Five Minds for the Future by Gardner: An Evidence Check

Every claim here is checked against the primary research. How we check.

TL;DR

«Five Minds for the Future» names five dispositions and says schools and employers must cultivate them. Nineteen years on, we found exactly one instrument in the literature that claims to measure all five: a 66-item scale from a single Turkish sample of 528 secondary students (Gelen, Educational Research and Reviews, 2015), never replicated and never independently validated.
Gardner concedes the measurement problem himself, in writing, about the mind he considers most important: «synthesizing capacities are neither systematically trained nor regularly and reliably assessed» (howardgardner.com).
The book’s authority rests on multiple intelligences. When that theory was tested directly — 16 tests, 200 adults — the cognitive intelligences loaded heavily on one general factor, which is the opposite of independence (Visser, Ashton & Vernon, Intelligence, 2006).
Gardner says plainly that multiple intelligences are not learning styles, and he is right to. It has not helped: 89.1% of educators across 37 studies and 15,405 respondents still endorse matching instruction to styles (Newton & Salvi, Frontiers in Education, 2020), a practice with essentially no valid supporting evidence.
What survives is narrower than the book and was established elsewhere: creativity training raises divergent-thinking scores (Δ = 0.75 across 70 studies) but real performance far less (Δ = 0.35), and business ethics instruction has «minimal impact» across 25 programmes.

Five Minds for the Future by Howard Gardner — cover

Verdict

Read our notes instead. Gardner is a serious scholar and «Five Minds for the Future» is an honest book — it is explicitly labelled as a normative essay, not a research report. The problem is what happened next. The five minds entered corporate training decks, curriculum documents and leadership workshops as though they were a validated framework. They are not. They have never been shown to be five separable things, and no one has demonstrated that teaching one of them produces the disposition in any setting other than the one it was taught in.

The vocabulary is useful. The 200 pages that deliver it are not necessary. See our other book reviews for the same triage applied elsewhere.

The claim on trial

Gardner’s claim in the 2007 book is that five cognitive dispositions — disciplined, synthesizing, creating, respectful, ethical — are the capacities education and workplaces must cultivate, and that these five can be named, distinguished from each other, and taught.

Three separable assertions sit inside that. First, the five are distinct constructs rather than five labels on one underlying thing. Second, each is measurable, because you cannot claim to cultivate what you cannot detect. Third, cultivating them produces dispositions that travel — a synthesizing mind trained in school should synthesize at work. All three are testable. We checked all three.

One instrument in nineteen years

In August 2026 we searched Crossref, PubMed and ERIC for validation studies of the five minds framework. PubMed returns two records for «five minds» and Gardner, both false matches about unrelated clinical trials. Crossref returns Gardner’s own essays, reprints of them, and book reviews. ERIC returns two book reviews and one empirical paper.

That paper is the entire empirical base. Gelen developed a 66-item «Scale for Determining Mind Areas», administered it to 528 senior secondary students, and reported a Cronbach’s alpha of 0.81 with variance explained per subscale of 52.7, 46.5, 46.6, 53.0 and 46.6 for the disciplined, synthesizing, creating, respectful and ethical minds (Gelen, Educational Research and Reviews, 2015). The abstract, which is what ERIC indexes, reports no comparison of the five-factor model against competing structures, no discriminant validity showing the five subscales measure different things from each other, and no check against general ability or conscientiousness. There is no independent replication in any of the three indexes.

Untested is not the same as refuted. But this is not a young idea waiting its turn. «Five Minds for the Future» has been in print since 2007 and is widely assigned in education and management programmes. Nineteen years of exposure produced one scale, one sample, one country. For a framework sold as the thing institutions should organise themselves around, that is a checked negative result, and it is the one that matters most.

Gardner does not hide from this. Writing later about the synthesizing mind, he notes that Benjamin Bloom’s taxonomy originally listed synthesis as education’s highest goal and that it was later dropped because «it was considered too difficult to assess», and he states directly that «synthesizing capacities are neither systematically trained nor regularly and reliably assessed» (howardgardner.com). He then proposes adapting the Remote Associates Test and offers seven sample items. No reliability coefficient, no sample, no validation data accompanies the proposal. That is a scholar being candid about a gap. It also confirms the gap is real.

The foundation: multiple intelligences

The five minds inherit their credibility from multiple intelligences. So the honest question is whether MI theory itself ever passed a test.

It has been tested. Visser, Ashton and Vernon built two tests for each of Gardner’s eight intelligences — 16 tests — and gave them to 200 adults. Factor analysis produced a large general factor with substantial loadings for the purely cognitive intelligences: linguistic, logical-mathematical, spatial, naturalistic and interpersonal. Only bodily-kinesthetic and musical loaded weakly, and those are the ones involving sensory and motor demands rather than cognition. The authors concluded the results fit a hierarchical model, not eight independent domains (Visser, Ashton & Vernon, Intelligence, 2006).

Gardner replied that g is «simply a statistical entity» and that the tests were contaminated by verbal and logical demands, so they were not «intelligence-fair» (Gardner, Intelligence, 2006). The authors answered that g predicts job performance and socioeconomic outcomes and has biological correlates, and asked Gardner for «falsifiable, testable, MI-based hypotheses» that would predict something a hierarchical model does not (Visser, Ashton & Vernon, Intelligence, 2006). Twenty years later that request is still open.

The pattern replicates. A confirmatory factor analysis of 294 children aged 5 to 7 given both a psychometric battery and an MI-style battery found a general factor within the MI tasks themselves — the intelligences were not independent of one another (Almeida et al., Learning and Individual Differences, 2010). Waterhouse, reviewing the field twice, found no researcher had directly looked for a brain basis for the separate intelligences, and that MI teaching studies generally lacked adequate samples, control groups or valid outcome measures (Waterhouse, Educational Psychologist, 2006; Waterhouse, Frontiers in Psychology, 2023). Meanwhile the general factor Gardner dismisses keeps turning up: a single factor emerged in 73.2% of 97 samples from 31 non-Western nations totalling 52,340 people, and in 88.5% of the rest at second order (Warne & Burningham, Psychological Bulletin, 2019).

Learning styles, and what Gardner cannot control

Gardner has been explicit and public that multiple intelligences are not learning styles. Intelligences, in his account, are relatively autonomous faculties for processing different content; styles are claims about individual preferences for how material is presented (Harvard Graduate School of Education, 2013). He is right that these are different claims, and he deserves credit for saying so loudly while the conflation was making his ideas more popular.

The distinction does not survive contact with practice. The meshing hypothesis — teach visual learners visually and they learn more — was reviewed by Pashler, McDaniel, Rohrer and Bjork, who required the one design that can actually test it: a crossed interaction between style and instruction. They found essentially none. Of the studies using an appropriate design, the results contradicted the meshing hypothesis, and for the commercial learning-styles programmes being sold to schools they found no such studies at all (Pashler et al., Psychological Science in the Public Interest, 2008). Willingham, Hughes and Dobolyi reached the same place: scientific support is lacking, and educators’ time is better spent elsewhere (Willingham et al., Teaching of Psychology, 2015).

Belief did not move. A systematic review of 37 studies covering 15,405 educators in 18 countries found weighted endorsement of matching instruction to learning styles at 89.1%, with no sign of decline (Newton & Salvi, Frontiers in Education, 2020). Waterhouse reports 90% of US teacher trainees planning to use MI strategies and 94% of surveyed Quebec teachers saying they already do (Waterhouse, 2023). Gardner’s disclaimer is honest and it is losing.

Grading each mind against its own literature

Take the five minds not as a framework but as five separate empirical bets, and they come apart at different rates.

Disciplined. The strongest. Ericsson’s violinists at a Berlin music academy showed the top group accumulating over 10,000 hours of solitary practice by age 20 (Ericsson, Krampe & Tesch-Römer, Psychological Review, 1993). But the effect is domain-bound and smaller than the popular version: across 88 studies and 11,135 participants, deliberate practice explained 26% of performance variance in games, 21% in music, 18% in sports, 4% in education and under 1% in professions (Macnamara, Hambrick & Oswald, Psychological Science, 2014). Discipline in one domain buys you that domain.
Synthesizing. No instrument, by Gardner’s own account. The nearest measured relative is critical thinking instruction, which does work: a weighted mean effect of 0.30 across 341 effect sizes, largest with dialogue, authentic problems and mentoring (Abrami et al., Review of Educational Research, 2015). That is a real finding about a different, narrower construct.
Creating. Training works, unevenly. Across 70 studies and 4,210 participants, creativity training produced Δ = 0.75 on divergent thinking and 0.84 on problem solving, but only 0.35 on actual performance and 0.24 on attitudes and behaviour. The approaches that worked were cognitive and domain-specific — problem identification, idea generation, realistic exercises — not open-ended exploration (Scott, Leritz & Mumford, Creativity Research Journal, 2004). The gap between 0.75 on the test and 0.35 in the world is the whole issue.
Respectful. The mechanism is real, and it is not instruction. Across 515 studies and 713 independent samples covering 250,089 people in 38 nations, intergroup contact reduced prejudice at a mean r of about .21, rising to .34 in experiments (Pettigrew & Tropp, Journal of Personality and Social Psychology, 2006). Respect is produced by arranging contact, not by cultivating a mind.
Ethical. The weakest. A meta-analysis of 25 business ethics instructional programmes found «minimal impact» on ethical perceptions, behaviour or awareness (Waples et al., Journal of Business Ethics, 2009). The one mind the book most wants institutions to teach is the one with the clearest evidence that teaching it does not do much.

Transfer is the load-bearing assumption

Every prescription in the book assumes that a disposition trained in one context shows up in another. That assumption has been tested harder than anything else here, and it keeps failing. Two meta-analyses of chess and music instruction plus one of working memory training found effect sizes inversely related to design quality — the better the control group, the smaller the transfer — and concluded that far transfer of learning rarely occurs (Sala & Gobet, Current Directions in Psychological Science, 2017). Across 87 publications and 145 experimental comparisons, working memory training improved working memory and nothing else, with no relationship between how much the trained skill improved and how much transferred (Melby-Lervåg, Redick & Hulme, Perspectives on Psychological Science, 2016).

This is why Scott’s creativity numbers drop from 0.75 to 0.35, and why deliberate practice explains 26% in games and under 1% in professions. Broad dispositions are the hardest thing in cognitive science to produce deliberately. «Five Minds for the Future» proposes five of them at once.

Who should actually read it

Curriculum designers and school leaders who need shared vocabulary for a strategy document will get something from the full text. The writing is clear and the categories are memorable, which is a real service.

Skip it if you want a validated model, an assessment, or an implementation method. The book contains none of the three and does not claim to. Skip it also if you already accept that depth beats breadth; that is the one operational instruction inside it, and you have just read it.

One thing to try

Pick one domain and set a deliberate-practice target inside it for 12 weeks, then measure something domain-specific. Do not measure whether you have become more «synthesizing». The evidence says the gain will be local, so the metric should be local too: pieces performed from memory, defect rate in your own code, cases argued. If the number moves, the practice worked in that domain, which is the only claim the data supports.

Get the book

Find «Five Minds for the Future» on Amazon — as an Amazon Associate, The Boring Work earns from qualifying purchases (disclosure).

The boring bottom line

Gardner named five things worth wanting and never claimed to have measured them. That is a defensible position for an essayist and an indefensible one for a training programme. The five minds have one scale between them in nineteen years, from one sample of 528 students, and their author has stated in print that the most important of the five is not reliably assessed. The theory holding them up produced a general factor when it was finally tested, and its most influential misreading is believed by 89% of teachers. Keep the vocabulary if it helps you think. Do not build an institution on it.

Five Minds for the Future: One Scale in Nineteen Years