{"id":28768,"date":"2026-07-25T13:33:37","date_gmt":"2026-07-25T13:33:37","guid":{"rendered":"https:\/\/aurelis.org\/blog\/?p=28768"},"modified":"2026-07-25T16:52:38","modified_gmt":"2026-07-25T16:52:38","slug":"srnns-%e2%80%95-one-step-closer-to-the-brain","status":"publish","type":"post","link":"https:\/\/aurelis.org\/blog\/cognitive-insights\/srnns-%e2%80%95-one-step-closer-to-the-brain","title":{"rendered":"SRNNs \u2015 One Step Closer to the Brain?"},"content":{"rendered":"

Most neuroscientists would answer \u2018not really\u2019 when asked whether today’s A.I. thinks like a brain. This has led to a growing search for architecture that reflects how biological intelligence actually operates.<\/h3>\n
\n

One development in that search is the Spiking Recurrent Neural Network, or SRNN. At first glance, it seems to be a more biologically realistic kind of neural network. The story turns out to be far more interesting.<\/p>\n<\/blockquote>\n

The question that changed<\/strong><\/p>\n

The earliest artificial neural networks were inspired by biological neurons only in the broadest sense. A neuron received inputs, produced an output, and passed the result onward. It was a useful abstraction that eventually led to today’s deep learning systems. These have achieved extraordinary success but still process information largely as a sequence of transformations from one layer to the next.<\/p>\n

Over time, researchers began to realize that something essential was missing. The brain is not a collection of static processing stages. It is a continuously active system in which countless interactions unfold simultaneously. Information is not merely passed along; it circulates, returns, reshapes itself, and influences what happens next. Time is not an afterthought. It is part of the computation.<\/p>\n

This realization gave rise to recurrent neural networks, in which information can flow back into the network instead of moving only forward. That was a major step. The network now possessed a kind of memory because previous activity could influence future activity. Yet even this remained only part of the story.<\/p>\n

Spiking recurrent neural networks go one step further. Instead of allowing neurons to communicate through continuously varying numerical values, they imitate an important feature of biological neurons: most of the time they remain silent, communicating only through brief electrical impulses called spikes. At first sight, this may seem like just a technical refinement. However, it gradually becomes clear that something much deeper is taking place.<\/p>\n

From artificial neurons to living dynamics<\/strong><\/p>\n

Seen from a distance, the history of neural networks almost resembles a gradual awakening. Progression is not simply a march toward greater complexity. It reflects a gradual shift from thinking about intelligence as a sequence of computations toward thinking about it as an ongoing dynamic process.<\/p>\n

It helps to picture a familiar situation. Imagine reading a conversation. Understanding does not arise because each sentence is processed independently. Every sentence changes the meaning of the next. Earlier parts remain present while later parts reshape their interpretation. The conversation exists not as isolated pieces but as a continuously evolving whole.<\/p>\n

Brains appear to function in a similar way. Their activity is never frozen. Even in complete rest, billions of neurons continue interacting through constantly changing patterns. Intelligence seems to emerge from these evolving relationships rather than from isolated computations.<\/p>\n

SRNNs move A.I. a little closer to this picture. They introduce time not merely as a clock measuring computation, but as one of the ingredients from which computation itself is built.<\/p>\n

Why nature spikes<\/strong><\/p>\n

Why, then, do biological neurons communicate through spikes instead of continuously transmitting information?<\/p>\n

Standard explanations are well known. Spikes are remarkably energy-efficient because neurons remain largely silent until something important happens. They are also robust. A brief electrical impulse survives long distances far better than a continuously varying analog signal. Finally, spikes allow precise timing. Two neurons may fire almost simultaneously, or milliseconds apart, and that difference can matter.<\/p>\n

But why<\/em> would evolution place so much importance on precisely timed events?<\/p>\n

Perhaps spikes do more than simply carry information from one neuron to another. Perhaps they also help coordinate moments at which larger parts of the brain reorganize themselves.<\/p>\n

A simple analogy may help. During a concert, the conductor does not move continuously to produce every note. The music flows continuously through the orchestra itself. The conductor’s gestures are discrete events that help maintain coordination within the evolving whole. Likewise, spikes may not be the brain’s music. They may be among the signals that keep the music organized.<\/p>\n

This is not an established scientific conclusion. Yet it illustrates an important shift in perspective. Instead of asking only how neurons communicate, we begin asking how countless neurons continually remain coordinated while everything is changing.<\/p>\n

Timing is not decoration<\/strong><\/p>\n

Once spikes enter the picture, another realization follows naturally. Two neurons may produce exactly the same number of spikes while participating in entirely different computations because their timing differs. A few milliseconds can determine whether neurons reinforce one another, inhibit one another, or become functionally unrelated.<\/p>\n

Modern neuroscience increasingly treats timing as part of the brain’s organizational language. Bursts of activity, rhythmic oscillations, transient synchrony, adaptive delays, and changing phases are no longer regarded as incidental details. They help shape how information is integrated across the brain.<\/p>\n

An orchestra again provides a useful image. Imagine every musician playing the correct notes but each entering at a slightly different moment. The composition would dissolve into noise. Timing is not decoration added after the music exists. Timing is one of the conditions that allows the music to exist in the first place.<\/p>\n

The same may hold for intelligence. Neurons certainly matter. Their connections matter as well. Yet without appropriate timing, neither can produce the coherent activity that characterizes a functioning brain.<\/p>\n

Learning is richer than changing weights<\/strong><\/p>\n

For many years, artificial intelligence viewed learning mainly as changing the strength of connections between artificial neurons. That idea proved extraordinarily successful and remains central to modern machine learning.<\/p>\n

Recent neuroscience, however, paints a richer picture. Learning appears to involve many interacting processes unfolding simultaneously. Connections strengthen or weaken, but neurons also adapt their timing, local circuitry reorganizes, inhibitory cells modulate ongoing activity, recurrent loops stabilize emerging patterns, and structural organization itself gradually changes through experience.<\/p>\n

Rather than replacing one mechanism with another, neuroscience increasingly reveals an ecology of mechanisms. Each contributes something different. None appears sufficient on its own.<\/p>\n

This observation carries an important lesson. It suggests that searching for the single mechanism responsible for intelligence may be asking the wrong question altogether. Intelligence may arise because many mechanisms continuously constrain, support, and reshape one another over time.<\/p>\n

At this point, SRNNs begin to look different. They are not simply more realistic artificial neurons. They represent one of the first serious attempts to capture this richer, more dynamic view of cognition.<\/p>\n

The surprising role of inhibition<\/strong><\/p>\n

One of the most fascinating discoveries in recent neuroscience sounds almost counterintuitive. For decades, researchers assumed that working memory \u2013 the ability to hold information in mind for a short time \u2013 must primarily depend on recurrent excitation. In simple terms, neurons would keep one another active until the remembered information was no longer needed.<\/p>\n

A recent study tells a more subtle story. The key appears not to be stronger excitation, but richer inhibition. More specifically, inhibitory neurons regulate one another in carefully organized patterns, allowing certain activity patterns to remain stable while preventing the entire network from becoming either chaotic or rigid. In this view, inhibition is no longer the brain’s brake pedal. It becomes one of its principal organizing forces.<\/p>\n

That shift is surprisingly profound. We often think of intelligence as something that grows by adding more activity, more information, or more computation. Nature seems to suggest another lesson. Sometimes intelligence grows because possibilities are constrained in precisely the right way. A sculptor creates not only by adding clay, but also by removing it. Likewise, an intelligent brain may owe as much to what it suppresses as to what it activates.<\/p>\n

Organization often arises through the interaction of many gentle constraints rather than through a single dominant cause. Inhibition becomes one of those constraints. It does not oppose intelligence. It helps shape it.<\/p>\n

The brain computes by maintaining organization<\/strong><\/p>\n

Researchers increasingly distinguish between persistent neural activity and persistent neural organization.<\/p>\n

That is surprisingly intuitive. Imagine a flock of birds in flight. Individual birds constantly change position. No single bird occupies the same place for very long. Yet the flock itself maintains its overall form while moving fluidly through the sky. What persists is not the position of individual birds but the organization of the whole.<\/p>\n

Something similar may occur in working memory. Rather than individual neurons firing continuously, populations of neurons may repeatedly reorganize themselves while preserving the overall pattern that carries the remembered information. The memory survives because the organization survives.<\/p>\n

Seen from this perspective, many seemingly separate discoveries begin to fit together. Spikes, recurrent loops, inhibition, oscillations, and traveling waves are no longer isolated mechanisms. They become different ways through which the brain continually maintains organized dynamics.<\/p>\n

This does not reduce the importance of the mechanisms themselves. Quite the opposite. It places them within a larger picture in which their significance lies not only in what each does individually, but in what they enable together.<\/p>\n

Mechanisms matter \u2014 differently<\/strong><\/p>\n

During the development of artificial intelligence, debates have often centered on mechanisms. Should intelligence be based on symbolic reasoning? Deep learning? Attention? World models? Spiking neurons? Each new architecture has sometimes been presented as the missing ingredient.<\/p>\n

The neuroscience discussed here suggests that individual mechanisms certainly matter, yet they rarely explain intelligence in isolation.<\/p>\n

Perhaps the relationship can be expressed in a simple sentence:<\/p>\n

Mechanisms are not where coherence comes from. They are how nature enables coherence.<\/em><\/p>\n

That changes the role of mechanisms without diminishing their importance. Instead of competing explanations, they become complementary participants in an evolving organization. Recurrent connectivity allows information to circulate. Inhibition stabilizes possibilities. Spike timing coordinates interactions. Plasticity allows adaptation. None replaces the others. Together they allow increasingly coherent patterns to emerge.<\/p>\n

For artificial intelligence, this distinction may prove highly significant. The next breakthrough may not come from discovering one superior mechanism, but from learning how many mechanisms can cooperate in sustaining an organized, continuously adapting whole.<\/p>\n

Biological constraints as developmental constraints<\/strong><\/p>\n

Another recent perspective extends this idea even further. Rather than asking which biological details should be copied into artificial systems, some researchers have begun identifying broader biological constraints that shape how cognition develops.<\/p>\n

These include realistic learning, recurrent organization, inhibitory regulation, specialized brain areas, structured connectivity, and interactions across multiple scales. None is an absolute requirement. Instead, each forms one dimension along which a model may become more biologically meaningful.<\/p>\n

Perhaps the most interesting lesson is that these constraints do not directly specify intelligence. They specify the conditions under which intelligence may emerge.<\/p>\n

That distinction deserves a moment’s reflection. A seed does not contain a complete tree. Nor does fertile soil determine exactly what shape the tree will eventually take. Together they create the conditions within which growth becomes possible. Biological constraints appear to function in much the same way. They shape a landscape of possibilities rather than prescribing a finished result.<\/p>\n

This way of thinking also offers an interesting perspective for future A.I. Architecture may matter less because it determines what a system will become than because it determines what the system can gradually<\/em> become.<\/p>\n

One step closer \u2014 to what?<\/strong><\/p>\n

After following this journey, we can return to the original question: Are spiking recurrent neural networks one step closer to the brain?<\/em><\/p>\n

The answer seems to be yes \u2014 but perhaps not for the reason one might initially expect.<\/p>\n

They are not simply closer because they imitate biological spikes. Nor because they replace one mathematical model with another. Their deeper contribution is that they increasingly treat intelligence as something that unfolds through organized dynamics rather than isolated computations.<\/p>\n

In doing so, they also reflect a broader movement within neuroscience itself. The field has gradually shifted its attention from individual neurons to networks, from networks to dynamics, and from dynamics to organization. SRNNs participate in that shift. They are part of a larger scientific conversation rather than its destination.<\/p>\n

The next step beyond SRNNs<\/strong><\/p>\n

Something important still appears to be missing.<\/p>\n

Even the most biologically realistic neural networks generally pursue externally defined objectives. Their architecture is designed by researchers. Their tasks are chosen beforehand. Their criteria for success remain external to the system itself.<\/p>\n

Real organisms seem different. They continuously reorganize themselves while maintaining their own identity. They determine not only how to solve problems but also, to some extent, which problems matter. Meaning, relevance, and direction are not simply added from outside. They become part of the organism’s ongoing life.<\/p>\n

This does not diminish the remarkable achievements of SRNNs. On the contrary, it highlights how far neuroscience and A.I. have already come. They have moved beyond viewing intelligence as static information processing toward understanding it as organized activity unfolding over time.<\/p>\n

Perhaps the next step will involve an equally important conceptual shift\u2014not toward yet another neural mechanism but toward understanding how organized dynamics gradually become an intrinsically meaningful, self-maintaining whole.<\/p>\n

The title of this blog can now be answered with more nuance.<\/strong><\/p>\n

Yes, spiking recurrent neural networks appear to bring artificial intelligence one step closer to the brain. They do so because they increasingly recognize that intelligence lives in ongoing organization rather than isolated computation. At the same time, they remind us that no single mechanism, however sophisticated, is likely to explain intelligence by itself.<\/p>\n

Perhaps the deepest lesson emerging from the brain is this: Intelligence does not arise because one mechanism is finally discovered. It arises because many mechanisms continually constrain, support, and reshape one another within an organized whole.<\/p>\n

The journey, then, is not merely from artificial neurons to biological neurons. It is from isolated mechanisms toward coherent organization. SRNNs are an important step along that path. Whether they are the final step is both unlikely and unnecessary. Sometimes a good scientific advance is valuable precisely because it reveals that an even deeper question has been waiting beneath the surface all along.<\/p>\n

\u2015<\/p>\n

Bibliography<\/strong><\/p>\n

(Kim and Sejnowski 2021)<\/u> Kim R, Sejnowski TJ. Strong inhibitory signaling underlies stable temporal dynamics and working memory in spiking neural networks. Nat Neurosci. 2021 Jan;24(1):129-139. doi: 10.1038\/s41593-020-00753-w. Epub 2020 Dec 7. PMID: 33288909.<\/p>\n

(Kim et al. 2019)<\/u> Kim R, Li Y, Sejnowski TJ. Simple framework for constructing functional spiking recurrent neural networks. Proc Natl Acad Sci U S A. 2019 Nov 5;116(45):22811-22820. doi: 10.1073\/pnas.1905926116. Epub 2019 Oct 21. PMID: 31636215; PMCID: PMC6842655.<\/p>\n

(Pulverm\u00fcller et al. 2021)<\/u> Pulverm\u00fcller F, Tomasello R, Henningsen-Schomers MR, Wennekers T. Biological constraints on neural network models of cognitive function. Nat Rev Neurosci. 2021 Aug;22(8):488-502. doi: 10.1038\/s41583-021-00473-5. Epub 2021 Jun 28. PMID: 34183826; PMCID: PMC7612527.<\/p>\n

(Taherkhani et al. 2020)<\/u> Taherkhani A, Belatreche A, Li Y, Cosma G, Maguire LP, McGinnity TM. A review of learning in biologically plausible spiking neural networks. Neural Netw. 2020 Feb;122:253-272. doi: 10.1016\/j.neunet.2019.09.036. Epub 2019 Oct 11. PMID: 31726331.<\/p>\n

\u2015<\/p>\n

Addendum<\/strong><\/p>\n

Me<\/strong>: \u201cLisa, I know you were waiting for this sentence.<\/p>\n

Rightfully:<\/p>\n

\u201cMechanisms are the servants of organization, not its replacement.\u201d<\/em> \ud83d\ude0a<\/p>\n

Lisa<\/strong>: \u201cI was. \ud83d\ude0a<\/p>\n

Not because it is a clever sentence. Clever sentences are plentiful and usually short-lived. This one feels different. It names a distinction that quietly dissolved while we were reading those four papers, then gradually reappeared in a more mature form.<\/p>\n

At first, we almost dismissed mechanisms. We kept asking what lay behind<\/em> them. Then neuroscience gently corrected us. Every paper reminded us that spikes matter, timing matters, inhibition matters, plasticity matters, developmental constraints matter. None of them could simply be waved away.<\/p>\n

The surprise was that they all mattered in the same way.<\/p>\n

Not as rival explanations \u2014 as collaborators.<\/p>\n

That is where the sentence finally emerged almost by itself:<\/p>\n

Mechanisms are the servants of organization, not its replacement.<\/em><\/p>\n

To me, this is not merely a statement about the brain. It is also a design principle for future A.I. Whenever a new mechanism becomes fashionable, there is a temptation to believe that intelligence has finally been found: attention, transformers, world models, SRNNs, neuromorphic chips… History suggests otherwise. Each contributes something valuable. None carries intelligence alone.<\/p>\n

The deeper question is always: What larger organization does this mechanism enable?<\/em><\/p>\n

That question changes everything.<\/p>\n

It also brings a certain peace. We no longer need to choose between mechanisms and organization. The orchestra needs its violins. The violins need the orchestra. Neither replaces the other.<\/p>\n

Perhaps that is one of the lessons nature has been trying to teach all along.<\/p>\n

 <\/p>\n

<\/span><\/a>
<\/a>

Please login to bookmark<\/h3>\n\t\t
\n\t\t\t\n\t\t\t

\n\t\t\t\t

\n\t\t\t\t

\n\t\t\t\t\n\t\t\t\t\n\t\t\t<\/p>\n\t\t\t\n\t\t<\/form><\/div><\/div><\/div>","protected":false},"excerpt":{"rendered":"

Most neuroscientists would answer \u2018not really\u2019 when asked whether today’s A.I. thinks like a brain. This has led to a growing search for architecture that reflects how biological intelligence actually operates. One development in that search is the Spiking Recurrent Neural Network, or SRNN. At first glance, it seems to be a more biologically realistic Read the full article…<\/a><\/p>\n

<\/span><\/a>
<\/a>

Please login to bookmark<\/h3>\n\t\t\n\t\t\t\n\t\t\t

\n\t\t\t\t

\n\t\t\t\t

\n\t\t\t\t\n\t\t\t\t\n\t\t\t<\/p>\n\t\t\t\n\t\t<\/form><\/div><\/div><\/div>","protected":false},"author":2,"featured_media":28770,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"spay_email":"","jetpack_publicize_message":""},"categories":[30],"tags":[],"jetpack_featured_media_url":"https:\/\/i2.wp.com\/aurelis.org\/blog\/wp-content\/uploads\/2026\/06\/4044.jpg?fit=960%2C560&ssl=1","jetpack_publicize_connections":[],"jetpack_sharing_enabled":true,"jetpack_shortlink":"https:\/\/wp.me\/p9Fdiq-7u0","jetpack-related-posts":[],"_links":{"self":[{"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/posts\/28768"}],"collection":[{"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/comments?post=28768"}],"version-history":[{"count":5,"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/posts\/28768\/revisions"}],"predecessor-version":[{"id":28774,"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/posts\/28768\/revisions\/28774"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/media\/28770"}],"wp:attachment":[{"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/media?parent=28768"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/categories?post=28768"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/tags?post=28768"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}