18–41. https://doi.org/10.1038/sj.npp.1301559
Dehaene, S.
(2014). Consciousness and the brain: Deciphering how the brain codes our
thoughts. Viking.
European
Parliament & Council of the European Union. (2024). Artificial
Intelligence Act. EUR-Lex. https://eur-lex.europa.eu/
Faisal, A.
A., Selen, L. P. J., & Wolpert, D. M. (2008). Noise in the nervous system. Nature
Reviews Neuroscience, 9(4), 292–303. https://doi.org/10.1038/nrn2258
Goodfellow,
I., Bengio, Y., & Courville, A. (2016). Deep learning. MIT Press. https://www.deeplearningbook.org
Herculano-Houzel,
S. (2009). The human brain in numbers: A linearly scaled-up primate brain. Frontiers
in Human Neuroscience, 3, Article 31. https://doi.org/10.3389/neuro.09.031.2009
Human Brain
Project. (2023, September 4). Learning from the brain to make AI more
energy-efficient. https://www.humanbrainproject.eu/en/follow-hbp/news/2023/09/04/learning-brain-make-ai-more-energy-efficient/
Kandel, E.
R., Koester, J. D., Mack, S. H., & Siegelbaum, S. A. (2021). Principles
of neural science (6th ed.). McGraw-Hill.
https://accessmedicine.mhmedical.com/book.aspx?bookid=3249
LeCun, Y.,
Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553),
436–444. https://doi.org/10.1038/nature14539
People’s
Republic of China. Cyberspace Administration of China. (2023). Global AI
Governance Initiative. Ministry of Foreign Affairs of the People’s Republic
of China. https://www.mfa.gov.cn/eng/wjdt_665385/2649_665393/202310/t20231018_11159238.html
Sporns, O.
(2011). Networks of
the brain. MIT
Press. https://mitpress.mit.edu/9780262528986/networks-of-the-brain/
Tulving,
E. (1983). Elements
of episodic memory.
Oxford University Press.
UNESCO.
(2021). Recommendation on the Ethics of Artificial Intelligence. UNESCO.
https://unesdoc.unesco.org/ark:/48223/pf0000381137
United
Nations General Assembly. (2024). Resolution A/RES/78/265. Seizing the
opportunities of safe, secure and trustworthy artificial intelligence systems
for sustainable development. https://documents-dds-ny.un.org/doc/UNDOC/GEN/N24/060/68/PDF/N2406068.pdf