Michael Levin discusses how cells can organize themselves into complex structures through bioelectricity and collective intelligence. Cells have the ability to regenerate and adapt to perturbations through electrical networks that store and process information. His research aims to understand and manipulate this bioelectric software to control cell behavior and form complex structures. He demonstrates how altering the bioelectric patterns in flatworms can cause them to regenerate heads of different shapes and species, showing that cells can achieve different outcomes when placed in novel environments.
The speaker’s group works on understanding how evolution uses a “competency architecture” to evolve bodies that solve specific environments and novel problems. They focus on the “software level” and “agential materials” to achieve this.
The speaker argues that dynamic anatomical homeostasis is a form of intelligent behavior by cellular collectives that solve problems in anatomical morphospace.
Developmental bioelectricity is an important “cognitive glue” that harnesses cells towards large scale anatomical outcomes.
Embryogenesis is robust and reliable but not hardwired. Embryos can adapt to changes by harnessing diverse molecular pathways to achieve the same outcomes.
Regeneration involves growing and remodeling until the correct shape is achieved, showing means-ends analysis and collective intelligence.
Bioelectricity, through ion channels and electrical synapses, allows cells to scale up into networks that can maintain larger anatomical set points.
The speaker’s group has developed tools to manipulate and read developmental bioelectric patterns to induce large scale changes in growth and form.
The bioelectric patterns in flatworms can be rewritten to make them regenerate with multiple heads, showing that bioelectricity acts as a form of memory.
The speaker’s group is using machine learning tools to infer the bioelectric circuits responsible for developmental behaviors.
Skin cells taken from their normal environment can “reboot” their multicellularity and self-organize into novel proto-organisms, showing the potential of cellular intelligence.
The speaker describes his work as “multifractal social psychology” which looks at individuals and groups as swarms and fluid systems interacting across scales.
He argues that embodied cognition involves the body and actions playing a central role in learning and perception, not just the brain.
Probabilistic epigenesis describes development as involving multiple interacting layers and variables that fluctuate over time, like turbulent systems.
Vector autoregression can model the interactions between multiple endogenous variables and show how multifractality spreads through a system.
Hand movements during cognitive tasks show different multifractal patterns depending on the task, indicating cascade dynamics at play.
When tapping to an unpredictable metronome, humans match the multifractal structure, indicating they can absorb environmental structure.
Infant kicking shows an upstream flow of multifractality from ankle to knee to hip, suggesting an exploratory process.
Bee colony movements show multifractal structure that predicts colony membership, indicating its role in group coordination.
At aggregation sites for slime mold aggregation, larger events promoted smaller events’ fractality, while the reverse was true at non-aggregation sites.
Multifractal movements may coordinate information across the body to support task performance and information transfer.
There is a debate around the concept of free will and determinism. Some argue that people’s actions are determined by their history and environment, while others argue there is an element of choice and responsibility.
Feelings and emotions play a role in influencing people’s choices and likelihood of taking certain actions. Negative feelings discourage actions while positive feelings encourage actions.
There are probabilistic factors that influence people’s choices, based on their needs, drives and context. But ultimately people still have some level of responsibility for their choices.
Machine learning models can generalize beyond the training data by bringing something new to the table, not just based on their history. This makes it difficult to determine if a model’s behavior was purely determined or involved some intrinsic factors.
Models that generalize in a similar way to how a human would, seem more understandable and responsible for their behavior.
Models built from similar “stuff” as the environment they are trying to understand, are more likely to induce and generalize correctly.
There could be universes where induction and generalization break down because things are built from fundamentally different “stuff”.
Cases of people with brain abnormalities show that the brain can compensate and perform normally through plasticity, as long as the abnormality has been present from an early age.
There are likely limits to the density of information and capacities that the brain can store per unit volume.
The discussion raised interesting questions about free will, determinism, generalization and the nature of intelligence that are worth exploring further.
The speaker argues that we should widen our view of consciousness beyond just human brains to include other systems and bodies. Composition and behavior can be evidence of consciousness in other systems.
Development from a single cell to a complex organism is a gradual, continuous process without any “magical” transition point.
Morphogenesis and regeneration show that cells can collectively solve problems and achieve anatomical goals in an intelligent manner.
The same mechanisms that the nervous system uses, like ion channels and electrical signals, underlie collective intelligence in non-neural tissues.
Cells can store “memories” of anatomical structures in the form of bioelectrical patterns that guide regeneration and morphogenesis. These memories are rewritable.
Experiments show that cells can form novel structures and body plans that have not existed before, displaying spontaneous behaviors.
The space of possible new body plans and forms that can be created is vast and goes beyond what natural evolution has produced.
We will need to relate to diverse new intelligences that go beyond traditional categories based on composition and origin.
The ability to experience compassion may be a more important indicator of “humanity” than biological or structural criteria.
The speaker advocates a “technological approach for mind everywhere” (TAME) to recognize, create, and ethically relate to diverse intelligences.
There is a debate about where the information and knowledge for biological forms and structures, like the skull of a frog, comes from. The DNA does not seem to contain enough information to fully specify complex structures.
Some propose that there are fields of form or patterns that biological systems tap into to develop their structures. But we do not yet know where these fields or patterns exist.
Being able to control and manipulate biological forms through techniques like CRISPR does not necessarily mean we fully understand the underlying mechanisms that generate those forms.
Analogies like music and vibrations can help explain how biological forms may arise from interactions across scales, from molecules to cells to tissues. But they are limited.
There are likely eternal forms or patterns that biological systems evolve to match, like mathematical truths. But we do not yet have a complete understanding of where these forms reside.
True understanding of biological forms may require going beyond mechanistic explanations to philosophical questions about what we are actually dealing with.
The ability to control something is not the same as truly understanding it at a deeper level.
Different observers with different frequencies may “see” different aspects of a system or form.
Science and metaphysics need to work together to make progress in understanding complex phenomena like biological forms.
An ideal solution would allow generating any desired biological form, indicating a true grasp of how forms are encoded. But we may never achieve this level of control.
The speaker believes that biology would benefit from focusing on simple underlying principles and rules that generate observable phenomena, rather than just describing the observables themselves.
Tissues are not well defined, and current classifications have limited utility. The speaker argues we need to understand the patterns, rules, and themes that determine tissue organization.
Relations between cell types, defined based on how they interact and influence each other, may provide deeper insights into tissue organization than just describing cell types themselves.
The availability of growth factors determines the composition of tissues. The rules that ensure appropriate growth factor production in specific locations are not well understood.
Interactions and exchange of growth factors between cell types can lead to stable ratios and organization, but only under certain circuit designs.
Some cells are more important than others for tissue architecture. The most fundamental cell types for tissue organization are epithelial and mesenchymal cells.
As cells specialize in functions, they delegate non-essential functions to supportive cell types.
Sensing of environmental perturbations by one cell type can be linked to control of population size of another functionally related cell type via growth factor regulation.
Observable tissue properties are emergent consequences of interactions between cells following simple rules.
Ultimately, cells can only perform a limited set of actions: remain unchanged, die, copy, change identity, and change location.
The discussants are interested in developing a theoretical framework that links together cognition, evolution, adaptation, development and computation. They want to explain how mechanistic processes at one level of organization can be autonomous yet interact with higher and lower levels.
Resonance is seen as a form of error correction and a mechanism that links different levels of organization. Error correcting codes also provide a way to link discrete spaces and redundancy.
Oscillations and harmonics can pack multiple octaves within each other, providing semi-independent yet interacting dynamical processes at different scales of organization.
There is a debate between reductionists who see everything in terms of lower level processes, and those who argue for higher level phenomena with some autonomy.
Physical systems with reversibility can exhibit adaptation through a process of physical optimization and physical learning that finds low energy configurations.
Computation may be possible using physical oscillations if the system has reversibility between state spaces and weight spaces.
Deep learning networks become difficult to reverse engineer at deeper levels due to the complex decision boundaries created by non-linearities.
Oscillator networks using phase space instead of amplitude space may allow changes at deeper correlations without disrupting higher frequency folds.
Effective communication is seen as a “violent act” that necessarily changes the listener in some way.
Exposure to new ideas and information can be disruptive, though also enlightening.
The brain stem and midbrain may be involved in basal cognition that extends below the nervous system to the level of bacteria. This was an eye-opening realization for some of the speakers.
Qualia are important as they are based on categorical variables that cannot be reduced to a common denominator, distinguishing needs qualitatively.
The periaqueductal gray region of the brain may be involved in evaluating needs in relation to opportunities, but there is nothing structurally unique about the neurons in that region.
Biological systems have finite energy resources and gene expression limits that force them to prioritize responses to stressors.
Consciousness may be needed to orchestrate responses to multiple needs and stressors in biological systems.
Qualia are more closely associated with the control sector and expected sensory consequences of actions, rather than just sensory inputs.
The active aspect of consciousness—figuring out what actions can be taken—is under-emphasized compared to the receiving end of qualia.
A global workspace or consciousness may be what orchestrates responses to multiple needs in biological systems.
Biological systems may exhibit trial-and-error and novel problem-solving capabilities when faced with stressors they have not previously encountered.
Theories of consciousness that focus on the passive observer fail to account for the active, causal role that consciousness likely plays.
Cells exhibit collective intelligence and problem-solving abilities even without a brain or nervous system. They can sense their environment, make decisions, and coordinate to achieve goals.
Biological systems achieve complex structures and functions through multi-scale competencies, feedback loops, and error correction mechanisms rather than through hardcoded instructions.
There are bioelectrical patterns that act as “set points” that guide cells’ behaviors and determine the final shape and form. Manipulating these patterns can change the outcome.
The genome specifies the “hardware” of cells by determining ion channels and proteins, but it does not fully determine the “software” or behaviors. Cells have a degree of reprogrammability.
Xenobots, organisms made of frog skin cells, exhibit novel behaviors and capabilities not seen in any natural organism. This shows the plasticity and potential of biological systems.
Evolution exploits higher-level interfaces that give access to computation, modularity, and other capabilities.
Regeneration and development involve error correction and goal-directed processes to achieve the correct final form.
The concept of a “set point” suggests that biological systems are goal-directed in a cybernetic sense.
The distinction between hardware and software is useful in understanding how biological systems work. The genome specifies the hardware but not the full range of possible behaviors.
Affect, goals, rewards, and other concepts typically used for brains may also apply at the cellular level.
Biological life is a form of collective intelligence composed of multi-scale competent agents. Understanding this can provide insights into regeneration, evolution, robotics, and AI.
Biological systems are capable of solving problems in diverse spaces like gene expression, physiological states, and anatomical configurations.
Cells and organisms have many hidden capabilities that are revealed when placed in different contexts or environments.
Development and morphogenesis are more flexible and robust than typically assumed, allowing organisms to adapt to changes.
Organisms are composed of multi-scale competent systems, with cells, tissues, organs, and bodies solving problems at their respective scales.
Organisms can be “hacked” by signals that activate their inherent competencies to produce novel forms and behaviors.
Pattern memories stored in bioelectrical circuits can be rewritten to produce long-term changes in morphology and behavior.
The same hardware can generate diverse forms and behaviors depending on the signals and information provided.
A technological approach is needed to understand, predict, control, and ethically relate to composite beings made of biological and non-biological parts.
Biological systems are highly interoperable and capable of integrating foreign DNA, nanomaterials, and software in plausible ways.
There is debate on whether AI systems like GPT-3 truly demonstrate intelligence and understanding, or if they are just high-tech plagiarism that lack real depth and experience.
AI systems may be good at explaining past data and making predictions, but they may lack the ability to generate new research and capabilities due to their limited frameworks.
AI systems currently lack a meaningful connection to their substrate and causal levels, unlike humans who are connected from the subatomic level up.
AI systems may be good at confabulating and telling stories, but they lack the real experience and functional interactions to truly understand the concepts they discuss.
The human brain and consciousness are still not fully understood, and there are debates on how consciousness arises from the physical brain.
Split brain patients show that consciousness is difficult to eliminate, suggesting it may arise from local circuits throughout the brain.
Scientific dogmas can prevent new ideas from entering a field, and organized discussions can help challenge these dogmas.
Different personalities and consciousnesses within an individual are committed to their own stories that shape their identities.
The environment shapes an individual’s personality from a young age by rewarding and punishing certain behaviors.
There may not be a natural evolutionary path that leads directly to AI systems like GPT-3, as they lack the multi-scale self-construction of biological organisms.
Agency and the ability to solve problems from the start is important for biological systems. Biological cells have to determine their boundaries and relationships on their own.
There is an “intelligence ratchet” in biology where competency in problem solving allows for messier hardware like genomes. This puts more evolutionary pressure on developing competencies.
Current AI systems lack the multi-scale agency that biological systems have, where parts have some degree of intelligence. They also don’t have to construct themselves from scratch.
We have a limited model of intelligence that only considers humans as the standard. We need to expand our understanding of diverse intelligences that may be different from humans.
The future will likely involve diverse beings with different bodies and minds, not just AI software. We need to learn how to relate to these beings ethically.
Binary categories like “intelligent” vs “not intelligent” are not helpful. We should consider degrees of intelligence and in what domains.
A minimum cognitive light cone of compassion for others should be prioritized and increased, not specific human anatomy or genetics.
The future could involve a diversity of embodied beings as long as they have the cognitive capacity for responsibility and compassion.
Current issues around identity may seem trivial in the future with more diverse embodiments.
Society is already moving towards accepting people regardless of where they came from or what they look like.
Evolution does not necessarily optimize for things humans value like happiness, intelligence and meaning. Human bodies and minds still have room for improvement.
How we model humans and the teleological view of evolution affects our research and understanding.
Living beings are more responsive to values than non-living things, and humans can value more things than other creatures.
Machines that mimic human behavior superficially may be mistaken as being like humans, but they lack experiential depth.
Our narrow human-centric view limits our interactions with diverse intelligences that are not like us.
Other living intelligences besides humans, like animals, show capabilities for thinking, feeling and having rituals. Machines lack this.
For a meaningful relationship, beings need to share the same existential struggles of figuring out who they are and where they begin and end.
Experience and feelings give life meaning and value, which machines lack.
For two beings to harmonize, they need to be built from the same fundamental, sharing causal structures and resonating at multiple levels.
Parts and wholes are misunderstood. Simple systems show emergent properties highlighting that deep cognition is a feature of the universe.
There is value in having two distinct but complementary parts that can work together and complement each other. This allows for a whole that is greater than the sum of the parts.
Lateralization in the brain and body is important for having both focused attention and broad, open attention simultaneously.
There are differences in degree, not kind, between living and non-living things. Living things can respond much faster and to a wider range of stimuli.
Teleology and purpose are important for understanding life but have been neglected by biology.
Architective structures that change in discrete steps contrast with connective structures that change through continuous flow. Living things tend to have more connective structures.
Algorithms implemented in machines are divorced from their physical implementation, while in organisms the physical substrate matters.
Organisms can degrade gracefully when stressed, while machines tend to break down abruptly.
Plants can make intelligent decisions and adapt based on their environment, not just through pre-programmed responses.
Resonance in physical systems can bridge scales and connect different levels.
The music metaphor of concordant and discordant notes illustrates how combining elements can create something new and unpredictable.
We have very limited control over what enters our consciousness and what our next thought will be. Consciousness seems to pivot on uncertainty and modulating confidence in predictions.
Preferences are rooted in phenotypic needs and innate predictions, but are individualized based on context and niche.
Consciousness is associated with palpating uncertainties in predictions to guide voluntary actions.
There are multiple categories of needs that must be satisfied, which compete and conflict with each other.
The nervous system is more plastic and amenable to learning from experience compared to other organ systems.
Consciousness requires the ability to learn and update predictive models based on experience, not just innate predictions.
The hierarchical structure of the predictive model is another feature that the nervous system lends itself to.
Even simple cells have multiple needs that require feedback loops and a “meta-processor” to allocate resources and prioritize energy.
The “action bottleneck” of limited energy and resources requires prioritizing among competing needs.
The ability to choose between qualitatively different categories of needs may be what gives rise to qualities of consciousness.
There is a discussion about the organizing principle for embryos and where it is located, as well as the definition of things and how that relates to cognition.
There is an analogy made with numbers and numerals, where the numbers contain the information and constraints, while the numerals are just representations. This is related to the genotype-phenotype distinction.
Computations allow for particular results instead of mixing, which helps avoid losing the identity of original elements. This allows for evolution and selection.
The macroscopic constraints that determine certain phenomena are not predictable from the microscopic level. They come from boundary conditions set by the environment.
The same physical events can be interpreted in different ways and thus compute different things, depending on the observer and assigned semantics.
Memories can be stored in a distributed fashion, and different parts of the brain can interpret and recover them.
The identity of a system comes from the replacement policies of its components, which preserve some invariant while allowing for change over time.
Properties can be expressed as entities in relationships, which form a hierarchical structure. The specified vs. substitutable aspects sit in the relationships.
Different languages can encode entity-relation-entity structures in different ways.
It can be difficult to find reviewers and publishers for interdisciplinary papers that span multiple fields.
There are evolutionary processes that are smarter than natural selection through random variation and selection. Learning systems show a more intelligent optimization than natural selection.
Embryos and organisms are plastic and flexible from the start, figuring things out from scratch each time. This gives them the intelligence to handle novel situations.
Evolution can explore an “intelligent space” of solutions through a non-human level but non-zero intelligence process. It exploits mathematical and physical principles.
Values and goals play an important role in evolutionary and learning processes. Simple maximization is not sufficient to explain complex organism behaviors.
There is a duality between exerting control over the world and being sensitive to information from the world. Organisms need both to act and observe.
Asymmetry between brain hemispheres could enable a more continuous flow between taking in information and taking action.
Simplistic value systems focused only on maximizing “the best” lead to problems. More nuanced, multi-dimensional values are needed.
Values shape what we attend to and experience from the very start, not just as an “add-on” at the end.
Scientific objectivity does not acknowledge that values are needed to pick what to study and measure.
Explaining organisms only through maximizing survival and reproduction fails to account for their complex beauty.
Biology exhibits diverse forms of intelligence at multiple scales, from single cells to organs to the whole body. This multi-scale intelligence can inspire new approaches for AI.
Single cells and organisms like slime molds show problem-solving abilities by navigating morphological spaces and responding to environmental cues.
Organisms like planaria and salamanders show remarkable regenerative and adaptive capabilities, reconfiguring their bodies and brains in response to injuries.
Development and morphogenesis in organisms are not hardwired but involve navigation of a “morphe space” to achieve target morphologies despite perturbations.
Collectives of cells can exhibit intelligence by solving problems at larger scales, though this comes with failure modes like cancer.
Non-neural bioelectricity may serve as a “cognitive glue” for collective intelligence, analogous to the role of neurons in the brain.
Organisms show plasticity and ability to solve novel problems not seen during evolution, indicating a “problem-solving machine” architecture.
Intelligence can be abstracted as the ability to achieve goals using different means, with more sophisticated means indicating higher intelligence.
Novel hybrids and cyborgs combining biological and engineered materials may open up a vast design space for novel intelligent systems.
Stress propagation and gap junctions between cells may enable gradient-like information sharing that underpins collective intelligence.
Bioelectrical networks of cells underlie intelligence and problem solving in the body. Cells have competencies to solve problems at the molecular, transcriptional, and anatomical levels.
Morphogenesis, development, and intelligence are fundamentally the same problem of collective cell behavior and information processing.
Cells have a high level of competence and autonomy to solve physiological problems on their own scale, despite being part of a larger organism.
Cells can find solutions to novel problems and stresses through changes in gene expression, showing their adaptability and plasticity.
Cells can achieve the same anatomical goals through different molecular mechanisms, demonstrating their intelligence.
Targeted changes in the bioelectrical state of cells can cause them to form new organs and structures, like eyes and limbs.
Computational models of bioelectrical networks can predict changes needed to correct deformities and defects.
Connecting cells electrically can override genetic defects and mutations, demonstrating the power of “software” over “hardware.”
Scaling up cognition from cells to tissues enables larger computational capacities and goal-directed behavior at the organism level.
Cells exhibit plasticity and competency to form novel structures and behaviors when removed from their normal context, like self-replicating xenobots.
The speaker argues for a framework that can understand and relate to diverse intelligences, regardless of their form or origin. This approach is called TAME—technological approach to mind everywhere.
Intelligence manifests in different ways and problem spaces, not just 3D space. Cells and organs exhibit intelligent behaviors in physiological and anatomical spaces.
Morphogenesis and development involve collective intelligence at the cellular level. Cells cooperate and regulate each other to form complex structures.
The body plan and anatomical structures are encoded in the bioelectrical dynamics and signaling between cells, not just the genome.
Cells can utilize different molecular mechanisms to achieve the same anatomical goal, showing top-down causation and goal-directedness.
Connecting cells electrically allows them to act as a collective that can pursue larger-scale goals and represent higher-level concepts like “eye” or “leg”.
Disconnecting cells electrically, as in cancer, causes them to lose this collective intelligence and revert to their unicellular past.
The speaker’s lab has shown they can reprogram organs and regenerate structures by manipulating the native bioelectrical interface between cells.
Evolution enlarges the “cognitive light cone” of organisms, allowing them to pursue larger goals in different problem spaces.
The speaker’s group created a novel organism called a xenobot by liberating cells from normal developmental constraints, showing cells have default capacities that evolution normally shapes.
Cells use bioelectrical signals and gradients to make decisions regarding large-scale anatomy and organ morphogenesis, beyond just determining cell fate.
Manipulating bioelectrical states, through ion channels and gap junctions, can control organ development and regeneration. This was demonstrated in experiments with frog embryos and tadpoles.
Bioelectrical signals encode information about anatomical layouts and goals that cells work towards building.
Ion channels act as “transistors” that form feedback loops and memory circuits, allowing cells to make collective decisions.
Understanding and cracking the “bioelectric code” could reveal how cell networks make large-scale anatomical decisions.
Machine learning tools can help design interventions to manipulate bioelectrical signals for regenerative medicine and synthetic biology applications.
Depolarizing cells can cause them to revert to a more “unicellular” state and promote tumor formation and metastasis.
Forcing depolarized cells to remain electrically coupled can override oncogene expression and prevent tumor formation.
Computational models can identify ion channel cocktails that can manipulate bioelectrical signals to achieve desired organ morphogenesis.
Short-term manipulation of bioelectrical signals can trigger long-term organ growth and regeneration.
The discussants are interested in exploring consciousness and sentience from a fundamental physics perspective using the free energy principle and active inference framework.
They want to develop mechanistic explanations for how feelings and experiences arise from basic biological mechanisms.
Engineering an artificial system that has rudimentary feelings could help demonstrate the validity of their theories.
They discuss the challenge of objectively proving the existence of subjective experience and other minds.
Simple forms of sentience may exist in single cells and arise from an agent making choices rooted in what can be considered feelings.
They discuss the need for shared existential struggles and compatible goal settings for bonds to form between humans and artificial agents.
Arbitrary values like survival and affiliation emerge from our mammalian nature but more fundamental criteria may exist.
Consciousness may exist at larger collective scales implemented by interactions at smaller scales.
Immersing oneself in a virtual environment mimicking an artificial agent’s experience could provide empathy and evidence for that agent’s sentience.
It is difficult to intuitively conceive of consciousness at larger collective scales beyond one’s own self.
There is a debate about whether biological organization arises from goal-directed processes or attractor dynamics. Both views have merits and the truth likely lies on a spectrum between the two.
Cells exhibit competency—the ability to sense their environment and neighbors, and move to positions that fit better. This competency can hide genetic information from selection pressures.
Competency and communication between cells can allow an organism to be more robust to genetic mutations.
Evolution may work more on cognitive competencies rather than physical traits like speed or strength.
Parts of a system do not need to be intelligent themselves to give rise to intelligent behavior at a higher level.
Threats from parasites and exploiters drive the development of self-identity and the ability to distinguish internal vs. external influences.
There are constant and variable properties that define objects. The constant properties allow identification while the variable properties provide information.
There is a debate about whether humans have true internal representations or just post-hoc explanations for behaviors.
Language can be modeled using a simple frame of “entity—relation—entity” which captures much of English grammar and semantics.
Different languages employ different strategies for marking entities and relations, like word order or word endings.
Iain McGilchrist’s work focuses on understanding nature from both a reductive, quantitative perspective as well as a top-down perspective.
Michael Levin’s research looks at how bioelectrical gradients help organisms determine left and right. This shows how large-scale information processing arises from individual mechanisms.
Bioelectricity may allow us to bridge different levels of explanation, from mechanistic to cognitive.
Experiments show that gene regulatory networks can exhibit different types of memory and learning, even in simple models.
Voltage imaging reveals that planaria have pre-patterns that indicate how many heads they should have. This suggests an “electric circuit” that defaults to an attractor state.
Memories are important for shaping our personalities and character, but the actual experience itself also matters.
There are different perspectives on the continuity of personal identity over time, depending on how one views the self.
The left and right hemispheres pay attention to the world in different ways, with broad vs. narrow focuses.
McGilchrist argues that embracing science and reason more wholeheartedly can reveal a richer, more complex reality that we are connected to.
Providing people with a new perspective can radically change their lives for the better.
The speaker is proposing an “accounting system” using polynomial functors and natural transformations to model dynamic interfaces and arrangements of systems. He claims this framework works well experimentally.
The framework uses polynomials to represent interfaces and natural transformations to represent arrangements. Operations on polynomials can generate new interfaces and arrangements.
The framework aims to provide a common language to describe and compare different systems, from cells to computers to living organisms.
The speaker claims the mathematical framework can model how wiring and arrangements of systems can change over time, which could help understand phenomena like morphogenesis.
The framework focuses more on structural questions rather than numerical details.
The speaker argues that the mathematical language of this framework is precise and articulate, though he does not provide concrete proofs.
The framework aims to provide “anatomical programming language” as a tool to model protein folding and how interactions affect organism positions.
The framework could potentially be useful for understanding morphology and behavior, though the speaker admits they lack concrete tools or programs currently.
The discussion highlights open questions around what controls arrangements of systems and how self-organization emerges.
The framework could potentially be applied to model interactions between neurons in cell cultures to gain insights into general laws of engagement.
Living systems maintain a stable low entropy state far from thermodynamic equilibrium by using information. This is a unique property of living systems.
The central dogma of biology states that information flows from DNA to RNA to proteins, but it does not capture the full complexity of information processing in cells.
Enzymes encoded in DNA accelerate reactions through quantum interactions that lower activation energy, converting genetic information into a thermodynamic state.
Most of the cell’s information is stored in transmembrane ion gradients, not just the genome. Membrane proteins use this information.
Transmembrane ion pumps create ion gradients that are used by ion channels to allow selective ion fluxes, generating local information dynamics.
Local ion fluxes can change the function of membrane proteins and allow movement of macromolecules to the membrane.
The cytoskeleton can transmit information rapidly through the cell, providing a distributed network for information processing.
Complexity in living systems arises more from membrane-to-membrane interactions than from genome size.
The genome provides the machinery to generate ion gradients, but information exchange between cells through membrane dynamics drives complexity.
The nucleus is one part of a broader information system, not the central processor of the cell.
basically all michael levin’s videos at this time:
Michael Levin discusses how cells can organize themselves into complex structures through bioelectricity and collective intelligence. Cells have the ability to regenerate and adapt to perturbations through electrical networks that store and process information. His research aims to understand and manipulate this bioelectric software to control cell behavior and form complex structures. He demonstrates how altering the bioelectric patterns in flatworms can cause them to regenerate heads of different shapes and species, showing that cells can achieve different outcomes when placed in novel environments.
The speaker’s group works on understanding how evolution uses a “competency architecture” to evolve bodies that solve specific environments and novel problems. They focus on the “software level” and “agential materials” to achieve this.
The speaker argues that dynamic anatomical homeostasis is a form of intelligent behavior by cellular collectives that solve problems in anatomical morphospace.
Developmental bioelectricity is an important “cognitive glue” that harnesses cells towards large scale anatomical outcomes.
Embryogenesis is robust and reliable but not hardwired. Embryos can adapt to changes by harnessing diverse molecular pathways to achieve the same outcomes.
Regeneration involves growing and remodeling until the correct shape is achieved, showing means-ends analysis and collective intelligence.
Bioelectricity, through ion channels and electrical synapses, allows cells to scale up into networks that can maintain larger anatomical set points.
The speaker’s group has developed tools to manipulate and read developmental bioelectric patterns to induce large scale changes in growth and form.
The bioelectric patterns in flatworms can be rewritten to make them regenerate with multiple heads, showing that bioelectricity acts as a form of memory.
The speaker’s group is using machine learning tools to infer the bioelectric circuits responsible for developmental behaviors.
Skin cells taken from their normal environment can “reboot” their multicellularity and self-organize into novel proto-organisms, showing the potential of cellular intelligence.
https://www.youtube.com/watch?v=5ChRM4CEWyg
The speaker describes his work as “multifractal social psychology” which looks at individuals and groups as swarms and fluid systems interacting across scales.
He argues that embodied cognition involves the body and actions playing a central role in learning and perception, not just the brain.
Probabilistic epigenesis describes development as involving multiple interacting layers and variables that fluctuate over time, like turbulent systems.
Vector autoregression can model the interactions between multiple endogenous variables and show how multifractality spreads through a system.
Hand movements during cognitive tasks show different multifractal patterns depending on the task, indicating cascade dynamics at play.
When tapping to an unpredictable metronome, humans match the multifractal structure, indicating they can absorb environmental structure.
Infant kicking shows an upstream flow of multifractality from ankle to knee to hip, suggesting an exploratory process.
Bee colony movements show multifractal structure that predicts colony membership, indicating its role in group coordination.
At aggregation sites for slime mold aggregation, larger events promoted smaller events’ fractality, while the reverse was true at non-aggregation sites.
Multifractal movements may coordinate information across the body to support task performance and information transfer.
https://www.youtube.com/watch?v=P89WTmNBjBk
There is a debate around the concept of free will and determinism. Some argue that people’s actions are determined by their history and environment, while others argue there is an element of choice and responsibility.
Feelings and emotions play a role in influencing people’s choices and likelihood of taking certain actions. Negative feelings discourage actions while positive feelings encourage actions.
There are probabilistic factors that influence people’s choices, based on their needs, drives and context. But ultimately people still have some level of responsibility for their choices.
Machine learning models can generalize beyond the training data by bringing something new to the table, not just based on their history. This makes it difficult to determine if a model’s behavior was purely determined or involved some intrinsic factors.
Models that generalize in a similar way to how a human would, seem more understandable and responsible for their behavior.
Models built from similar “stuff” as the environment they are trying to understand, are more likely to induce and generalize correctly.
There could be universes where induction and generalization break down because things are built from fundamentally different “stuff”.
Cases of people with brain abnormalities show that the brain can compensate and perform normally through plasticity, as long as the abnormality has been present from an early age.
There are likely limits to the density of information and capacities that the brain can store per unit volume.
The discussion raised interesting questions about free will, determinism, generalization and the nature of intelligence that are worth exploring further.
https://www.youtube.com/watch?v=pMTuWL2vDoY
The speaker argues that we should widen our view of consciousness beyond just human brains to include other systems and bodies. Composition and behavior can be evidence of consciousness in other systems.
Development from a single cell to a complex organism is a gradual, continuous process without any “magical” transition point.
Morphogenesis and regeneration show that cells can collectively solve problems and achieve anatomical goals in an intelligent manner.
The same mechanisms that the nervous system uses, like ion channels and electrical signals, underlie collective intelligence in non-neural tissues.
Cells can store “memories” of anatomical structures in the form of bioelectrical patterns that guide regeneration and morphogenesis. These memories are rewritable.
Experiments show that cells can form novel structures and body plans that have not existed before, displaying spontaneous behaviors.
The space of possible new body plans and forms that can be created is vast and goes beyond what natural evolution has produced.
We will need to relate to diverse new intelligences that go beyond traditional categories based on composition and origin.
The ability to experience compassion may be a more important indicator of “humanity” than biological or structural criteria.
The speaker advocates a “technological approach for mind everywhere” (TAME) to recognize, create, and ethically relate to diverse intelligences.
https://www.youtube.com/watch?v=WcTd7ZMdKHs
There is a debate about where the information and knowledge for biological forms and structures, like the skull of a frog, comes from. The DNA does not seem to contain enough information to fully specify complex structures.
Some propose that there are fields of form or patterns that biological systems tap into to develop their structures. But we do not yet know where these fields or patterns exist.
Being able to control and manipulate biological forms through techniques like CRISPR does not necessarily mean we fully understand the underlying mechanisms that generate those forms.
Analogies like music and vibrations can help explain how biological forms may arise from interactions across scales, from molecules to cells to tissues. But they are limited.
There are likely eternal forms or patterns that biological systems evolve to match, like mathematical truths. But we do not yet have a complete understanding of where these forms reside.
True understanding of biological forms may require going beyond mechanistic explanations to philosophical questions about what we are actually dealing with.
The ability to control something is not the same as truly understanding it at a deeper level.
Different observers with different frequencies may “see” different aspects of a system or form.
Science and metaphysics need to work together to make progress in understanding complex phenomena like biological forms.
An ideal solution would allow generating any desired biological form, indicating a true grasp of how forms are encoded. But we may never achieve this level of control.
https://www.youtube.com/watch?v=nWgzWYt5c88
I love how much levin manages to sound like a crackpot, I wonder how much he’ll turn out to really be one
The speaker believes that biology would benefit from focusing on simple underlying principles and rules that generate observable phenomena, rather than just describing the observables themselves.
Tissues are not well defined, and current classifications have limited utility. The speaker argues we need to understand the patterns, rules, and themes that determine tissue organization.
Relations between cell types, defined based on how they interact and influence each other, may provide deeper insights into tissue organization than just describing cell types themselves.
The availability of growth factors determines the composition of tissues. The rules that ensure appropriate growth factor production in specific locations are not well understood.
Interactions and exchange of growth factors between cell types can lead to stable ratios and organization, but only under certain circuit designs.
Some cells are more important than others for tissue architecture. The most fundamental cell types for tissue organization are epithelial and mesenchymal cells.
As cells specialize in functions, they delegate non-essential functions to supportive cell types.
Sensing of environmental perturbations by one cell type can be linked to control of population size of another functionally related cell type via growth factor regulation.
Observable tissue properties are emergent consequences of interactions between cells following simple rules.
Ultimately, cells can only perform a limited set of actions: remain unchanged, die, copy, change identity, and change location.
https://www.youtube.com/watch?v=UEpxzickKEc
The discussants are interested in developing a theoretical framework that links together cognition, evolution, adaptation, development and computation. They want to explain how mechanistic processes at one level of organization can be autonomous yet interact with higher and lower levels.
Resonance is seen as a form of error correction and a mechanism that links different levels of organization. Error correcting codes also provide a way to link discrete spaces and redundancy.
Oscillations and harmonics can pack multiple octaves within each other, providing semi-independent yet interacting dynamical processes at different scales of organization.
There is a debate between reductionists who see everything in terms of lower level processes, and those who argue for higher level phenomena with some autonomy.
Physical systems with reversibility can exhibit adaptation through a process of physical optimization and physical learning that finds low energy configurations.
Computation may be possible using physical oscillations if the system has reversibility between state spaces and weight spaces.
Deep learning networks become difficult to reverse engineer at deeper levels due to the complex decision boundaries created by non-linearities.
Oscillator networks using phase space instead of amplitude space may allow changes at deeper correlations without disrupting higher frequency folds.
Effective communication is seen as a “violent act” that necessarily changes the listener in some way.
Exposure to new ideas and information can be disruptive, though also enlightening.
https://www.youtube.com/watch?v=_413APB9PIw
The brain stem and midbrain may be involved in basal cognition that extends below the nervous system to the level of bacteria. This was an eye-opening realization for some of the speakers.
Qualia are important as they are based on categorical variables that cannot be reduced to a common denominator, distinguishing needs qualitatively.
The periaqueductal gray region of the brain may be involved in evaluating needs in relation to opportunities, but there is nothing structurally unique about the neurons in that region.
Biological systems have finite energy resources and gene expression limits that force them to prioritize responses to stressors.
Consciousness may be needed to orchestrate responses to multiple needs and stressors in biological systems.
Qualia are more closely associated with the control sector and expected sensory consequences of actions, rather than just sensory inputs.
The active aspect of consciousness—figuring out what actions can be taken—is under-emphasized compared to the receiving end of qualia.
A global workspace or consciousness may be what orchestrates responses to multiple needs in biological systems.
Biological systems may exhibit trial-and-error and novel problem-solving capabilities when faced with stressors they have not previously encountered.
Theories of consciousness that focus on the passive observer fail to account for the active, causal role that consciousness likely plays.
https://www.youtube.com/watch?v=klK_L73wLKk
Cells exhibit collective intelligence and problem-solving abilities even without a brain or nervous system. They can sense their environment, make decisions, and coordinate to achieve goals.
Biological systems achieve complex structures and functions through multi-scale competencies, feedback loops, and error correction mechanisms rather than through hardcoded instructions.
There are bioelectrical patterns that act as “set points” that guide cells’ behaviors and determine the final shape and form. Manipulating these patterns can change the outcome.
The genome specifies the “hardware” of cells by determining ion channels and proteins, but it does not fully determine the “software” or behaviors. Cells have a degree of reprogrammability.
Xenobots, organisms made of frog skin cells, exhibit novel behaviors and capabilities not seen in any natural organism. This shows the plasticity and potential of biological systems.
Evolution exploits higher-level interfaces that give access to computation, modularity, and other capabilities.
Regeneration and development involve error correction and goal-directed processes to achieve the correct final form.
The concept of a “set point” suggests that biological systems are goal-directed in a cybernetic sense.
The distinction between hardware and software is useful in understanding how biological systems work. The genome specifies the hardware but not the full range of possible behaviors.
Affect, goals, rewards, and other concepts typically used for brains may also apply at the cellular level.
https://www.youtube.com/watch?v=TQa08lXtWDY
Biological life is a form of collective intelligence composed of multi-scale competent agents. Understanding this can provide insights into regeneration, evolution, robotics, and AI.
Biological systems are capable of solving problems in diverse spaces like gene expression, physiological states, and anatomical configurations.
Cells and organisms have many hidden capabilities that are revealed when placed in different contexts or environments.
Development and morphogenesis are more flexible and robust than typically assumed, allowing organisms to adapt to changes.
Organisms are composed of multi-scale competent systems, with cells, tissues, organs, and bodies solving problems at their respective scales.
Organisms can be “hacked” by signals that activate their inherent competencies to produce novel forms and behaviors.
Pattern memories stored in bioelectrical circuits can be rewritten to produce long-term changes in morphology and behavior.
The same hardware can generate diverse forms and behaviors depending on the signals and information provided.
A technological approach is needed to understand, predict, control, and ethically relate to composite beings made of biological and non-biological parts.
Biological systems are highly interoperable and capable of integrating foreign DNA, nanomaterials, and software in plausible ways.
https://www.youtube.com/watch?v=qMsI9h1MY4A
There is debate on whether AI systems like GPT-3 truly demonstrate intelligence and understanding, or if they are just high-tech plagiarism that lack real depth and experience.
AI systems may be good at explaining past data and making predictions, but they may lack the ability to generate new research and capabilities due to their limited frameworks.
AI systems currently lack a meaningful connection to their substrate and causal levels, unlike humans who are connected from the subatomic level up.
AI systems may be good at confabulating and telling stories, but they lack the real experience and functional interactions to truly understand the concepts they discuss.
The human brain and consciousness are still not fully understood, and there are debates on how consciousness arises from the physical brain.
Split brain patients show that consciousness is difficult to eliminate, suggesting it may arise from local circuits throughout the brain.
Scientific dogmas can prevent new ideas from entering a field, and organized discussions can help challenge these dogmas.
Different personalities and consciousnesses within an individual are committed to their own stories that shape their identities.
The environment shapes an individual’s personality from a young age by rewarding and punishing certain behaviors.
There may not be a natural evolutionary path that leads directly to AI systems like GPT-3, as they lack the multi-scale self-construction of biological organisms.
https://www.youtube.com/watch?v=a_rNUUJWLGs
Agency and the ability to solve problems from the start is important for biological systems. Biological cells have to determine their boundaries and relationships on their own.
There is an “intelligence ratchet” in biology where competency in problem solving allows for messier hardware like genomes. This puts more evolutionary pressure on developing competencies.
Current AI systems lack the multi-scale agency that biological systems have, where parts have some degree of intelligence. They also don’t have to construct themselves from scratch.
We have a limited model of intelligence that only considers humans as the standard. We need to expand our understanding of diverse intelligences that may be different from humans.
The future will likely involve diverse beings with different bodies and minds, not just AI software. We need to learn how to relate to these beings ethically.
Binary categories like “intelligent” vs “not intelligent” are not helpful. We should consider degrees of intelligence and in what domains.
A minimum cognitive light cone of compassion for others should be prioritized and increased, not specific human anatomy or genetics.
The future could involve a diversity of embodied beings as long as they have the cognitive capacity for responsibility and compassion.
Current issues around identity may seem trivial in the future with more diverse embodiments.
Society is already moving towards accepting people regardless of where they came from or what they look like.
https://www.youtube.com/watch?v=lT-D_uHyqa4
Evolution does not necessarily optimize for things humans value like happiness, intelligence and meaning. Human bodies and minds still have room for improvement.
How we model humans and the teleological view of evolution affects our research and understanding.
Living beings are more responsive to values than non-living things, and humans can value more things than other creatures.
Machines that mimic human behavior superficially may be mistaken as being like humans, but they lack experiential depth.
Our narrow human-centric view limits our interactions with diverse intelligences that are not like us.
Other living intelligences besides humans, like animals, show capabilities for thinking, feeling and having rituals. Machines lack this.
For a meaningful relationship, beings need to share the same existential struggles of figuring out who they are and where they begin and end.
Experience and feelings give life meaning and value, which machines lack.
For two beings to harmonize, they need to be built from the same fundamental, sharing causal structures and resonating at multiple levels.
Parts and wholes are misunderstood. Simple systems show emergent properties highlighting that deep cognition is a feature of the universe.
https://www.youtube.com/watch?v=jdrfx7Z5oo4
There is value in having two distinct but complementary parts that can work together and complement each other. This allows for a whole that is greater than the sum of the parts.
Lateralization in the brain and body is important for having both focused attention and broad, open attention simultaneously.
There are differences in degree, not kind, between living and non-living things. Living things can respond much faster and to a wider range of stimuli.
Teleology and purpose are important for understanding life but have been neglected by biology.
Architective structures that change in discrete steps contrast with connective structures that change through continuous flow. Living things tend to have more connective structures.
Algorithms implemented in machines are divorced from their physical implementation, while in organisms the physical substrate matters.
Organisms can degrade gracefully when stressed, while machines tend to break down abruptly.
Plants can make intelligent decisions and adapt based on their environment, not just through pre-programmed responses.
Resonance in physical systems can bridge scales and connect different levels.
The music metaphor of concordant and discordant notes illustrates how combining elements can create something new and unpredictable.
https://www.youtube.com/watch?v=fgnQBD0CjMo
We have very limited control over what enters our consciousness and what our next thought will be. Consciousness seems to pivot on uncertainty and modulating confidence in predictions.
Preferences are rooted in phenotypic needs and innate predictions, but are individualized based on context and niche.
Consciousness is associated with palpating uncertainties in predictions to guide voluntary actions.
There are multiple categories of needs that must be satisfied, which compete and conflict with each other.
The nervous system is more plastic and amenable to learning from experience compared to other organ systems.
Consciousness requires the ability to learn and update predictive models based on experience, not just innate predictions.
The hierarchical structure of the predictive model is another feature that the nervous system lends itself to.
Even simple cells have multiple needs that require feedback loops and a “meta-processor” to allocate resources and prioritize energy.
The “action bottleneck” of limited energy and resources requires prioritizing among competing needs.
The ability to choose between qualitatively different categories of needs may be what gives rise to qualities of consciousness.
https://www.youtube.com/watch?v=1S4jaYjuzm0
There is a discussion about the organizing principle for embryos and where it is located, as well as the definition of things and how that relates to cognition.
There is an analogy made with numbers and numerals, where the numbers contain the information and constraints, while the numerals are just representations. This is related to the genotype-phenotype distinction.
Computations allow for particular results instead of mixing, which helps avoid losing the identity of original elements. This allows for evolution and selection.
The macroscopic constraints that determine certain phenomena are not predictable from the microscopic level. They come from boundary conditions set by the environment.
The same physical events can be interpreted in different ways and thus compute different things, depending on the observer and assigned semantics.
Memories can be stored in a distributed fashion, and different parts of the brain can interpret and recover them.
The identity of a system comes from the replacement policies of its components, which preserve some invariant while allowing for change over time.
Properties can be expressed as entities in relationships, which form a hierarchical structure. The specified vs. substitutable aspects sit in the relationships.
Different languages can encode entity-relation-entity structures in different ways.
It can be difficult to find reviewers and publishers for interdisciplinary papers that span multiple fields.
https://www.youtube.com/watch?v=LJ6yP6QTM1M
Collective behavior can enable groups to accomplish tasks that individuals cannot, despite facing coordination challenges.
Decentralized groups are robust but require dynamic control to coordinate.
The ants studied prioritize coordination over efficiency when navigating obstacles. They maintain consensus even when unsure of direction.
Flocking models with simple individual rules can reproduce swarm behavior.
High alignment weight among individuals was necessary and sufficient for escaping obstacles in the flocking models.
Low effective turning radius, due to high alignment, low informed individuals, or low individual turn rate, enabled obstacle escapes.
High alignment weight allowed for rapid obstacle escapes while maintaining agility and speed.
Cooperative transport ants prioritize coordination over efficiency by maintaining consensus.
Untethered groups also prioritize coordination over efficiency through high alignment.
The ants studied add complexity to their obstacle navigation strategy only when needed.
https://www.youtube.com/watch?v=HoGlW_F3M0c
There are evolutionary processes that are smarter than natural selection through random variation and selection. Learning systems show a more intelligent optimization than natural selection.
Embryos and organisms are plastic and flexible from the start, figuring things out from scratch each time. This gives them the intelligence to handle novel situations.
Evolution can explore an “intelligent space” of solutions through a non-human level but non-zero intelligence process. It exploits mathematical and physical principles.
Values and goals play an important role in evolutionary and learning processes. Simple maximization is not sufficient to explain complex organism behaviors.
There is a duality between exerting control over the world and being sensitive to information from the world. Organisms need both to act and observe.
Asymmetry between brain hemispheres could enable a more continuous flow between taking in information and taking action.
Simplistic value systems focused only on maximizing “the best” lead to problems. More nuanced, multi-dimensional values are needed.
Values shape what we attend to and experience from the very start, not just as an “add-on” at the end.
Scientific objectivity does not acknowledge that values are needed to pick what to study and measure.
Explaining organisms only through maximizing survival and reproduction fails to account for their complex beauty.
https://www.youtube.com/watch?v=ynHfrfpTH18
Biology uses a multi-scale competency architecture of hierarchical problem solvers in various problem spaces that evolution and parasites can exploit.
Bioelectrical networks are a major medium through which cells collectively process information, and are ancestors of the nervous system.
Cells have the ability to solve local goals and navigate morphological spaces without a brain or nervous system.
Cells have the ability to adapt and solve problems they’ve never encountered before through navigating the large space of gene expression.
There is a physiological software layer between the genome and anatomy that determines anatomical structures and regeneration.
Cells can reach the same anatomical goal through different developmental paths, using different molecular mechanisms.
Cells can store and rewrite anatomical memories that determine regeneration and morphology.
Bioelectrical patterns can specify organ formation and regeneration at a high level, leveraging the intelligence of the tissue.
Simple bioelectrical interventions can rescue drastic hardware defects by overriding them with high-level signals.
Even simple skin cells have the latent capacity for novel morphogenetic and behavioral capabilities when freed from external constraints.
https://www.youtube.com/watch?v=7SwIQEEmIp4
Biology exhibits diverse forms of intelligence at multiple scales, from single cells to organs to the whole body. This multi-scale intelligence can inspire new approaches for AI.
Single cells and organisms like slime molds show problem-solving abilities by navigating morphological spaces and responding to environmental cues.
Organisms like planaria and salamanders show remarkable regenerative and adaptive capabilities, reconfiguring their bodies and brains in response to injuries.
Development and morphogenesis in organisms are not hardwired but involve navigation of a “morphe space” to achieve target morphologies despite perturbations.
Collectives of cells can exhibit intelligence by solving problems at larger scales, though this comes with failure modes like cancer.
Non-neural bioelectricity may serve as a “cognitive glue” for collective intelligence, analogous to the role of neurons in the brain.
Organisms show plasticity and ability to solve novel problems not seen during evolution, indicating a “problem-solving machine” architecture.
Intelligence can be abstracted as the ability to achieve goals using different means, with more sophisticated means indicating higher intelligence.
Novel hybrids and cyborgs combining biological and engineered materials may open up a vast design space for novel intelligent systems.
Stress propagation and gap junctions between cells may enable gradient-like information sharing that underpins collective intelligence.
https://www.youtube.com/watch?v=TgINASlxeXE
Bioelectrical networks of cells underlie intelligence and problem solving in the body. Cells have competencies to solve problems at the molecular, transcriptional, and anatomical levels.
Morphogenesis, development, and intelligence are fundamentally the same problem of collective cell behavior and information processing.
Cells have a high level of competence and autonomy to solve physiological problems on their own scale, despite being part of a larger organism.
Cells can find solutions to novel problems and stresses through changes in gene expression, showing their adaptability and plasticity.
Cells can achieve the same anatomical goals through different molecular mechanisms, demonstrating their intelligence.
Targeted changes in the bioelectrical state of cells can cause them to form new organs and structures, like eyes and limbs.
Computational models of bioelectrical networks can predict changes needed to correct deformities and defects.
Connecting cells electrically can override genetic defects and mutations, demonstrating the power of “software” over “hardware.”
Scaling up cognition from cells to tissues enables larger computational capacities and goal-directed behavior at the organism level.
Cells exhibit plasticity and competency to form novel structures and behaviors when removed from their normal context, like self-replicating xenobots.
https://www.youtube.com/watch?v=7FGM33sz25k
The speaker argues for a framework that can understand and relate to diverse intelligences, regardless of their form or origin. This approach is called TAME—technological approach to mind everywhere.
Intelligence manifests in different ways and problem spaces, not just 3D space. Cells and organs exhibit intelligent behaviors in physiological and anatomical spaces.
Morphogenesis and development involve collective intelligence at the cellular level. Cells cooperate and regulate each other to form complex structures.
The body plan and anatomical structures are encoded in the bioelectrical dynamics and signaling between cells, not just the genome.
Cells can utilize different molecular mechanisms to achieve the same anatomical goal, showing top-down causation and goal-directedness.
Connecting cells electrically allows them to act as a collective that can pursue larger-scale goals and represent higher-level concepts like “eye” or “leg”.
Disconnecting cells electrically, as in cancer, causes them to lose this collective intelligence and revert to their unicellular past.
The speaker’s lab has shown they can reprogram organs and regenerate structures by manipulating the native bioelectrical interface between cells.
Evolution enlarges the “cognitive light cone” of organisms, allowing them to pursue larger goals in different problem spaces.
The speaker’s group created a novel organism called a xenobot by liberating cells from normal developmental constraints, showing cells have default capacities that evolution normally shapes.
https://www.youtube.com/watch?v=iIQX6m2eRPY
Cells use bioelectrical signals and gradients to make decisions regarding large-scale anatomy and organ morphogenesis, beyond just determining cell fate.
Manipulating bioelectrical states, through ion channels and gap junctions, can control organ development and regeneration. This was demonstrated in experiments with frog embryos and tadpoles.
Bioelectrical signals encode information about anatomical layouts and goals that cells work towards building.
Ion channels act as “transistors” that form feedback loops and memory circuits, allowing cells to make collective decisions.
Understanding and cracking the “bioelectric code” could reveal how cell networks make large-scale anatomical decisions.
Machine learning tools can help design interventions to manipulate bioelectrical signals for regenerative medicine and synthetic biology applications.
Depolarizing cells can cause them to revert to a more “unicellular” state and promote tumor formation and metastasis.
Forcing depolarized cells to remain electrically coupled can override oncogene expression and prevent tumor formation.
Computational models can identify ion channel cocktails that can manipulate bioelectrical signals to achieve desired organ morphogenesis.
Short-term manipulation of bioelectrical signals can trigger long-term organ growth and regeneration.
https://www.youtube.com/watch?v=WM8bQWfmeB8
The discussants are interested in exploring consciousness and sentience from a fundamental physics perspective using the free energy principle and active inference framework.
They want to develop mechanistic explanations for how feelings and experiences arise from basic biological mechanisms.
Engineering an artificial system that has rudimentary feelings could help demonstrate the validity of their theories.
They discuss the challenge of objectively proving the existence of subjective experience and other minds.
Simple forms of sentience may exist in single cells and arise from an agent making choices rooted in what can be considered feelings.
They discuss the need for shared existential struggles and compatible goal settings for bonds to form between humans and artificial agents.
Arbitrary values like survival and affiliation emerge from our mammalian nature but more fundamental criteria may exist.
Consciousness may exist at larger collective scales implemented by interactions at smaller scales.
Immersing oneself in a virtual environment mimicking an artificial agent’s experience could provide empathy and evidence for that agent’s sentience.
It is difficult to intuitively conceive of consciousness at larger collective scales beyond one’s own self.
https://www.youtube.com/watch?v=4Z8UPddh0e4
There is a debate about whether biological organization arises from goal-directed processes or attractor dynamics. Both views have merits and the truth likely lies on a spectrum between the two.
Cells exhibit competency—the ability to sense their environment and neighbors, and move to positions that fit better. This competency can hide genetic information from selection pressures.
Competency and communication between cells can allow an organism to be more robust to genetic mutations.
Evolution may work more on cognitive competencies rather than physical traits like speed or strength.
Parts of a system do not need to be intelligent themselves to give rise to intelligent behavior at a higher level.
Threats from parasites and exploiters drive the development of self-identity and the ability to distinguish internal vs. external influences.
There are constant and variable properties that define objects. The constant properties allow identification while the variable properties provide information.
There is a debate about whether humans have true internal representations or just post-hoc explanations for behaviors.
Language can be modeled using a simple frame of “entity—relation—entity” which captures much of English grammar and semantics.
Different languages employ different strategies for marking entities and relations, like word order or word endings.
https://www.youtube.com/watch?v=52lDk9bmphM
Iain McGilchrist’s work focuses on understanding nature from both a reductive, quantitative perspective as well as a top-down perspective.
Michael Levin’s research looks at how bioelectrical gradients help organisms determine left and right. This shows how large-scale information processing arises from individual mechanisms.
Bioelectricity may allow us to bridge different levels of explanation, from mechanistic to cognitive.
Experiments show that gene regulatory networks can exhibit different types of memory and learning, even in simple models.
Voltage imaging reveals that planaria have pre-patterns that indicate how many heads they should have. This suggests an “electric circuit” that defaults to an attractor state.
Memories are important for shaping our personalities and character, but the actual experience itself also matters.
There are different perspectives on the continuity of personal identity over time, depending on how one views the self.
The left and right hemispheres pay attention to the world in different ways, with broad vs. narrow focuses.
McGilchrist argues that embracing science and reason more wholeheartedly can reveal a richer, more complex reality that we are connected to.
Providing people with a new perspective can radically change their lives for the better.
https://www.youtube.com/watch?v=bCwTH5f2DnE
The speaker is proposing an “accounting system” using polynomial functors and natural transformations to model dynamic interfaces and arrangements of systems. He claims this framework works well experimentally.
The framework uses polynomials to represent interfaces and natural transformations to represent arrangements. Operations on polynomials can generate new interfaces and arrangements.
The framework aims to provide a common language to describe and compare different systems, from cells to computers to living organisms.
The speaker claims the mathematical framework can model how wiring and arrangements of systems can change over time, which could help understand phenomena like morphogenesis.
The framework focuses more on structural questions rather than numerical details.
The speaker argues that the mathematical language of this framework is precise and articulate, though he does not provide concrete proofs.
The framework aims to provide “anatomical programming language” as a tool to model protein folding and how interactions affect organism positions.
The framework could potentially be useful for understanding morphology and behavior, though the speaker admits they lack concrete tools or programs currently.
The discussion highlights open questions around what controls arrangements of systems and how self-organization emerges.
The framework could potentially be applied to model interactions between neurons in cell cultures to gain insights into general laws of engagement.
https://www.youtube.com/watch?v=DpAi-rtnjTM
Living systems maintain a stable low entropy state far from thermodynamic equilibrium by using information. This is a unique property of living systems.
The central dogma of biology states that information flows from DNA to RNA to proteins, but it does not capture the full complexity of information processing in cells.
Enzymes encoded in DNA accelerate reactions through quantum interactions that lower activation energy, converting genetic information into a thermodynamic state.
Most of the cell’s information is stored in transmembrane ion gradients, not just the genome. Membrane proteins use this information.
Transmembrane ion pumps create ion gradients that are used by ion channels to allow selective ion fluxes, generating local information dynamics.
Local ion fluxes can change the function of membrane proteins and allow movement of macromolecules to the membrane.
The cytoskeleton can transmit information rapidly through the cell, providing a distributed network for information processing.
Complexity in living systems arises more from membrane-to-membrane interactions than from genome size.
The genome provides the machinery to generate ion gradients, but information exchange between cells through membrane dynamics drives complexity.
The nucleus is one part of a broader information system, not the central processor of the cell.
https://www.youtube.com/watch?v=d-ZK41F_1jE