I'm not sure it's entirely fair to characterize "hypothesiz[ing a] link between fixed physical characteristics and some hypothesized inborn tendency or trait" as pseudoscientific "phrenology", since in some cases there are known correlations between these things. Various genetic diseases (e.g. Down syndrome, Williams syndrome, hydrocephalus) cause both mental disabilities and noticeable differences in physical characteristics. There is some evidence (reviewed in https://web.archive.org/web/20210611221151/https://fileleaks.com/file/b41c83251265fdfb29fe062563ca7ea4757a2914/yeoal2007developmentalInstabilityindividualVariationInBrainDevelopment.pdf) that bodily asymmetry is associated with lower IQ & higher rates of some mental illnesses, since harmful mutations in genes with diverse effects can contribute to each of these.
I agree, though, that any prediction algorithm based on such correlations should be used only as a cue for a human to investigate, rather than as a decision-maker in itself, unless it is precise enough that its accuracy wouldn't be significantly improved by human review.
I'm not sure it's entirely fair to characterize "hypothesiz[ing a] link between fixed physical characteristics and some hypothesized inborn tendency or trait" as pseudoscientific "phrenology", since in some cases there are known correlations between these things. Various genetic diseases (e.g. Down syndrome, Williams syndrome, hydrocephalus) cause both mental disabilities and noticeable differences in physical characteristics. There is some evidence (reviewed in https://web.archive.org/web/20210611221151/https://fileleaks.com/file/b41c83251265fdfb29fe062563ca7ea4757a2914/yeoal2007developmentalInstabilityindividualVariationInBrainDevelopment.pdf) that bodily asymmetry is associated with lower IQ & higher rates of some mental illnesses, since harmful mutations in genes with diverse effects can contribute to each of these.
I agree, though, that any prediction algorithm based on such correlations should be used only as a cue for a human to investigate, rather than as a decision-maker in itself, unless it is precise enough that its accuracy wouldn't be significantly improved by human review.