Can We Understand How LLMs Reason? — ACM, Causal Abstraction, and J-Space
Communications of the ACM (Jul 7, 2026) on Thomas Icard, Atticus Geiger, and causal abstraction — plus Goodfire's "Arithmetic in the Wild." explainx.ai connects academic mechanistic interpretability to Anthropic's J-lens and J-space.
Large language models write essays, prove lemmas, and ship code — but we still cannot fully explain how.
That is the opening of Communications of the ACM, July 7, 2026: engineering is ahead of science. Stanford philosopher-computer scientist Thomas Icard and the mechanistic interpretability community are trying to close the gap with causal abstraction — asking when a billion-parameter network implements a higher-level algorithm you would recognize from logic or cognitive science.
Six days earlier, Anthropic published J-space — a privileged internal workspace in Claude, instrumented with the J-lens, with swap interventions that change answers. Same scientific program, different venue: academia on Llama, frontier lab on Claude.
This post maps the ACM story, embeds explainx.ai's J-lens and J-space deep dives, and states what builders should do with partial understanding.
Silent fake/fictional before output; swap France→China
Anthropic J-lens / J-space
Icard's bottom line: interpretability may never yield a closed-form equation for GPT-class models — but hidden algorithms can become partly legible.
Causal abstraction — ideal gas for neural nets
Icard's analogy (via ACM): physicists abstract colliding molecules to pressure and temperature. Neural nets can likewise be described above matrix multiplies — with much messier overlap, because one neuron can participate in many functions at once.
The precision tool is causal abstraction:
When do two causal models in different languages describe the same underlying reality — one more abstract than the other?
Applied to LLMs: you hypothesize a high-level algorithm (logical inference, cyclic arithmetic), then test whether internal states factor like that algorithm's variables. Not "it got the answer right" — the mechanism matches.
Atticus Geiger (Icard's former Ph.D. student, now a leading interpretability researcher at Goodfire) helped build this framework. The field's progress since 2021 spans training-data reflection, post-training reshaping, and links between internal abstractions and generalization — the same arc Mandy Lu's "why does AI work?" debate highlights from the theory side.
BERT learned logic (2021)
In a 2021 study, Icard, Geiger, and collaborators showed a BERT-based language model internally implements pieces of a logical reasoning system — including inferences with quantifiers (every, some, not) and negation.
That was early proof the question is not only philosophical ("does it mimic reasoning?") but empirical: you can align internal computation to symbolic steps.
"Arithmetic in the Wild" — August + 6 = February
Goodfire's recent paper (Geiger-led, cited by Icard in ACM) asks: What month is six months after August?
Naive expectation: reason directly on a 12-month cycle.
Observed mechanism in a Llama-based model:
snippet
August → internal month index 8
→ decimal addition: 6 + 8 = 14
→ map 14 back: 14 = 12 + 2 → February
The same template appears for weekdays and clock time — one general strategy, multiple domains, never explicitly trained as "use decimal first."
Icard: "A beautiful example of using causal abstraction methods to uncover how a large language model reasons about cyclic concepts."
explainx.ai note: This is why private enterprise evals must include weird edge cases — models can pass benchmarks while running surprising internal routes that break under distribution shift.
Open weights vs closed frontier — who can look inside?
From ACM:
Access
Models
Intervention depth
Academia
Llama, OLMo (~10B params)
Change weights / activations
Frontier labs
Claude, Gemini-class
In-house teams; no public weights
Builders
Open weights + tools
J-lens repo, Neuronpedia demos
Icard: commercial models are off-limits, but Anthropic and Google DeepMind have dedicated interpretability groups — "we regularly talk with both."
That asymmetry feeds Nadella's trust boundary argument: vendors see more of the model than you do; you must own evals, traces, and monitoring on your side.
The J-lens (Jacobian lens) is Anthropic's July 2026 instrument for the same scientific goal as causal abstraction — but operationalized for frontier Claude:
text
Forward: activations → layers → logits → (maybe) token
J-lens: which activation direction pushes logits toward word W later?
For each vocabulary token, find the internal pattern that increases future probability of saying that word. Stack layer readouts → a live list of silent "words on Claude's mind."
Question
Answer
J-lens vs J-space?
Tool vs workspace the tool reveals
vs chain of thought?
CoT is written; J-lens reads pre-output activations
"For every word in Claude's vocabulary, the J-lens finds the internal activity pattern that makes Claude more likely to say that word at some point in the future."
Causal-abstraction parallel: Goodfire explains month arithmetic as a high-level two-step algorithm; J-lens explains France capital questions as workspace variables you can surgically replace — same epistemology, different granularity.
Embedded guide — Anthropic's J-space global workspace
On July 6, 2026, Anthropic argued Claude has a strikingly similar divide to human conscious access: most processing is automatic; a small privileged channel holds reportable, controllable concepts — the J-space.
Anthropic's July 2026 video on J-space — companion to the global-workspace research post and explainx.ai's embedded guides.
TL;DR — J-space at a glance
Question
Answer
What is it?
Dozens of word-linked patterns forming an internal global workspace
How found?
J-lens readouts across layers
vs CoT
Silent — steps can appear without visible text
Scale
Under 10% of internal activity; dozens of concepts at once
Academia and frontier labs are converging on causality, not just saliency maps. Enterprises sit on the consumption side: you get behavior from APIs, occasionally research blog visibility, rarely internal graphs.
explainx.ai read
"Can we understand reasoning?" — Partially, and improving fast; not completely
Causal abstraction — The right standard (mechanism match, not answer match)
Open weights — Science runs on Llama; production runs on Claude/GPT — plan for opacity
Enterprise moat — Own evals + traces; treat vendor interpretability as R&D signal
Science is catching up to engineering. Your job is not to wait for the full theory — it is to log, test, and intervene on your workflows while the field maps the rest.
Research citations and model access reflect publication-week status, July 2026. Open-weight tooling and commercial model internals change frequently — verify repos and terms before production monitoring designs.
