Reading
Deep learning gets the glory, deep fact checking gets ignored
Deep learning is glamorous and highly rewarded. If you train and evaluate a Transformer (a state-of-the-art language model) on a dataset of 22 million enzymes and then use it to predict the function of 450 unknown enzymes, you can publish your results in Nature Communications (a very well-regarded publication). Your paper will be viewed 22,000 times and will be in the top 5% of all research outputs scored by Altmetric (a rating of how much attention online articles receive). However, if you do the painstaking work of combing through someone else’s published work, and discovering that they are riddled with serious errors, including hundreds of incorrect predictions, you can post a pre-print to bioRxiv that will not receive even a fraction of the citations or views of the original. In fact, this is exactly what happened in the case of these two papers:
Human Brain Cells on a Chip for Sale
In a development straight out of science fiction, Australian startup Cortical Labs has released what it calls the world’s first code-deployable biological computer. The CL1, which debuted in March, fuses human brain cells on a silicon chip to process information via sub-millisecond electrical feedback loops.
The recent successes of generative AI models have convinced some that AGI is imminent. While these models appear to capture the essence of human intelligence, they defy even our most basic intuitions about it. They have emerged not because they are thoughtful solutions to the problem of intelligence, but because they scaled effectively on hardware we already had. Seduced by the fruits of scale, some have come to believe that it provides a clear pathway to AGI. The most emblematic case of this is the multimodal approach, in which massive modular networks are optimized for an array of modalities that, taken together, appear general. However, I argue that this strategy is sure to fail in the near term; it will not lead to human-level AGI that can, e.g., perform sensorimotor reasoning, motion planning, and social coordination. Instead of trying to glue modalities together into a patchwork AGI, we should pursue approaches to intelligence that treat embodiment and interaction with the environment as primary, and see modality-centered processing as emergent phenomena.
How Much Energy Does It Take To Think?
Sharna Jamadar, a neuroscientist at Monash University in Australia, and her colleagues reviewed research from her lab and others around the world to estimate the metabolic cost of cognition — that is, how much energy it takes to power the human brain. Surprisingly, they concluded that effortful, goal-directed tasks use only 5% more energy than restful brain activity. In other words, we use our brain just a small fraction more when engaging in focused cognition than when the engine is idling.
The Best Illusion of the Year Contest is a celebration of illusions and perception, created by the ingenuity of the world’s premier illusion creators. Illusions are perceptual experiences that do not match the physical reality. How we see the outside world―our perception―is generated indirectly by brain mechanisms, and so all perception is illusory to some extent. The study of illusions is critical to how we understand sensory perception, and many ophthalmic and neurological diseases.
Revisiting Minsky’s Society of Mind in 2025
Minsky’s core proposal in The Society of Mind is elegant and radical: “The power of intelligence stems from our vast diversity, not from any single, perfect principle.”
Rather than a single, unified “genius” solving problems, our minds are portrayed as assemblies of many tiny agents, each with limited ability. These agents form hierarchies and teams (which Minsky calls “agencies”), where each sub-agent handles a piece of a task, and higher-level agents coordinate or choose which agents should act. Intelligence, in this view, emerges from the interplay of lots of simple parts, much like a society’s culture emerges from many individuals. No single component understands the whole, yet together they achieve complex, adaptive behavior.
Importantly, Minsky also anticipated the need for oversight and self-regulation within this agent society. He described how minds avoid mistakes by employing what he called “negative expertise” – essentially knowledge about what not to do. In Minsky’s model this takes the form of special “censor” and “suppressor” agents that watch for dangerous or unproductive impulses. “Censors suppress the mental activity that precedes unproductive or dangerous actions, while suppressors suppress those unproductive or dangerous actions themselves,” he wrote . In other words, one part of the mind can veto or inhibit another, providing a safety check against runaway behaviors or known pitfalls.
Attractor Networks, (A bit of) Computational Neuroscience Part III
Brains are comprised of networks of neurons connected by synapses, and these networks have greater computational properties than the neurons and synapses themselves. In this post, I am going to talk about a class of neural networks which I think are fascinating: attractor networks. These are recurrent neural networks with attractor states; these states and the dynamics governing an attractor networks evolution between attractor states endow these networks with powerful computational properties. Some attractor networks are useful models of neural circuits. It would be helpful to have a little knowledge of neuroscience and dynamical systems — fortunately for you my previous posts cover those topics: for an introduction to dynamical systems, you can read Part I, for an introduction to synapses, you can read Part II. You can probably get by without it though.
The Disunity of Consciousness in Everyday Experience
Is there a structural, cognitive-architecture argument that our experiences are generally unified? Maybe yes. But only under some highly specific theoretical assumptions. For example, if you subscribe to a global workspace theory, according to which cognitive processes are conscious if and only if they are shared to a functional workspace that is accessible to a wide range of downstream cognitive processes and if you hold that this workspace normally unifies whatever is being processed into a single representational whole, then you have a structural argument for the unity of consciousness. Alternatively, you might accept a higher-order theory of consciousness and hold that in ordinary cognition the relevant higher-order representation is generally a single representation with complex conjoined contents (e.g., “visual and tactile and philosophical-thought processes are all going on”). But it’s not clear why we should accept such views – especially the part after the “and” in my characterizations. (For example, David Rosenthal’s higher-order account of phenomenal unity is different and more complicated.)
I’m inclined to think, in fact, that the balance of structural considerations tilt against unity. Our various cognitive processes run to a substantial extent independently. They influence each other, but they aren’t tightly integrated. Arguably, this is true even for conscious processes, such as thoughts of philosophy and visual experiences of a road. Even on relatively thin or sparse views of consciousness, on which only one or a few modalities can be conscious in a moment, this is probably true; but it seems proportionately more plausible the richer and more abundant conscious experience is. Suppose we have constant tactile experience of our feet in our shoes, constant auditory experience of the background noises in our environment, constant proprioceptive experience of the position of our body, constant experience of our levels of hunger, sleepiness/energy, our emotional experiences, our cognitive experiences and inner speech, etc. – a dozen or more very different phenomenal types all at once. You adventurously outrun the currently available evidence of cognitive psychology if you suggest that there’s also constantly some unifying cognitive process that stitches this multitude together into a cognitive unity. This isn’t to deny that modalities sometimes cooperate tightly (e.g., the McGurk effect). But to treat tight integration as the standard condition of all aspects of experience all the time is a much stronger claim. Sensorimotor integration among modalities is common and important, yes. But overall, the human mind is loosely strung together.
Schizophrenia Is the Price We Pay for Minds Poised Near the Edge of a Cliff
One proposed solution is to think of schizophrenia in terms of cliff-edged fitness functions, a hypothesis originally applied to schizophrenia by Randolph Nesse in 2004, and subsequently formalized by Philipp Mitteroecker and Giuseppe Pierpaolo Merola. As Nesse describes it succinctly, “I recognized that traits with an asymmetric fitness function are special; they give greater and greater advantages up to a point where the system fails and fitness falls off a cliff edge.”
In the case of schizophrenia, it can be hypothesized that certain cognitive, linguistic, or social traits increase fitness and are positively selected by evolution but lead to catastrophic consequences when expressed beyond a critical threshold. The model envisions a continuous, heritable trait that enhances reproductive fitness up to a point, beyond which it sharply decreases. Natural selection stabilizes the trait just below the cliff, but a small minority overshoot and suffer a dramatic fitness loss.
Transhumanism in a Technofeudal Society
The novel’s main character Siri is a “synthesist”. Siri has the remarkable ability to translate ideas without having to actually understand them himself. Instead, he prompts his interlocutors to speak, and rather than attempting to process the words directly, he observes their “topology”, the connections between the speakers language, their body movements and their interactions with their environment. Keaton describes it like understanding the shape of an underwater object by observing the waves it creates on the surface. Others compare it to the “Chinese Room” thought experiment.
I let this concept sit with me for a while. Pondering what it must be like to process information without understanding. Then I realised, he was just describing surveillance capitalism.*
The systems which observe us do not understand us. They don’t need to. Targets Customer Loyalty Program does not know what a baby is, but, it knows when a particular customer will need to buy baby stuff, even before the customer. The system doesn’t know the customer is pregnant. It doesn’t know what pregnant is. It doesn’t need to know. It just sees the connections in the customers purchase history, recognises the pattern and acts on it.
The experience continues until you stop experiencing it.
- This is a very interesting ad.
Research
Object personification in autism: This paper will be very sad if you don’t read it
Object personification is the attribution of human characteristics to non-human agents. In online forums, autistic individuals commonly report experiencing this phenomenon. Given that approximately half of all autistic individuals experience difficulties identifying their own emotions, the suggestion that object personification may be a feature of autism seems almost paradoxical. Why would a person experience sympathy for objects, when they struggle to understand and verbalise the emotions of other people as well as their own? An online survey was used to assess tendency for personification in 87 autistic and 263 non-autistic adults. Together, our results indicate that object personification occurs commonly among autistic individuals, and perhaps more often (and later in life) than in the general population. Given that in many cases, autistic people report their personification experiences as distressing, it is important to consider the reasons for the increased personification and identify structures for support.
Experimental and Theoretical Approaches to Conscious Processing
Recent experimental studies and theoretical models have begun to address the challenge of establishing a causal link between subjective conscious experience and measurable neuronal activity. The present review focuses on the well-delimited issue of how an external or internal piece of information goes beyond nonconscious processing and gains access to conscious processing, a transition characterized by the existence of a reportable subjective experience. Converging neuroimaging and neurophysiological data, acquired during minimal experimental contrasts between conscious and nonconscious processing, point to objective neural measures of conscious access: late amplification of relevant sensory activity, long-distance cortico-cortical synchronization at beta and gamma frequencies, and “ignition” of a large-scale prefronto-parietal network. We compare these findings to current theoretical models of conscious processing, including the Global Neuronal Workspace (GNW) model according to which conscious access occurs when incoming information is made globally available to multiple brain systems through a network of neurons with long-range axons densely distributed in prefrontal, parieto-temporal, and cingulate cortices. The clinical implications of these results for general anesthesia, coma, vegetative state, and schizophrenia are discussed.
Deep temporal models and active inference
How do we navigate a deeply structured world? Why are you reading this sentence first – and did you actually look at the fifth word? This review offers some answers by appealing to active inference based on deep temporal models. It builds on previous formulations of active inference to simulate behavioural and electrophysiological responses under hierarchical generative models of state transitions. Inverting these models corresponds to sequential inference, such that the state at any hierarchical level entails a sequence of transitions in the level below. The deep temporal aspect of these models means that evidence is accumulated over nested time scales, enabling inferences about narratives (i.e., temporal scenes). We illustrate this behaviour with Bayesian belief updating – and neuronal process theories – to simulate the epistemic foraging seen in reading. These simulations reproduce perisaccadic delay period activity and local field potentials seen empirically. Finally, we exploit the deep structure of these models to simulate responses to local (e.g., font type) and global (e.g., semantic) violations; reproducing mismatch negativity and P300 responses respectively.
The dynamics of production and destruction: Analytic insight into complex behavior
This paper analytically explores the properties of simple differential-difference equations that represent dynamic processes with feedback dependent on prior states of the system. Systems with pure negative and positive feedback are examined, as well as those with mixed (positive/negative) feedback characteristics. Very complex time dependent behaviors may arise from these processes. Indeed, the same mechanism may, depending on system parameters and initial conditions, produce simple, regular, repetitive patterns and completely irregular random-like fluctuations.
For the differential-delay equations considered here we prove the existence of: (i) stable and unstable limit cycles, where the stable cycles may have an arbitrary number of extrema per period; and (ii) chaos, meaning the presence of infinitely many periodic solutions of different period and of infinitely many irregular and mixing solutions.
Neuronal Oscillations in Cortical Networks
Clocks tick, bridges and skyscrapers vibrate, neuronal networks oscillate. Are neuronal oscillations an inevitable by-product, similar to bridge vibrations, or an essential part of the brain’s design? Mammalian cortical neurons form behavior-dependent oscillating networks of various sizes, which span five orders of magnitude in frequency. These oscillations are phylogenetically preserved, suggesting that they are functionally relevant. Recent findings indicate that network oscillations bias input selection, temporally link neurons into assemblies, and facilitate synaptic plasticity, mechanisms that cooperatively support temporal representation and long-term consolidation of information.
The chemical basis of morphogenesis
It is suggested that a system of chemical substances, called morphogens, reacting together and diffusing through a tissue, is adequate to account for the main phenomena of morphogenesis. Such a system, although it may originally be quite homogeneous, may later develop a pattern or structure due to an instability of the homogeneous equilibrium, which is triggered off by random disturbances. Such reaction-diffusion systems are considered in some detail in the case of an isolated ring of cells, a mathematically convenient, though biologically unusual system. The investigation is chiefly concerned with the onset of instability. It is found that there are six essentially different forms which this may take. In the most interesting form stationary waves appear on the ring. It is suggested that this might account, for instance, for the tentacle patterns on Hydra and for whorled leaves. A system of reactions and diffusion on a sphere is also considered. Such a system appears to account for gastrulation. Another reaction system in two dimensions gives rise to patterns reminiscent of dappling. It is also suggested that stationary waves in two dimensions could account for the phenomena of phyllotaxis. The purpose of this paper is to discuss a possible mechanism by which the genes of a zygote may determine the anatomical structure of the resulting organism. The theory does not make any new hypotheses; it merely suggests that certain well-known physical laws are sufficient to account for many of the facts. The full understanding of the paper requires a good knowledge of mathematics, some biology, and some elementary chemistry. Since readers cannot be expected to be experts in all of these subjects, a number of elementary facts are explained, which can be found in text-books, but whose omission would make the paper difficult reading.
Central pattern generators and the control of rhythmic movements
Central pattern generators are neuronal circuits that when activated can produce rhythmic motor patterns such as walking, breathing, flying, and swimming in the absence of sensory or descending inputs that carry specific timing information. General principles of the organization of these circuits and their control by higher brain centers have come from the study of smaller circuits found in invertebrates. Recent work on vertebrates highlights the importance of neuro-modulatory control pathways in enabling spinal cord and brain stem circuits to generate meaningful motor patterns. Because rhythmic motor patterns are easily quantified and studied, central pattern generators will provide important testing grounds for understanding the effects of numerous genetic mutations on behavior. Moreover, further understanding of the modulation of spinal cord circuitry used in rhythmic behaviors should facilitate the development of new treatments to enhance recovery after spinal cord damage.
The Kuramoto model: A simple paradigm for synchronization phenomena
Synchronization phenomena in large populations of interacting elements are the subject of intense research efforts in physical, biological, chemical, and social systems. A successful approach to the problem of synchronization consists of modeling each member of the population as a phase oscillator. In this review, synchronization is analyzed in one of the most representative models of coupled phase oscillators, the Kuramoto model. A rigorous mathematical treatment, specific numerical methods, and many variations and extensions of the original model that have appeared in the last few years are presented. Relevant applications of the model in different contexts are also included.
Steps Toward Artificial Intelligence
The problems of heuristic programming-of making computers solve really difficult problems-are divided into five main areas: Search, Pattern-Recognition, Learning, Planning, and Induction.
A computer can do, in a sense, only what it is told to do. But even when we do not know how to solve a certain problem, we may program a machine (computer) to Search through some large space of solution attempts. Unfortunately, this usually leads to an enormously inefficient process. With Pattern-Recognition techniques, efficiency can often be improved, by restricting the application of the machine’s methods to appropriate problems. Pattern-Recognition, together with Learning, can be used to exploit generalizations based on accumulated experience, further reducing search. By analyzing the situation, using Planning methods, we may obtain a fundamental improvement by replacing the given search with a much smaller, more appropriate exploration. To manage broad classes of problems, machines will need to construct models of their environments, using some scheme for Induction.
Wherever appropriate, the discussion is supported by extensive citation of the literature and by descriptions of a few of the most successful heuristic (problem-solving) programs constructed to date
Despite their apparently divergent accounts of higher cognition, cognitive theories based on neural computation and those employing symbolic computation can in fact strengthen one another. To substantiate this controversial claim, this landmark work develops in depth a cognitive architecture based in neural computation but supporting formally explicit higher-level symbolic descriptions, including new grammar formalisms. Detailed studies in both phonology and syntax provide arguments that these grammatical theories and their neural network realizations enable deeper explanations of early acquisition, processing difficulty, cross-linguistic typology, and the possibility of genomically encoding universal principles of grammar. Foundational questions concerning the explanatory status of symbols for central problems such as the unbounded productivity of higher cognition are also given proper treatment. The work is made accessible to scholars in different fields of cognitive science through tutorial chapters and numerous expository boxes providing background material from several disciplines. Examples common to different chapters facilitate the transition from more basic to more sophisticated treatments. Details of method, formalism, and foundation are presented in later chapters, offering a wealth of new results to specialists in psycholinguistics, language acquisition, theoretical linguistics, computational linguistics, computational neuroscience, connectionist modeling, and philosophy of mind.
The metabolic costs of cognition
Cognition and behavior are emergent properties of brain systems that seek to maximize complex and adaptive behaviors while minimizing energy utilization. Different species reconcile this trade-off in different ways, but in humans the outcome is biased towards complex behaviors and hence relatively high energy use. However, even in energy-intensive brains, numerous parsimonious processes operate to optimize energy use. We review how this balance manifests in both homeostatic processes and task-associated cognition. We also consider the perturbations and disruptions of metabolism in neurocognitive diseases.
Our brains, as well as those of many mammals, are capable of efficiently controlling organisms with low resource requirements. We wish to reproduce these capabilities in computer software.Instead of blindly guessing as to what kind of algorithms the brain might use, we try to operate within some biological limitations in order to ensure that we are exploring in the right areas. While we don’t need to reproduce biological algorithms exactly, we wish to capture the essence of the algorithms the brain implements and reduce them to programs with low complexity that can exhibit similar properties.
Sparse Predictive Hierarchies (SPH) are a general model of how we think the human brain functions. It can perform tasks involving sequence prediction, world model building, and reinforcement learning. It is based on online and incremental learning algorithms developed at Ogma Corp. SPH is: • Sparse: It uses only small fractions of the available resources at a time. • Predictive: It models the world as sequences of events, where previous events are causally related to subsequent events. • Hierarchies: It consists of several layers of abstraction, both spatially and temporally.
While SPH is “deep” in the sense that it uses these multiple layers of abstraction in a hierarchy, we do not consider it to be “Deep Learning” (DL). Deep learning is not biologically plausible, mostly due to continuous, dense (not sparse) representations and the use of the backpropagation algorithm
How To Build Conscious Machines
I must know what it is I want to build before I can build it. I want to build a mind, so that means I have to take concrete positions on disputed issues within philosophy of mind, psychology, cognitive science and neuroscience. The following is a survey of some relevant material from those fields. It is based on the introductory sections of my publications on enactive and ethical AI, communication and consciousness. Topics covered include the mind body problem functionalism, the “hard problem” of consciousness, various theories of consciousness, self-organisation and the free energy principle, enactivism, epistemology, semiotics, structuralism, post-structuralism and theories of meaning. Though this is a very broad ranging survey, I try to tie these concepts together into a coherent, sequential story from beginning to end
Tools
A web extension that redirects YouTube, Instagram, Reddit, TikTok and other websites to alternative privacy-friendly frontends.
LibreChat brings together the future of assistant AIs with the revolutionary technology of OpenAI’s ChatGPT. Celebrating the original styling, LibreChat gives you the ability to integrate multiple AI models. It also integrates and enhances original client features such as conversation and message search, prompt templates and plugins.
A self-hosted BitTorrent indexer, DHT crawler, content classifier and torrent search engine with web UI, GraphQL API and Servarr stack integration.
Joplin - the privacy-focused note taking app with sync capabilities for Windows, macOS, Linux, Android and iOS.
News
- Introducing the V-JEPA 2 world model and new benchmarks for physical reasoning
Today, we’re excited to share V-JEPA 2, the first world model trained on video that enables state-of-the-art understanding and prediction, as well as zero-shot planning and robot control in new environments. As we work toward our goal of achieving advanced machine intelligence (AMI), it will be important that we have AI systems that can learn about the world as humans do, plan how to execute unfamiliar tasks, and efficiently adapt to the ever-changing world around us.
V-JEPA 2 is a 1.2 billion-parameter model that was built using Meta Joint Embedding Predictive Architecture (JEPA), which we first shared in 2022. Our previous work has shown that JEPA performs well for modalities like images and 3D point clouds. Building on V-JEPA, our first model trained on video that we released last year, V-JEPA 2 improves action prediction and world modeling capabilities that enable robots to interact with unfamiliar objects and environments to complete a task. We’re also sharing three new benchmarks to help the research community evaluate how well their existing models learn and reason about the world using video. By sharing this work, we aim to give researchers and developers access to the best models and benchmarks to help accelerate research and progress—ultimately leading to better and more capable AI systems that will help enhance people’s lives.