The frame problem with 'how smart'
There is no IQ test for crows. There can't be one, because human IQ tests are built around the cognitive capacities humans developed under selection pressures crows did not face. Asking 'how smart is a crow' relative to a human is like asking how fast a deer is relative to a fish — meaningful only if you've specified what 'fast' or 'smart' means in a context that fairly captures both. Researchers who study animal cognition have largely abandoned the ranking question and replaced it with capacity-by-capacity assessment: what specific cognitive operations can this species perform, and how do those compare to other species along the same operation?
Individual recognition: American crows recognize and remember individual human faces for years (Marzluff mask experiments).
Capacities crows demonstrably have
Individual recognition: American crows recognize and remember individual human faces for years (Marzluff[1] mask experiments). Social information transfer: face-recognition information moves between crows, including to crows not present at the original event. Episodic-like memory: scrub jays (close cousins) demonstrate WHAT, WHERE, and WHEN memory in cache-recovery tasks (Clayton et al.); American crows show analogous patterns less rigorously tested. Tool use: New Caledonian crows shape hooked tools from materials; American crows show occasional but not species-typical tool use. Numerical competence: jackdaws and other corvids have been trained to perform numerical discrimination tasks (Pepperberg's work establishes the avian ceiling). Future planning: scrub jays demonstrate caching behavior consistent with anticipated future need.
Cognitive capacities crows almost certainly have but haven't been rigorously tested for
Long-term grudges (anecdote-strong, experiment-weak for American crows specifically). Mental representation of absent objects (suggested by various behaviors but not rigorously isolated). Theory of mind to some degree (scrub jay caching-while-watched experiments hint at it; American crow extensions less established). Cultural transmission of non-tool behavior (face recognition spreads, suggesting other social information may also). Vocal-meaning compositionality (statistical hints, no rigorous behavioral test). The list of 'almost certainly has but not yet rigorously tested' is uncomfortable for science — it admits that cognitive capacity outpaces the experimental literature in this species.
The brain-size question
American crows weigh about 450 grams with a brain weight of roughly 9 grams. That puts brain-to-body weight ratio on the high end for birds, comparable to or exceeding many small primates. More importantly, the corvid brain has an unusually developed nidopallium caudolaterale — the avian functional analogue of mammalian prefrontal cortex, the region most associated with cognitive flexibility and decision-making. Convergent evolution between mammalian and avian brain architectures means brains that look very different from outside can support similar cognitive operations. Crows aren't smart 'despite' their bird brains; they're smart because their bird brains are specifically built for the capacities we're measuring.
What 'seven-year-old child' actually means
The popular comparison 'crows are as smart as a seven-year-old child' traces back to a few specific experimental results, mostly Sarah Jelbert and Nicola Clayton's water-displacement experiments showing New Caledonian crows can solve Aesop's-fable-style problems at roughly the level of a seven-year-old. The comparison is real but narrow — it covers physical-cognition problem-solving, not language, social cognition, abstract reasoning, or most of what we mean when we say a child is smart. Generalizing from 'comparable on this task' to 'as smart as overall' is the kind of move careful science avoids and popular coverage usually doesn't.
What the AI bioacoustics work adds
Contemporary -based AI methods don't directly measure cognition. They measure acoustic structure. But they bear on the question indirectly by characterizing the vocal repertoire's complexity. A species with nine acoustically distinct emergent clusters, individual signatures recoverable from a single call, and statistical evidence of sequence structure has more communicative bandwidth than a species with three coarse alarm-call categories. Whether that bandwidth gets used for the cognitive operations the brain is capable of remains open. But the bandwidth is there, and the bandwidth is a precondition for any future demonstration that the cognition uses it.
The honest position
Crows are highly intelligent by every operational measure researchers have applied to them across multiple species. The ranking 'how intelligent compared to X' is more often a popular-journalism move than a scientific one. What's defensible: crows demonstrate individual recognition, social learning, episodic memory, problem-solving, and (in New Caledonian crows specifically) tool culture at levels that establish the family as cognitively sophisticated. What's not yet defensible: any specific ranking against great apes, cetaceans, or human children that doesn't carefully specify the cognitive operation being compared. CrowLingo's editorial floor: take the demonstrated capacities seriously; don't make ranking claims the science doesn't yet support.