Author: Iñigo Careaga, strategy & communication manager, CIC energiGUNE

Lithium and sodium are neighbours in the periodic table. Both are alkali metals, both have a single electron in their outer shell, and both can be used to shuttle ions between two electrodes to store and release electrical energy. At first sight, replacing lithium with sodium in a rechargeable battery might therefore seem straightforward.

There are good reasons to try. Sodium is widely available and sodium-ion batteries can rely on electrode materials containing abundant elements such as iron and manganese. But chemical similarity does not mean electrochemical equivalence. A sodium ion is larger and heavier than a lithium ion, and its interaction with the materials surrounding it is different. As a consequence, replacing Li⁺ with Na⁺ changes much more than the identity of the charge carrier.

The electrodes have to accommodate a different ion. Their structures may respond differently as sodium is inserted and removed. The electrolyte has to transport and solvate it. New interfaces form between the electrolyte and the electrodes. Even graphite, the standard anode material of most commercial lithium-ion batteries, does not store sodium under conventional conditions nearly as effectively as it stores lithium ¹.

A sodium-ion battery is therefore not simply a lithium-ion battery containing a different element. Understanding why requires looking at what happens to sodium once it enters the battery.

Graphite does not play by the same rules

Charging a conventional lithium-ion battery involves moving lithium ions from the positive electrode through the electrolyte and storing them in the negative electrode. Graphite is particularly good at this job. Its carbon atoms are arranged in sheets, and lithium can reversibly occupy the spaces between them, a process known as intercalation. It would be convenient if sodium behaved in the same way. It largely does not.

The poor ability of conventional graphite to accommodate sodium has long been one of the reasons why lithium-ion concepts cannot simply be transferred to sodium-ion batteries. Instead, one of the most widely studied negative-electrode materials is hard carbon ².

Hard carbon is still carbon, but structurally it is very different from graphite. Rather than a regular stack of extended carbon sheets, it contains disordered graphitic domains, defects and pores. This disordered architecture provides environments in which sodium can be stored, but it also makes the underlying mechanism considerably harder to describe.

Where exactly does the sodium go? Researchers have proposed contributions from adsorption at defects and surfaces, insertion between graphitic layers and storage associated with nanopores. The relative importance of these mechanisms depends on the structure and microstructure of the carbon. This is more than a microscopic detail: where and how sodium is stored determines the voltage profile, reversible capacity, first-cycle losses and, ultimately, the energy that a complete cell can deliver.

Researchers at CIC energiGUNE and collaborators have studied this connection between charge-storage mechanisms, carbon structure and electrochemical performance ³. Their analysis highlights an important point: maximizing the amount of sodium that an anode can store is not enough. The potential at which that storage occurs and the irreversible losses during the first cycle are equally important when the material is incorporated into a full battery.

The problem is therefore not simply to find a carbon that accommodates more sodium. It is to understand how its structure stores sodium and then engineer that structure so that capacity, voltage and efficiency improve together. This has led researchers to explore different ways of controlling hard-carbon structure, including the use of lignocellulosic biomass as a precursor. Work has shown that even apparently simple processing decisions, such as acid treatment of the biomass precursor, can affect the resulting carbon and its behaviour as a sodium-ion anode ⁴.

And the negative electrode is only half of the problem. At the positive electrode, sodium ions have to repeatedly leave and re-enter a solid structure. Layered transition-metal oxides are attractive candidates because sodium can be reversibly extracted from and inserted into spaces between layers of metal and oxygen atoms. Compositions based on relatively abundant elements such as iron and manganese are particularly interesting ⁵.

Again, however, accommodating sodium is not enough. The host structure has to survive the process. A useful example is the layered oxide P2-Na₂/₃[Fe₁/₂Mn₁/₂]O₂, investigated by researchers from CIC energiGUNE and collaborators ⁶. The material can provide high capacity, but its long-term stability strongly depends on the voltage range over which it is operated.

This produces a familiar battery dilemma. Increasing the operating voltage can increase the amount of energy extracted from a material. But pushing the electrode further can also trigger structural transformations that make subsequent cycles less reversible.

The researchers compared cells operating in different voltage windows and combined electrochemical measurements with structural characterization and theoretical calculations. Above approximately 4.0 V versus Na+/Na, the cathode undergoes a high-voltage phase transition. Avoiding this transition improves capacity retention, revealing a direct connection between the structural state of the electrode and its long-term electrochemical behaviour.

Changing the chemical composition can also alter this structural response. In another study, researchers investigated the effect of introducing titanium into sodium manganese-rich layered oxides ⁷. A certain degree of Ti substitution helped the crystal structure accommodate electrochemically induced strain. Too little was insufficient to stabilize it, while further substitution introduced other destabilizing effects.

Other studies have approached the same problem from a different angle, examining how the precise stoichiometry of P2 layered oxides affects their performance ⁸, or following their structural evolution under different current rates and dopants using operando techniques ⁹. The lesson is that there is no single property called “a good sodium cathode”. Composition, atomic arrangement, voltage and cycling conditions are interconnected. Even an element that does not directly provide the stored charge can determine how well the electrode survives sodium repeatedly moving through its structure.

A different chemistry

Layered oxides are not the only possible positive electrodes. Sodium chemistry allows other structural families to be explored.

One particularly instructive example is NaFePO₄, the sodium analogue of the well-known lithium iron phosphate chemistry. Researchers involving CIC energiGUNE have used in situ X-ray diffraction to follow how NaFePO₄ changes during sodium insertion and extraction ¹⁰. They have also shown that the reaction mechanism itself depends on how rapidly the electrode is charged or discharged ¹¹.

This is a subtle but important point. We often describe an electrode material as if it had a fixed reaction mechanism. In practice, what happens inside it can depend on how the battery is operated. Computational work has gone further by investigating sodium diffusion in NaₓFePO₄ ¹². Combining experiments and modelling in this way makes it possible to connect what is observed at the battery level with the motion of ions through an atomic structure.

But even understanding both electrodes is not enough. Between them lies the electrolyte, and the sodium ion spends much of its journey there. It becomes surrounded by solvent molecules, travels across the cell and eventually reaches an electrode surface. At that interface the electrolyte can decompose, producing a solid-electrolyte interphase, or SEI.

This apparently thin layer can have a disproportionate influence on battery behaviour. Researchers at CIC energiGUNE have investigated both the composition and evolution of the SEI in sodium-ion electrodes ¹³ and, more broadly, the relationship between electrode-electrolyte interphases, transport properties and electrochemical performance in non-aqueous sodium-ion batteries ¹⁴.

The interface illustrates particularly well why changing lithium for sodium is not a simple substitution. The chemistry of the ion influences its solvation, its transport and the decomposition reactions that occur when it reaches an electrode. A material that can reversibly store sodium may still perform poorly if the interface around it continuously consumes electrolyte or sodium during cycling.

This is why sodium-ion research has to consider the cell as a system. Graphite may have to give way to hard carbon. Positive electrodes require structures able to tolerate repeated sodium insertion and extraction. Electrolytes and additives have to create suitable transport conditions and stable interfaces.

And every attempt to solve one of these problems can affect the others.

Sodium is not the new lithium

Lithium has a fundamental advantage that materials engineering cannot simply remove: it is the lightest metal and has a very low electrochemical potential. Sodium is heavier and its electrochemical potential is less favourable, placing intrinsic constraints on the gravimetric energy density that sodium-ion batteries can reach.

That does not necessarily make sodium a worse battery chemistry. Energy density is only one parameter of an electrochemical storage system. Cost, raw-material availability, lifetime, power, safety and operating temperature can become equally important depending on the application. A battery designed for a mobile phone, an electric vehicle or stationary storage does not necessarily have to optimize the same variables. And sodium chemistry itself offers several possible ways of balancing them.

CIC energiGUNE researchers have explored, for example, low-cost phosphate-based cathodes for aqueous sodium-ion batteries ¹⁵, Prussian Blue compounds as positive electrodes in aqueous systems ¹⁶, alternative sodium layered oxides [5,8,9], hard-carbon anodes [2,3,4] and even materials such as NaTiOPO₄ designed to modify or potentially avoid some of the interfacial problems normally associated with sodium-ion anodes ¹⁷.

These are not different routes towards exactly the same battery. They represent different ways of deciding which compromises are acceptable.

This changes the original question. The relevant challenge may not be how to make sodium behave as much like lithium as possible. Research on hard carbons shows that sodium may require a different type of host structure. Work on layered oxides shows that extracting more energy from an electrode can alter its structure and compromise its stability. Studies of NaFePO₄ show that the reaction pathway itself can depend on cycling conditions. And research on interfaces shows that storing sodium reversibly is only part of the problem: the chemistry taking place around the active material matters as well.

Sodium-ion batteries are therefore not lithium-ion batteries in which one element has simply been exchanged for another. Changing the ion changes the materials that host it, the structures they adopt, the reactions they undergo and the interfaces they form.

Sodium may not need to beat lithium at its own game. The more interesting question is whether understanding its different chemistry will allow us to play a different one.

 

The BRTA is a consortium that remains a step ahead of future socio-economic challenges worldwide and in the Basque Autonomous Community; it addresses them through research and technological development, thus projecting itself internationally. The BRTA centres collaborate to generate knowledge and transfer it to Basque society and industry so as to make them more innovative and competitive. The BRTA is an alliance of 17 R&D centres and cooperative research centres with the support of the Basque Government, the SPRI and the Chartered Provincial Councils of Araba, Bizkaia and Gipuzkoa.

 

The post Why can’t we simply replace lithium with sodium in batteries? appeared first on Mapping Ignorance.

 

 

About the author: José Luis Granados Mateo is a former secondary-school and sixth-form STEM teacher, having taught physics, chemistry and mathematics. He is currently a postdoctoral researcher in the epistemology of science at the University of the Basque Country (UPV/EHU). 

 

When I taught physics, chemistry and mathematics to secondary-school and sixth-form students, a correct answer was sometimes where the interesting part began. I would ask why they had used a particular equation. A solution that had looked perfectly sound could become much less convincing; occasionally the reverse happened, and a student whose algebra had gone wrong could explain the physics rather well. Nothing on the page had changed. My estimate of what the student understood had.

I have been thinking about those exchanges while reading the research on generative AI in education. Students can now submit a convincing explanation, solve a numerical problem or produce working code with considerable help from a language model. It is easy to see the achievement and much harder to work out what the student has learnt along the way.

That difficulty cannot be settled by appealing to the virtues of an unaided mind. Scientific work depends on instruments, notation, diagrams, software and other people. Delegating part of a calculation may be entirely sensible. Yet teaching has a complication that is easy to overlook from the perspective of an expert: the operation being delegated may be one of the things through which a beginner is meant to acquire expertise.

An experiment by Hamsa Bastani and colleagues brings this problem into unusually sharp focus.¹ Nearly a thousand pupils at a secondary school in Turkey took part in a study comparing two GPT-4 systems with a control condition without AI. GPT Base offered assistance resembling a general-purpose chatbot. GPT Tutor had additional material from teachers, including solutions and common mistakes, and instructions to guide pupils with hints rather than give away the answer.

During practice, both systems helped. The estimated scores were 48% higher than the control mean with GPT Base and 127% higher with GPT Tutor. These are relative differences, not percentage-point gains. On the tests that followed, however, the pupils had to work without assistance. The GPT Base group then scored an estimated 17% below the control mean. GPT Tutor largely mitigated that penalty, but its users showed no statistically significant advantage over the control group. The tests took place within the study sessions, so they tell us about immediate independent performance rather than retention months later.

Had the researchers stopped at the practice exercises, they would have had an encouraging result for both systems. They did not stop there, and the educational interpretation changed. A pupil working with AI had become more successful at the task. Whether that success belonged to the pupil in any lasting sense was another matter.

Psychologists have long distinguished performance during acquisition from learning: a relatively durable change in knowledge or skill. Soderstrom and Bjork review the evidence for that distinction, including cases in which fluent practice leads to disappointing retention and others in which slower, more error-prone practice produces better results later.² Watching someone learn, we inevitably see performance. To find out what remains, we need to look again after a delay, withdraw some of the support or ask for an application in a different context. Even then, what we have is evidence to interpret, not a direct view of understanding.

In STEM, much of the important work happens before the calculation. A mechanics problem requires a choice of system, a judgement about the relevant interactions and some representation of the situation. Chemistry involves moving between observable phenomena, molecular accounts and symbols on a page. A program that runs is the endpoint of decisions about what to separate, what to abstract and how to test it. Nersessian’s work on model-based reasoning gives representations a central place in scientific thought: constructing and revising a model can be part of arriving at an explanation, rather than merely a way of presenting one.³

A language model can help at almost any of these points. Perhaps the student has made the difficult modelling decisions and wants an algebraic check. Perhaps the model has chosen the principle, drawn up the representation, done the algebra and supplied an explanation before the student has formed a view of the problem. A questionnaire asking whether both students “used AI” would record the same answer.

There is an obvious temptation here to defend struggle. I am wary of it. Spending half an hour looking for a formula can be frustrating without being especially instructive, and a teacher who withholds every useful clue has not necessarily protected the student’s learning.

Worked examples offer a familiar counterexample. A novice studying a solution can attend to the structure of a procedure without having to search blindly for it. Barbieri and colleagues’ meta-analysis, covering 43 articles, 55 studies and 181 effect sizes, found a medium benefit for mathematics performance (g = 0.48).⁴ Removing some work can help a learner do other work more productively. The judgement lies in deciding what to remove.

Nor will the right decision remain right indefinitely. In a 2025 meta-analysis of the expertise reversal effect, Tetzlaff and colleagues brought together 60 experiments involving 5,924 participants and 176 effect sizes. Learners with little prior knowledge benefited from greater instructional assistance (d = 0.505); for those with high prior knowledge, lower-assistance conditions were preferable (d = −0.428).⁵ These are heterogeneous results across educational contexts, not estimates specifically for generative AI. Still, they help explain a familiar experience: an explanation that once made a procedure possible can later get in the way of carrying it out.

This is part of what Koedinger and colleagues call the assistance dilemma: when should instruction provide information, and when should it require the learner to generate it?⁶ With a conversational model, the practical difficulty is that help need never run out. Another prompt can produce another hint, or an answer, or a complete solution to a problem the student has scarcely begun to consider.

Pablo Rodríguez Sánchez raised a related question in Mapping Ignorance, in Why teach calculus in the age of AI. Software can compute many standard derivatives and integrals almost immediately, and AI interfaces make such tools easier to consult. That does change the case for teaching manual procedures. It does not dispose of it. Working through integration by parts can make its relation to the product rule visible; substitution can bring the chain rule into focus from the other direction. Some of the value of a procedure lies in the relationships a student encounters while carrying it out. Requiring the procedure without regard to those relationships would miss the point just as thoroughly as abandoning it because a machine is faster.

We already move parts of our thinking into the world. We write a number down so that we do not have to remember it, or set a reminder for something we expect to forget. Risko and Gilbert describe such actions as cognitive offloading: changing the processing demands of a task through physical action and external resources.⁷ Its consequences are worth studying, but the fact that someone has offloaded a task is not in itself evidence of cognitive decline.

For history and philosophy of science (HPS), this should be familiar ground. Giere and Moffatt describe scientific cognition as distributed across people and artefacts.⁸ On this account, instruments, representations and social arrangements can be parts of the system through which scientific knowledge is produced. A physicist who uses a computer to integrate a differential equation has not thereby revealed a failure to understand physics.

She does, though, need some grounds for trusting the result. She must know what calculation is appropriate, which assumptions it involves and when the output ought to make her suspicious. A beginner may still be acquiring those resources. If every mechanics problem arrives already represented and ready for calculation, there may be very little occasion to learn how to represent one.

“Which formula do I use?” was a question I heard often. Usually, in the cases I remember, the difficulty was not simply a forgotten equation. The student had not yet made sense of the situation in a way that gave any equation a purpose. Supplying the formula could rescue the exercise while leaving that difficulty untouched. A question about the forces, or about what was being treated as the system, could sometimes help much more.

The records from Bastani’s experiment are suggestive in this respect. GPT Base users more often requested answers, and the authors interpreted their interaction and error analyses as consistent with copying solutions. GPT Tutor elicited more attempted answers and requests for help. The surveys, administered at the end of the sessions, revealed a further complication. GPT Base users did not perceive the disadvantage visible in their unassisted scores; GPT Tutor users rated their performance more favourably even though no significant exam advantage had been detected.¹ Students’ impressions were not a dependable guide to what the assistance had done for them.

But the evidence does not all point towards disappointment. In an introductory physics course at Harvard, Kestin and colleagues compared a purpose-built AI tutor with an established active-learning class.⁹ Their randomised crossover study covered two lessons, with students experiencing both conditions. The tutor incorporated scaffolding, targeted feedback and self-pacing. Median post-test scores were 4.5 in the AI condition and 3.5 in the classroom condition. Measured against the pooled pre-test median, the gain in the AI condition was more than twice as large. Seventy per cent of students spent less than the 60 minutes allocated to classroom learning on the AI activity; the median was 49 minutes. They also reported greater engagement and motivation.

That is an interesting result on its own terms. It concerns particular content, a carefully prepared tutor and tests taken after the lessons. The intervention also combined several differences, including the setting, pace and prerecorded introductions, so the study cannot tell us how much of the benefit came from the language model alone. Nor does it establish long-term retention. Within those limits, it is good reason to take the design of AI-supported instruction seriously. Efficiency and learning need not be enemies.

Programming provides a less reassuring case. Bassner and colleagues reported a three-arm randomised trial at the Technical University of Munich, with 275 introductory programming students in the reported sample.¹⁰ During a 90-minute exercise, students used either the scaffolded tutor Iris, unrestricted ChatGPT or conventional web resources without AI. Both AI groups achieved higher scores on automated tests of their code and reported less frustration. Neither showed a statistically significant advantage in knowledge gains or code comprehension. Only Iris showed an increase in intrinsic motivation. There was clearer evidence of better task performance than of better understanding, although the absence of a significant learning difference does not establish that the groups learnt exactly the same amount.

Three studies cannot carry a general theory of AI in education. Their students, tasks and measurements are too different, and their time horizons too short. What they do make difficult to ignore is the need to decide whose competence an assessment is measuring. If a tool remains available throughout, successful performance may tell us a great deal about the person–tool combination and rather less about a change in the person.

The larger reviews do not make this much easier. Boolzen and colleagues’ 2026 review identified 85 eligible controlled studies of cognitive outcomes in STEM; 49 studies, yielding 59 effect sizes, entered the meta-analysis.¹¹ A conventional random-effects model gave a positive mean effect, alongside extraordinary heterogeneity (I² = 96.32%). Its prediction interval ran from g = −1.52 to g = 3.20. For comparable settings, the model therefore allows for effects extending from substantial harm to substantial benefit. It would be a mistake to read the positive average as a reliable forecast for a new classroom intervention.

The authors also found funnel-plot asymmetry, consistent with possible publication bias. Using a robust Bayesian adjustment, they obtained an overall estimate close to zero (μ = 0.076), with substantial uncertainty. The heterogeneity remained. Asymmetry need not be caused by publication bias, and the adjusted estimate depends on modelling assumptions; it is not a final answer about whether AI helps students learn.

Their moderator analysis is more helpful for thinking about instruction. They coded the learning outcome and the change in students’ activity relative to the control condition. The patterns favoured augmentation over substitution, but these were comparisons across studies, not randomised tests of those design choices. Much variation remained unexplained, and no combination of the examined moderators reliably supplied a sufficient condition for a large effect.¹¹ The distinction is promising as a research question. We do not yet have a rule for applying it to every lesson.

Wu and colleagues’ 2026 meta-analysis sounds more optimistic: 35 studies, 4,193 participants and a positive mean effect of g = 0.670, with a 95% confidence interval of 0.495–0.844.¹² The two reviews are not estimating quite the same thing. Wu’s synthesis ranges across disciplines and includes motivation, engagement and self-efficacy alongside cognitive outcomes. It also reports very high heterogeneity (I² = 91.444%). Its tests did not detect publication bias, which is different from establishing its absence. Most included studies concerned higher education.

I would hesitate to choose between those reviews on the strength of their averages. An answer-giving chatbot, a tutor that asks for explanations and a debugging assistant may use closely related models while organising the student’s activity very differently. Asking for “the effect of ChatGPT on learning” begins to sound like asking for the effect of books. We would immediately want to know which books, who was reading them and what they were being asked to do.

Meanwhile, students are getting on with using the tools. PISA 2025 cannot resolve the causal questions raised by the experiments, but it gives some sense of their scale. On average across OECD countries, 46% of 15-year-olds reported using AI chatbots at least weekly to help them learn. In Spain, the figure was 54%. The Spanish percentages reported for preliminary research, summarising assigned texts and drafting written work were 42%, 40% and 33%, respectively. Only 8% reported never or almost never using AI for any of the schoolwork purposes examined.^13,14

The achievement patterns deserve attention, though they are easily overread. After adjustment for students’ socioeconomic profile, the OECD-average results generally favoured non-users over those reporting AI use for specific tasks such as drafting or summarising. Weekly use for the broader purpose of helping with learning was associated with roughly similar science performance to non-use. For some specific tasks, moderate users did better than very frequent users.¹³

A pupil who is struggling may seek more help. Schools may encourage AI use in response to difficulties that already exist. And frequency tells us very little about the encounter itself: requesting an answer, interrogating a hint and challenging an explanation can all count as using a chatbot. Those possibilities are not findings that explain away PISA’s associations; they are reasons why the associations cannot establish what AI caused.

PISA also examines opportunities to evaluate AI-generated information at school. Some descriptive patterns look favourable, but the overall positive association with that activity is not observed, on average, after adjustment for students’ and schools’ socioeconomic profiles.¹³ The report gives us no simple causal claim about the benefits of teaching students to check AI output.

There is nevertheless an educational question here that deserves more than a slogan. Mapping Ignorance has discussed it in Critical thinking for the unstoppable use of artificial intelligence. In STEM, asking a student to check an answer presupposes that something is available with which to check it. Does the student know what order of magnitude to expect? Which assumption matters? What evidence would bear on the explanation? A fluent paragraph can be physically impossible, but fluency alone will not tell a beginner where it goes wrong.

It is also worth keeping the chronology straight. Generative AI cannot be the sole explanation for declines in educational achievement that began before ChatGPT was publicly available. What PISA establishes more usefully here is that the practices under discussion are already part of ordinary schoolwork. Teachers have to make decisions while the evidence is still developing.

Some of those decisions will be quite small. A worked solution may be exactly what a pupil needs on Monday. A few weeks later, providing the whole solution may remove an opportunity to reconstruct the method. AI could make it easier to vary the support: ask for a representation before proposing an equation, invite a prediction before showing a simulation, or offer a flawed argument for the student to diagnose. These are plausible uses, not capabilities we should assume a system can deliver reliably simply because it holds a convincing conversation.

Assessment poses a related problem. Students will need to work competently with tools, as scientists and engineers do, and that competence deserves to be assessed. But if we also want to know whether they can perform a particular operation themselves, there must be an occasion when the tool is not doing it for them. We have to say what belongs in the learner’s own repertoire, and then find an assessment that gives us reasonable evidence of it.

For someone working in History and Philosophy of Science, the interest extends beyond the ownership of a correct answer. Scientific competence involves judgement about models, evidence, instruments and the reach of a procedure. A student may know a formula and still not know when it applies. An experienced scientist may quite reasonably delegate its numerical solution while retaining responsibility for the assumptions and interpretation. Learning to make those distinctions is part of becoming competent.

That brings me back to the classroom question. When I asked why a student had used an equation, I wanted access to the reasoning behind the calculation. With generative AI, the visible work may belong to a temporary partnership, and locating that reasoning becomes harder. The partnership itself need not be objectionable. What matters educationally is what repeated participation in it does to the student.

Over time, can the learner represent a problem more confidently, choose a method, recognise an implausible result or decide when help is needed? If so, the tool may be contributing to the acquisition of expertise. If those capacities scarcely change while the submitted answers become more polished, the same outward success has a different meaning.

I do not think the most useful question is whether students should use AI in STEM. It is what they should still be learning to do, how the assistance bears on that learning, and when it ought to recede. A correct answer was never a guarantee of understanding. We now have to be more deliberate about finding out what lies behind it.

 

 

 

References

1. Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö. & Mariman, R. (2025). ‘Generative AI without guardrails can harm learning: Evidence from high school mathematics’. Proceedings of the National Academy of Sciences, 122(26), e2422633122. 10.1073/pnas.2422633122
2. Soderstrom, N. C. & Bjork, R. A. (2015). ‘Learning versus performance: An integrative review’. Perspectives on Psychological Science, 10(2), 176–199. 10.1177/1745691615569000
3. Nersessian, N. J. (2002). ‘The cognitive basis of model-based reasoning in science’. In P. Carruthers, S. Stich & M. Siegal (eds), The Cognitive Basis of Science, pp. 133–153. Cambridge University Press. 10.1017/CBO9780511613517.008
4. Barbieri, C. A., Miller-Cotto, D., Clerjuste, S. N. & Chawla, K. (2023). ‘A meta-analysis of the worked examples effect on mathematics performance’. Educational Psychology Review, 35, Article 11. 10.1007/s10648-023-09745-1
5. Tetzlaff, L., Simonsmeier, B., Peters, T. & Brod, G. (2025). ‘A cornerstone of adaptivity – A meta-analysis of the expertise reversal effect’. Learning and Instruction, 98, 102142. 10.1016/j.learninstruc.2025.102142
6. Koedinger, K. R., Pavlik, P. I. Jr., McLaren, B. M. & Aleven, V. (2008). ‘Is it better to give than to receive? The assistance dilemma as a fundamental unsolved problem in the cognitive science of learning and instruction’. In B. C. Love, K. McRae & V. M. Sloutsky (eds), Proceedings of the 30th Annual Conference of the Cognitive Science Society, pp. 2155–2160. Full text
7. Risko, E. F. & Gilbert, S. J. (2016). ‘Cognitive offloading’. Trends in Cognitive Sciences, 20(9), 676–688. 10.1016/j.tics.2016.07.002
8. Giere, R. N. & Moffatt, B. (2003). ‘Distributed cognition: Where the cognitive and the social merge’. Social Studies of Science, 33(2), 301–310. 10.1177/03063127030332017
9. Kestin, G., Miller, K., Klales, A., Milbourne, T. & Ponti, G. (2025). ‘AI tutoring outperforms in-class active learning: An RCT introducing a novel research-based design in an authentic educational setting’. Scientific Reports, 15, 17458. 10.1038/s41598-025-97652-6
10. Bassner, P., Lenk-Ostendorf, B., Beinstingel, R., Wasner, T. & Krusche, S. (2026). ‘Less stress, better scores, same learning: The dissociation of performance and learning in AI-supported programming education’. Computers and Education: Artificial Intelligence, 10, 100537. 10.1016/j.caeai.2025.100537
11. Boolzen, C., Kuhn, J., Flegr, S., Rott, E.-M., Stausberg, N. & Küchemann, S. (2026). ‘Evidence of impact and interpretational limits of generative AI in STEM education: a systematic review and meta-analysis on cognitive learning outcomes’. Artificial Intelligence Review. Advance online publication. 10.1007/s10462-026-11665-9
12. Wu, X., Zhu, P., Zhang, J., Yin, M. & Wang, Y. (2026). ‘ChatGPT’s impact on student learning outcomes: a meta-analysis of 35 experimental studies’. Humanities and Social Sciences Communications, 13, 684. 10.1057/s41599-026-07019-z
13. OECD (2026). PISA 2025 Results (Volume I): Future-Ready Students. Paris: OECD Publishing. 10.1787/73451bc5-en
14. OECD (2026). PISA 2025 Results (Volume I): Spain. Country note. OECD. Country note

 

The post AI, problem solving, and the future of STEM education appeared first on Mapping Ignorance.

Author: José R. Pineda got his Ph.D. from University of Barcelona in 2006. Since 2007 he has worked for Institut Curie and The French Alternative Energies and Atomic Energy Commission. Currently he is a researcher of the EHU. He investigates the role of stem cells in physiologic and pathologic conditions.

 

One of the ideas most of us learn early in biology is that only a few specialized cells can communicate through electrical signals. Neurons do it. Muscle cells do it. Epithelial cells, the cells that form our skin and line most organs, supposedly do not. However, a recent study by Yu and Granick demonstrates that this picture may not be entirely correct. In fact, these seemingly ordinary cells appear capable of generating electrical spikes of their own when tissue integrity is disturbed. Despite  epithelial cells have traditionally been viewed as passive structures whose main functions are to form protective barriers, absorb nutrients, and maintain tissue integrity, Yu and Granick challenges this long-standing idea in a striking way. Their work demonstrates that epithelial cells can generate electrical spikes resembling action potentials, revealing a previously unrecognized form of bioelectrical communication that may help coordinate tissue responses after injury ¹.

The finding is surprising because epithelia are among the most abundant tissues in the body. They line the skin, digestive tract, airways, and many internal organs. Although these tissues are known to participate in wound healing and cellular communication, they were not believed to possess the kind of rapid electrical activity usually associated with the nervous system. The new study suggests that this assumption may have been incomplete. Under certain conditions, epithelial cells appear capable of producing coordinated electrical signals that spread across remarkably large distances. To investigate this phenomenon, the researchers employed two different epithelial models: the first model consisted of primary human skin keratinocytes (the major cellular component of the epidermis). The second model used Madin-Darby Canine Kidney (MDCK) cells (a classical epithelial cell line frequently employed in cell biology research). Both cell types were grown until they formed dense, confluent monolayers resembling miniature sheets of epithelial tissue.

The key experimental tool was a multielectrode array, often abbreviated as “MEA”. This technology can be imagined as a highly sensitive electronic “listening device” containing dozens of microscopic electrodes embedded in a culture surface. When cells generate electrical activity, even extremely small voltage fluctuations can be detected simultaneously at multiple positions. Multielectrode arrays are routinely used to study neurons, but their application to epithelial tissues is relatively uncommon. After establishing epithelial monolayers on these arrays, the researchers created a small and highly localized injury using a focused laser beam. The laser produced controlled damage in a tiny region at the center of the culture while leaving surrounding cells intact. This experimental design allowed the team to observe how cells located at various distances reacted to the injury. Electrical signals were then monitored continuously for one hour. What they observed was unexpected: within the epithelial sheets, electrical spikes appeared shortly after injury and persisted for more than sixty minutes!! These spikes displayed a characteristic sequence of depolarization, repolarization, and hyperpolarization, terms commonly used to describe action potentials in neurons. However, there was one major difference, while neuronal action potentials last a few milliseconds, the epithelial spikes unfolded over one to two seconds, making them roughly one thousand times slower. Despite their slower kinetics, the electrical signals were surprisingly robust, measurements showed that spikes could be detected as far as 740 micrometers from the injury site, a distance corresponding to dozens of cell diameters (Figure 1). Even more remarkably, their amplitude did not progressively decrease with distance as one would expect from a simple wave spreading outward from a central source. In some cases, signals recorded farther away were as large as, or even larger than, those detected closer to the wound. This observation suggested that distant epithelial cells were not merely receiving a weakened signal from injured neighbors. Instead, they may have been generating their own electrical responses as part of a collective tissue-wide process.

 

Discovering the spikes was only part of the story. The obvious next question was what actually produced them. Were epithelial cells using the same electrical machinery found in neurons, or was something entirely different at work? To answer this, the researchers began dissecting the molecular mechanisms one by one. They began by testing the role of calcium (one of the most important intracellular messengers in biology). Calcium ions regulate a wide range of cellular activities, including contraction, migration, secretion, and gene expression. To determine whether calcium was necessary for epithelial spiking, the researchers added EDTA (a molecule that binds calcium and reduces its availability). Surprisingly, once calcium availability was reduced, the electrical activity virtually vanished!! This immediately suggested that calcium was not just associated with the phenomenon but was an essential part of it. Next, the team examined the contribution of cellular mechanics, because it is known that epithelial cells continuously generate mechanical forces through interactions between actin filaments and myosin motor proteins, a system known as “the actomyosin network”. This machinery allows cells to contract, change shape, and exert forces on neighboring cells. To disrupt this process, the researchers treated cultures with blebbistatin (a selective inhibitor of myosin II activity). Once again, electrical spiking was largely abolished. The similarity between the effects of calcium depletion and myosin inhibition suggested that electrical activity was intimately linked to the mechanical state of the tissue. This observation raised an intriguing possibility: perhaps epithelial electrical signaling is not driven by voltage-gated ion channels (as in neurons), but instead by channels that respond to mechanical forces. To test this hypothesis, the investigators focused on mechanosensitive ion channels, specialized proteins embedded within cell membranes that open when cells are stretched, compressed, or subjected to mechanical stress. Using the inhibitor GsMTx4 (which blocks several mechanosensitive channels including Piezo1), the researchers observed a partial reduction in electrical activity. While spiking was not completely eliminated, both amplitude and coordination decreased substantially. This result implicated mechanosensitive channels as key components of the signaling pathway. To really prove their implication, next the researchers activated them directly using specific pharmacological compounds. One molecule, GSK1016790A, activates the TRPV4 channel. Another, Yoda-1, specifically activates Piezo1. Remarkably, activating these channels was sufficient to generate large electrical spikes even in the absence of any injury. In other words, simply stimulating mechanosensitive channels could reproduce the electrical behavior observed after tissue damage. This provided strong evidence that mechanical sensing lies at the heart of epithelial electrical excitability.

 

Based on their findings, the authors proposed a positive feedback mechanism: 1) mechanical stress activates mechanosensitive ion channels, allowing calcium to enter the cells. 2) rising calcium levels stimulate actomyosin contraction, generating additional mechanical stress. 3) this new stress then activates more mechanosensitive channels, sustaining the cycle. Such a feedback loop could explain why spiking persists for long periods and why cells distant from the wound remain electrically active. One particularly interesting aspect of the study is its potential relevance to wound healing. Successful tissue repair requires large numbers of cells to coordinate their behavior across considerable distances. Cells must migrate, proliferate, and remodel tissue structures in a synchronized manner. Traditional explanations have emphasized biochemical signaling molecules and calcium waves. The electrical spikes described here introduce an additional layer of communication that may operate on intermediate spatial and temporal scales. According to the authors, these signals could complement chemical and mechanical communication systems, helping tissues organize collective responses after injury. However, the researchers are careful to acknowledge the study’s limitations. Their experiments were performed in cultured epithelial monolayers rather than living organisms. Although the results clearly demonstrate intrinsic electrical excitability in epithelial cells, it remains unknown whether the same phenomenon occurs in intact tissues during natural wound healing. Furthermore, the exact relationship between these electrical spikes and cellular behavior has not yet been determined. It is still unclear whether the spikes actively drive tissue repair or simply accompany mechanical disruption. Additional work involving animal models, voltage-sensitive imaging, and targeted genetic manipulation of mechanosensitive channels will be required to answer these questions.

 

In conclusion, the study by Yu and Granick reminds us that biology still contains unexpected surprises, even in tissues that have been studied for decades. A simple layer of epithelial cells, once considered electrically silent, may in fact possess its own form of excitability. Rather than transmitting information through millisecond action potentials like neurons, epithelial tissues appear to communicate through slower, mechanically driven electrical spikes capable of coordinating activity across hundreds of micrometers. Whether these signals prove to be a central component of tissue physiology remains to be seen. What is already clear, however, is that the electrical life of cells extends far beyond the nervous system!!

The post When skin cells behave like neurons: The surprising discovery of electrical spikes in epithelia appeared first on Mapping Ignorance.

Authors: Mary M. Hausfeld, Assistant Professor in Management, University of Limerick and Jochen Menges, Director of the Center for Leadership in the Future of Work, University of Zurich

The post Women are more emotionally intelligent than men, and that’s a problem appeared first on Mapping Ignorance.

 

Authors: Eider Fernandez & Igor Baroja-Careaga, Esneki Zentroa (Leartiker S.COOP)

 

Probiotics are part of a rapidly expanding field of research. They are live microorganisms that, when administered in adequate amounts, may confer health benefits[1]. Among the most extensively studied probiotics, some species of Lactobacillus and Bifidobacterium have been associated with the maintenance of gut microbiota balance and with several physiological functions[2]. Producing these microorganisms usually requires nutrient-rich culture media. But what if such bacteria could be obtained from a food by-product nowadays regarded as waste? This question formed the starting point of a study carried out by Esneki Dairy Centre – Leartiker to assess the potential of cheese whey as a substrate for producing microbial biomass enriched in microorganisms with probiotic potential.

The challenge of cheese whey

Cheesemaking inevitably generates substantial amounts of whey: a yellowish liquid released from the curd that still carries lactose, soluble proteins, minerals and other valuable compounds. In fact, whey can represent around 85% of the original milk volume used in cheese production[3].

Despite this nutritional richness, whey remains a difficult stream to manage for many artisanal dairies. This is particularly true in the Basque Country, where many cheesemaking farms are small, family-run businesses located in rural areas[4]. Its highly perishable nature means that whey must be preserved or processed soon after production, making storage, collection and transport especially challenging when volumes are small and geographically dispersed.

Moreover, uncontrolled whey discharge results in a high organic load that can affect aquatic ecosystems. Paradoxically, the same nutrients that make cheese whey a potential pollutant are also what give it considerable biotechnological value.

A natural culture medium

From a microbiological standpoint, cheese whey is not merely a residual stream, but a naturally nutrient-rich matrix[5]. Its lactose content, together with soluble proteins, minerals and other available compounds, provides a favourable environment for the growth of lactic acid bacteria, widely used microorganisms in fermented foods and increasingly explored for their probiotic potential.

The working hypothesis was simple: by steering whey fermentation under controlled conditions, it could be possible to recover a biomass enriched in microorganisms of interest, while at the same time increasing the stability and value of the by-product itself.

To explore this possibility, several fermentation strategies were tested using whey generated during the production of blue cheese from pasteurised cow’s milk. The study compared three different approaches:

• Fermentation of previously pasteurised whey (Protocol 1).
• Fermentation of whey after a deproteinisation step (Protocol 2).
• Direct fermentation of raw whey, without any prior treatment (Protocol 3).

Two different blends of commercial microorganisms, inspired by the naturally diverse microbial communities found in kefir, were also evaluated. These included; Choozit Kefir DV LYO 1000L (Danisco) (starter culture A) and Ferlac Kefir Type C (Abiasa) (starter culture B). Additionally, unfermented whey (that is, whey to which no starter cultures were added and which was not subjected to fermentation) from each protocol was analysed as a control sample. Controlled fermentation processes were carried out under monitored conditions. Throughout the fermentation, key parameters were monitored, including temperature (25 ± 1 °C), incubation time (16 ± 1 h for starter culture A and 21 ± 1 h for starter culture B) and pH evolution, with fermentation being considered complete once the pH fell below 4,6. These measurements allowed the identification of the conditions that most effectively supported the growth and enrichment of the target microbial populations.

 

Exploring the microbiome through metagenomics

However, assessing the success of a fermentation process requires more than simply determining how many microorganisms grow. It is also essential to identify which species are present and how the microbial community changes over time.

Traditional culture-based microbiological methods remain fundamental tools for quantifying and isolating viable microorganisms. Yet they have important limitations when complex microbial communities need to be characterised, as they only detect microorganisms able to grow under the selected culture conditions. As a result, part of the microbial diversity present in the sample may remain undetected. To obtain a more comprehensive view of the microbial community, the study incorporated high-throughput amplicon sequencing, an approach that enables the direct analysis of DNA present in a sample and the identification of microorganisms regardless of their ability to grow under laboratory conditions.

Bacterial communities were characterised by amplifying the V4 region of the 16S rRNA gene, one of the most widely used markers in bacterial diversity studies. Fungal and yeast communities were analysed through the ITS region — the Internal Transcribed Spacer — which is considered the standard marker for fungal diversity studies.

Sequencing was performed using Illumina technology, generating approximately 100,000 reads per sample for bacteria and around 40,000 reads for fungi and yeasts. These sequences were then compared against specialised databases to determine the identity of the microorganisms present and estimate their relative abundance.

Before interpreting the results, sequencing depth was assessed using rarefaction curves. These curves showed a clear trend towards saturation, indicating that most of the microbial diversity present in the samples had been adequately captured.

Which protocol worked best?

High-throughput amplicon sequencing revealed clear differences among the fermentation strategies tested.

Fermentations carried out using raw whey (Protocol 3) maintained a more diverse bacterial community than those in which the whey had first been subjected to thermal treatment (Protocol 1 and Protocol 2), as can be seen in Figure 1, as can be seen in Figure 1, the number of observed microbial taxa increases with sequencing depth. This suggests that the partial preservation of the native microbiota originating from the cheesemaking process may play an important role in shaping the fermentation outcome.

 

 

 

In contrast, the diversity of moulds and yeasts decreased. This trend may be related to the origin of the whey, which was obtained from blue cheese production, where specific mould (Penicillium roquefoti) species are deliberately enriched during manufacture. The competitive advantage of these dominant fungi appears to favour the development of a more specialised microbial community, reducing the abundance and diversity of other microbial populations present in the whey (Figure 2).

 

 

Principal coordinates analysis (PCoA) based on Jaccard distances (Figure 3) showed a clear separation among the different fermentation protocols, indicating that each strategy gave rise to distinct microbial communities. In parallel, the relative abundance of profiles (Figure 4) made it possible to visualise how particular species increased or decreased depending on the conditions applied.

 

 

 

 

Amplicon sequencing also enabled the identification of the dominant microorganisms during the process. In most samples, Lactococcus lactis was particularly abundant. This result was expected, as this bacterium was included in the commercial starts cultures used and is widely employed in dairy fermentations. However, some fermentation conditions, particularly those involving starter culture B, also favoured the development of species such as Lacticaseibacillus rhamnosus and Bifidobacterium animalis, wich were also present in the commercial starter culture. These microorganisms are frequently associated with probiotic applications[6]^,[7]^,[8].

To identify the most suitable protocol for producing a fermented whey enriched in microorganisms with probiotic potential, both microbial abundance (including aerobic mesophilic bacteria, yeasts and moulds) and the diversity of the resulting microbial communities were evaluated. The highest microbial counts were obtained in whey samples produced using Protocol 3, which was based on raw whey, and fermented with starter culture B. (Average counts of 8.2 log CFU/mL were obtained with starter culture A, and average counts of 8.5 log CFU/mL with starter culture B).

Furthermore, as mentioned previously, the highest levels of microbial diversity were observed in the samples produced using Protocol 3, where fermentation was carried out directly on raw whey without prior heat treatment.

 

Overall, the results pointed to the most effective strategy also being the simplest from a technological perspective: the direct fermentation of raw whey, without prior thermal treatment using starter culture B. This approach maintained high microbial diversity while reducing energy requirements and simplifying the overall process.

An opportunity for the circular economy

Beyond the microbiological results, the broader relevance of this work lies in showing that the role of whey within the cheesemaking chain can be reconsidered.

Rather than being treated as a waste stream associated with management costs, whey can be viewed as a raw material capable of supporting valuable biotechnological processes. Controlled fermentation not only helps stabilise whey but also makes it possible to harness the nutrients it still contains to generate added-value microbial biomass.

Such strategies are closely aligned with the principles of the circular bioeconomy, in which by-products are no longer seen as the end of a production chain, but as the starting point for another. In a context where sustainability and efficient resource use are increasingly important, initiatives of this kind show how microbiology and omics technologies can help turn an environmental challenge into an opportunity for innovation.

 

The PRO2O project (grant agreement No. 00003-BIO2022-48) was funded through the 2022 Bioeconomy Innovation Projects funding programme of the Basque Government’s Department of Economic Development, Sustainability and Environment.

References

[1] Guarner F, Requena T, Marcos A. Consensus statements from the Workshop “Probiotics and Health: Scientific evidence”. Nutr Hosp. 2010 Sep-Oct;25(5):700-4. PMID: 21336423.

[2] Martin, A. J. M., Serebrinsky-Duek, K., Riquelme, E., Saa, P., & Garrido, D. (2023). Microbial interactions and the homeostasis of the gut microbiome: the role of Bifidobacterium. Microbiome Research Reports, 2. https://doi.org/10.20517/mrr.2023.10

[3] González Siso, M. I. (1996). The biotechnological utilization of cheese whey: A review. Bioresource Technology, 57(1), 1–11. https://doi.org/10.1016/0960-8524(96)00036-3

[4] Esneki Zentroa – Leartiker (2022). “El sector quesero en Euskadi 2021”. Available at: El-sector-quesero-en-Euskadi.pdf

[5] Figueroa Pires, A., garcía Marnotes, N., Díaz Rubio, O., Cobos García, A., Dias Pereira, C. (2021). Dairy by-products: A review on the valorization of whey and second cheese whey. Foods, 10: 1067

[6]Neethu, M., Bunt, C., Hussain, M., 2015. Comparison of Microbiological and Probiotic Characteristics of Lactobacilli Isolates from Daity FOod Products and Animal Rumen Contens Microorganims 3, 198-212.

[7] Rama, G.R., Kuhn, D., Beux, S., aciel, M.J., Volken de Souz, C.F., 2019. Potential apllications of dairy why for the production of lactic acid bacteira cultures. Int Dairy J.

[8] Rodriguez-Palacios, 1., Staemplfli, H.R., Duffield, T., Weese, J.S., 009. Isolation of bovine intestinal Lactobacillus plantarum and Pediococcus acidilactici with inhibitory activity against Escherichia coli O157 and F5. J. Appl. Mcrobiol. 106, 393-401

 

The BRTA is a consortium that remains a step ahead of future socio-economic challenges worldwide and in the Basque Autonomous Community; it addresses them through research and technological development, thus projecting itself internationally. The BRTA centres collaborate to generate knowledge and transfer it to Basque society and industry so as to make them more innovative and competitive. The BRTA is an alliance of 17 R&D centres and cooperative research centres with the support of the Basque Government, the SPRI and the Chartered Provincial Councils of Araba, Bizkaia and Gipuzkoa.

The post Turning cheese whey into value: a sustainable substrate for probiotic biomass production appeared first on Mapping Ignorance.

Author: Deven Desai, Professor of Business Law and Ethics; Associate Director for Law, Policy, and Ethics at the Machine Learning Center, Georgia Institute of Technology

The post The ‘WarGames’ problem: AI agents don’t go rogue appeared first on Mapping Ignorance.