AI can genuinely amplify human thinking, or it can quietly flatten it, and which one happens depends almost entirely on who is using it and at what point in the learning process. That's the central argument Dr. Sarah Chardonnens, a professor at the University of Freiburg with a PhD in the science of learning, makes on this episode of Unscripted. Her research draws a sharp line between AI as an amplifier for people who already have expertise to challenge it, and AI as a shortcut that can prevent novices from ever building the cognitive architecture they need in the first place.
Who is Sarah Chardonnens?
Sarah Chardonnens is a professor at the University of Freiburg specializing in the science of learning, and the author of "The Learning Revolution." Before her academic career, she trained as a concert musician across multiple instruments and as a martial artist, experiences she credits directly with shaping how she thinks about discipline, effort, and the emotional and physical dimensions of building a skill. She developed the Synapse model, a four-phase framework describing how learning actually happens in the brain, in direct response to the arrival of large language models, and now works both with future teachers in training and on broader AI literacy efforts for schools, industry, and policymakers.
Does AI make us smarter, or just make us feel smarter?
Chardonnens' research shows a clear pattern of cognitive erosion in learners who over-rely on large language models, but she's careful to note the effect isn't uniform. Experts can use AI to challenge their own thinking, catch errors, and extend their performance. Novices, by contrast, still need to build their own foundational knowledge and neural architecture before AI can safely enter the process. The core problem, in her view, isn't the technology itself, it's a mismatch in speed: AI tools evolve far faster than research can study them, publish findings, and translate those findings into guidance for schools, parents, and policymakers. That gap means most of society is making decisions about AI and learning without the science having caught up yet.
What is the Synapse model and how does it explain learning?
Chardonnens developed the Synapse model to answer a more fundamental question first: what does intelligence actually require, and what does a person need to develop in order to remain cognitively strong as AI becomes more capable. The model describes learning in four phases: sensory input, where a learner makes initial connections to existing knowledge; network adaptation, where practice, error, and struggle actually build neural pathways; participation, where a learner manages their own learning process, recognizing when to rest, change strategy, or ask for help; and storage and embodiment, where knowledge becomes consolidated enough to build new understanding on top of it. Motivation sits at the center of all four phases. Introducing AI at the wrong phase, particularly the sensory input stage, can overwhelm a learner with more information than they have the expertise to filter, short-circuiting the very process the other phases depend on.
Why can the same AI tool help an expert and hurt a novice?
The distinction comes down to what each person brings into the interaction. An expert already has the internal framework to evaluate an AI's output, spot its errors, and push it further. A novice doesn't yet have that framework, and using AI to skip the struggle of building it means skipping the part of the process where actual learning happens. Chardonnens is direct about this: asking an LLM to produce a text is receiving a product, not participating in a process. She sees the same principle apply outside education entirely, wherever someone is tempted to let AI complete a task before they've built the underlying capability themselves.
What does it look like when AI amplifies learning correctly?
Chardonnens offers a concrete example from her own teaching practice. She built a closed AI system trained on her didactic materials, references, and teaching competencies, and has trainee teachers describe a lesson they want to build, then interact with the system as it asks questions and helps them refine their thinking. The effort and the struggle stay with the student. The AI's role is to challenge and extend, not to generate the finished product. She applies the same principle to students writing a persuasive essay: they should draft their own initial argument first, then bring it to an AI to be challenged, offered alternatives, and pushed to justify their choices, arriving at a final piece of writing that reflects a real thinking process rather than a copy-paste result. The distinction, in her framing, is whether a teacher is evaluating a result or engaging with the process that produced it.
What's the one human capacity we most need to keep training on purpose?
Chardonnens' answer is human agency: the capacity to direct your own attention, exercise judgment, and take deliberate action, rather than defaulting to whatever an AI system produces. She's candid that this is difficult to protect precisely because language is so tightly linked to thought. When you delegate the search for the right words to a system, you're not just delegating a task, you're changing how your own thinking gets shaped. Her comparison is to the industrial revolution's effect on the body: as machines took over physical labor, gyms and fitness culture emerged as a societal response. She sees the same reckoning coming for cognitive fitness, a deliberate practice of keeping the brain's own reasoning, calculating, and reading capacity active even as AI becomes capable of doing much of that work.
FAQ
Does AI make us smarter or just make us feel smarter, according to Sarah Chardonnens? It depends on the learner. Experts can use AI to genuinely extend their thinking, while novices risk skipping the struggle that builds foundational cognitive skills, producing fluent-sounding output without real understanding underneath.
What is the Synapse model? A four-phase framework developed by Sarah Chardonnens describing how learning happens: sensory input, network adaptation, participation, and storage and embodiment, with motivation at the center, used to determine when AI helps versus when it interferes.
Who is Sarah Chardonnens? A professor at the University of Freiburg with a PhD in the science of learning, author of "The Learning Revolution," and a former concert musician and martial artist who developed the Synapse model in response to the arrival of large language models.
Why can the same AI tool amplify an expert and flatten a novice? Because experts already have the internal framework to evaluate, challenge, and extend an AI's output, while novices haven't yet built that framework, so using AI to skip the struggle of building it prevents the learning from happening at all.
What is the most important human capacity to protect as AI takes on more cognitive work? Human agency, meaning the ability to direct your own attention, exercise independent judgment, and take deliberate action, rather than defaulting to whatever an AI system produces.
Watch the full conversation and explore more episodes at Unscripted with Jeff Pedowitz.