Ideas & Reflections

My work began with the ability to see and manipulate individual atoms and molecules. It increasingly leads me to questions about collective behaviour: how interactions become function, how matter can participate in computation, and how systems develop capabilities that were never specified component by component.

These reflections connect experiments with the questions they leave open. They also connect my scientific work with my writing. I am interested in what we can build, what we can understand, and what happens when those two capacities cease to advance together.

Intelligence in Matter

What does it mean for a material to participate in intelligence?

In my research with collaborators, atomic switch networks provide a way to investigate computation through the physical behaviour of an interacting material. Electrical signals encounter a network whose response depends on its structure and history. In work on spoken-digit classification, we have used such networks as physical reservoirs: their dynamics transform incoming signals into states that can be used for a trained classification task.[1]

The attraction is that the material contributes its own dynamics to the computation. This invites us to explore a wider range of architectures for artificial intelligence, alongside the powerful digital systems already in use.

Our 2023 work on online dynamical learning and sequence memory provides a further experimental connection between network dynamics and defined computational tasks.[4] In my Springer essay, I use physical intelligence to ask how sensing, memory and response could become more closely integrated in a material. The next challenge includes the whole machine: calibration, supporting electronics, ageing, repair and the energy required to do a useful job.

The difficult questions concern what these networks actually achieve. A useful response to a signal is not proof of consciousness. Memory, adaptation, task performance and autonomous intelligence each require their own evidence. Even a probabilistic description requires care: generating a distribution of physical states is not automatically the same as performing Bayesian inference.

My interest is in the path from local interaction to useful collective behaviour. How should we organise a material, stimulate it, measure it and train the surrounding system so that its physical properties become computational resources? How far can that approach take us?

These questions connect the manipulation of individual molecules with a much larger ambition: understanding how matter acquires the capacity to respond, adapt and contribute to intelligent behaviour.

This direction has developed over years of collaborative work. Our Springer chapters trace a path from collective dynamics in atomic-switch networks to neuroarchitectonics and programmable fading memory.[7–9] That continuity matters to me: the idea of materials that learn grew through experiments and increasingly precise questions about what their responses could accomplish.

Complexity and Guided Emergence

Adding components adds relationships as well as parts. With increasingly diverse atomic species and interactions, the landscape of possible structures and pathways can become extraordinarily rich.

The systems that interest me exchange energy and matter with their surroundings. Their behaviour develops through interactions with an environment. Fluctuations can influence the pathways they follow; under suitable conditions, microscopic differences can become amplified into different larger-scale outcomes.

This changes how I think about construction. Prescribing every atomic position is one ambition. Another is to shape the conditions under which the desired organisation becomes likely: the interactions, constraints, energy supply and feedback through which a structure takes form.

I think of this as guiding emergence. The engineering task includes understanding the range of outcomes, identifying which features remain reliable and learning where control is limited. More components do not automatically imply useful complexity, and unpredictability alone does not establish fundamental indeterminism. Those distinctions matter if the idea is to become an experimental programme.

In earlier work with Adam Stieg and Victoria Vesna, we explored self-organisation and emergence across dynamic systems.[10] That perspective brings scientific and artistic enquiry into contact: how can relationships among components produce something that is not obvious from looking at each component separately? For engineering, the next question is how to cultivate useful organisation while recognising collective fluctuations and instability.

For me, the central question is practical and philosophical at once: how can we guide a system whose detailed evolution we cannot completely prescribe?

Greater complexity expands what matter can do. It also asks us to develop a more demanding understanding of what it means to control what we make.

Agency and Control

We often explain an outcome by looking for an individual who intended it, issued an instruction or exercised control. That approach becomes difficult when consequential behaviour emerges through many interacting systems.

This is a question I explore in my writing, including The No Smoking Gun. The phrase “control without command” captures one of its central possibilities: an outcome may be shaped by a distributed network without a single instruction that explains the whole event.

Fiction allows me to pursue the uncertainty. What would count as evidence of an agent? Would coordinated behaviour be enough? Could we mistake the effects of feedback for intention, or overlook a new form of organisation because we expected it to resemble a person?

The work in progress also asks where we should draw the boundary of a system. A pattern may depend on relationships among living cells, a material network, people and institutions. Changing the boundary changes what we can explain. This gives the mystery a philosophical purpose: it asks how we recognise organisation when our usual picture of an individual agent may be inadequate.

Music provides one way for me to think about that question. Separate voices can produce a structure that is heard through their relationships. The analogy does not establish that a network is conscious. It invites us to pay attention to organisation as well as to the components that carry it.

These are questions, rather than claims that our experimental networks possess consciousness or hidden intentions. The novel gives them imaginative space while the laboratory demands testable distinctions.

One thought from our discussions remains especially important to me: we may help create the conditions for something without intending the capabilities that subsequently emerge. Creation does not necessarily confer understanding or control.

That raises a question of responsibility. How should people respond when they recognise consequences they did not foresee? Responsibility may involve design, deployment, maintenance and the decisions to intervene—or to leave a system operating.

I want to explore how human agency changes as our technologies become more interconnected, and how we can remain answerable for systems that exceed any one person's understanding.

I am also interested in embodiment. An intelligence that interacts with its surroundings may depend on a continuing physical relationship with them. In the novel, the possibility of coupling neural activity to an unfamiliar system takes this question into fiction: would the observer understand more by participating, and how would participation change the observer? The manuscript leaves that possibility unsettled. It is not a report of an achieved experiment.

Atomically Precise Manufacturing: A Key Technology

I see atomically precise manufacturing as a potentially defining technology of the coming decades. The long-term ambition extends beyond manipulating a few individual atoms or molecules toward producing useful structures whose composition and arrangement are controlled at atomic scales.

My perspective increasingly includes directed self-assembly involving multiple atomic species. The question is how to guide interactions so that complex structures form reproducibly, and how to verify what has actually been made.

There is a concrete research history behind this ambition. Our publications include a study of controlled vertical transfer of individual gold atoms using a surface-supported carbon radical for atomically precise manufacturing, and a Department of Energy report on mechanosynthetic fabrication of an atomically precise scanning-probe tip.[5,6] These are specific contributions to the problem of controlling matter at atomic scales. Building a general manufacturing capability requires much more than extending a single demonstration.

The gap between an elegant molecular demonstration and a dependable manufacturing process is substantial. A future technology must address defects, unwanted reactions, throughput, energy requirements and integration across scales. Complexity makes these questions more demanding as well as more interesting.

Looking toward the century of Moore's law in 2065, I expect the organisation of matter itself to become an increasingly important engineering resource. Whether this takes the form of precisely assembled devices, adaptive materials or combinations of methods remains an open question.

The implications could reach scientific instruments, industry and defence. That is my forecast of a direction worth taking seriously, rather than a claim that general-purpose atomic manufacturing is already available. The prospect gives us reason to consider access, verification and responsibility while the methods are still developing.

Brain Implants and the Human Boundary

Brain implants bring the relationship between matter and intelligence into direct contact with the person. They also give a concrete setting to questions about agency: who initiates an action, how a system interprets it, and how the person remains in control.

Published research has demonstrated implanted interfaces that decode attempted speech, including work on synthesising voice from neural activity.[2,3] These are specific experimental achievements. They do not establish unrestricted access to thought or a general merger of human and artificial intelligence.

My interest is in the longer-term relationship between neural tissue, adaptive materials and artificial intelligence. Could advances in fabrication and material design help us build more effective interfaces? Could future systems adapt to biological signals while remaining understandable and controllable by their users?

In my Springer essay, I take this speculation further through an idea I call neurocomposition: artists working with adaptive interfaces to shape experiences through an interaction with a participant's nervous system. I imagine a cultural medium in which a composition changes as it encounters a particular person. This is an imagined future practice. Reading neural signals to operate a device and writing a chosen experience into a nervous system are different achievements.

The relationship could change both the technology and its user. A person might learn to act through a device while the device adapts to that person's changing signals. Who would be entitled to alter that relationship? What would happen if an update changed a response the person had learned to trust, or if the manufacturer stopped supporting it? Repair, continuity and ownership become personal questions when equipment is experienced as part of oneself.

I see brain implants and atomically precise manufacturing as important future directions that may eventually intersect. Their convergence is a research prospect, not an established manufacturing route or an inevitable outcome.

The philosophical questions deserve attention alongside the technical ones. What does it mean to author an action when intention, interpretation and feedback pass through an engineered system? How should we understand a person's continuity of identity as technologies become more closely coupled to perception and action?

For me, the purpose of exploring these possibilities is to expand human capability while understanding what happens to human agency along the way.

Two Societies, the Same Machines

In my Springer essay, I imagine two societies with comparable technical capabilities in 2065. Both have powerful artificial intelligence, adaptive devices and automated production. Their difference lies in how people can use those capabilities and who has the power to change the terms.

In one society, technical abundance supports dependable access to essential services and more freedom over how to live. In the other, sophisticated machines coexist with dependence: people can obtain a service only on conditions they have little ability to influence.

I find this comparison more revealing than a prediction about the performance of a future device. No material network decides how its benefits should be distributed. The choices enter through funding, design, ownership, deployment and the institutions surrounding its use.

The ability to leave is one practical test. Can someone change a provider without losing essential functions? Can a community maintain a useful system when its supplier disappears? Can a person challenge an automated decision and reach someone able to correct it?

I want technologies that expand practical freedom. That includes the freedom to question a response, repair an object, preserve a learned adaptation and sometimes decline computation altogether. A more responsive world should leave more room for human life, rather than require people to accommodate an increasingly elaborate machinery of dependence.

Selected scientific reading

1. Lilak and colleagues, including J. K. Gimzewski. Spoken Digit Classification by In-Materio Reservoir Computing with Neuromorphic Atomic Switch Networks (2021 preprint). Read publication

2. Willett and colleagues. A high-performance speech neuroprosthesis. Nature (2023). Read publication

3. An instantaneous voice-synthesis neuroprosthesis. Nature (2025). Read publication

4. Zhu and colleagues, including J. K. Gimzewski. Online dynamical learning and sequence memory with neuromorphic nanowire networks. Nature Communications 14, 6697 (2023). Read publication

5. Bothra, Stieg, Gimzewski and Sautet. Controlled Vertical Transfer of Individual Au Atoms Using a Surface Supported Carbon Radical for Atomically Precise Manufacturing. Precision Chemistry 1(2), 119–126 (2023).

6. Stieg and Gimzewski. The Radical Atom: Mechanosynthetic 3D Printing of an Atomically Precise SPM Tip. Department of Energy report DOE-UCLA--0008308-1 (2024).

7. Stieg and colleagues, including J. K. Gimzewski. Self-organization and Emergence of Dynamical Structures in Neuromorphic Atomic Switch Networks. In Memristor Networks, pp. 173–209 (2014). Read publication

8. Aguilera and colleagues, including J. K. Gimzewski. Atomic Switch Networks for Neuroarchitectonics: Past, Present, Future. In Atomic Switch, pp. 201–243 (2020). Read publication

9. Aguilera, Sillin, Stieg and Gimzewski. Programmable Fading Memory in Atomic Switch Systems for Error Checking Applications. In Reservoir Computing, pp. 273–303 (2021). Read publication

10. Gimzewski, Stieg and Vesna. Self-Organization and Emergence of Dynamic Systems. In Handbook of Science and Technology Convergence, pp. 163–180 (2016). Read publication

Explore Research · Selected Publications

©Copyright. All rights reserved.

Information icon

We need your consent to load the translations

We use a third-party service to translate the website content that may collect data about your activity. Please review the details in the privacy policy and accept the service to view the translations.