1 Introduction

What is the nature and structure of human intelligence? For more than a century, researchers have argued and discussed this question, often focusing on whether human intelligence is a general cognitive ability that influences all other cognitive abilities, or whether it is a set of different and independent cognitive abilities (Spearman, 1904, 1927). Most intelligence researchers today endorse the former perspective, according to which there is a g factor that emerges from the positive manifold—i.e., from the positive correlations of scores on different cognitive tasks—and that reflects a general cognitive ability in the human mind (Reeve & Charles, 2008; Rindermann et al., 2020).Footnote 1 This naturally became the dominant view in the 1990s, when Carroll (1993) factor-analyzed about 450 datasets from 19 different countries, almost all of which produced a hierarchical factor structure with g at the top. Moreover, only five years later, Arthur Jensen published The g Factor: The Science of Mental Ability (1998), convincingly arguing that the g factor receives a wealth of support from the empirical evidence, and that it is one of the most important psychological constructs ever discovered (cf. Bouchard, 2014; Panizzon et al., 2014; Ree & Carretta, 2022).Footnote 2

However, despite several decades of research supporting the idea that human intelligence is a general cognitive ability that contributes to one’s performance on all cognitive tasks, the opposing perspective claiming that there are different and independent intelligences remains remarkably popular, especially in educational circles (Hunt, 2011; Klein, 1997).Footnote 3 Indeed, theories of independent intelligences (such as Gardner’s multiple intelligences theory) are among the most commonly taught topics in introductory psychology textbooks (Griggs et al., 2004; Warne et al., 2018), and the vast majority of both teachers and members of the general public endorse the claim that there are many different types of intelligence (Schroth & Helfer, 2009; Warne & Burton, 2020).Footnote 4

Given that ideas about independent intelligences appear to be tremendously resistant to both theoretical problems and empirical refutations, the purpose of this article is twofold: first, by using Lakatos’s conceptual framework for understanding and evaluating scientific theories, it will argue that theories of independent intelligences constitute a degenerating, rather than progressive, research program; second, it will use the independent intelligences research program in order to illustrate three fundamental problems with Lakatos’s framework, having to do with his conceptions of the negative heuristic, the positive heuristic, and the relationship between scientific appraisal and advice. The upshot will thus be that theories of independent intelligences indeed teach us more about philosophy of science than about the nature of human intelligence.

Before doing this, however, three important points must be noted. First, although it is common to present the debate in question in a dichotomous manner (as a conflict between the independent intelligences perspective and the g perspective), this dichotomy may not fully capture all the relevant theories about the structure of human intelligence. For example, some researchers have recently argued that g is an emergent and epiphenomenal property (rather than the cause) of various domain-general executive functions (Kovacs & Conway, 2016) or of interacting cognitive processes during development (van der Maas et al., 2006). It is not entirely clear whether, or to what extent, these theories differ from more standard conceptions of g. For example, Arthur Jensen, who is one of the most famous proponents of the g perspective, at times offer seemingly conflicting reflections on the nature of g, some of which closely parallel the emergentist perspective. (For more on this, see the discussion by van der Maas et al., 2006, p. 843). Moreover, there are some data indicating that emergentist theories of g may be wrong (Gignac, 2014) and, furthermore, they may also have a hard time explaining the often replicated finding that cognitive training has very little transfer, and almost no influence on g (Protzko, 2017; Sala & Gobet, 2017).

Second, proponents of the g perspective generally acknowledge both the conceptual and practical importance of specialized cognitive abilities (cf. footnote 2). Indeed, a lot of recent research is focused on ability tilt, which refers to within-subject differences in specialized cognitive abilities, such as verbal and mathematical ability. Important findings include that ability tilt predicts occupation and college majors (Coyle, 2019), and that there are sex differences in cognitive abilities (Wai et al., 2018).

Third, those who endorse the independent intelligences perspective today usually agree that factor analyzing a battery of cognitive tests results in a g factor that explains about half of the variance in IQ test performance; they just offer a different interpretation of it (cf. footnote 3). More specifically, the proponents of the independent intelligences research program recognize that there is a distinction between the positive manifold and g, and they argue that only the former phenomenon—i.e., the positive correlations of scores on different cognitive tasks—really exists in any empirically meaningful sense of the term. This position is arguably inspired by some of Gould’s (1981) objections, such as that g should not be reified into a “thing”. Although Gould’s reification objection has been refuted (e.g., Bartholomew 2004; Bouchard, 2014),Footnote 5 the fact that there is a general agreement that the g factor (whatever its nature) explains approximately 50% of the variance in performance on well-constructed IQ tests shows that the aforementioned positions, presented in a somewhat dichotomous manner above, in some sense are less opposed to each other than they once were.

2 Theories of Independent Intelligences Constitute a Lakatosian Research Program

According to the philosopher of science Imre Lakatos, the proper unit of scientific evaluation is not individual theories or hypotheses, but rather a sequence of theories constituting a research program. Such a sequence of theories has the same hard core, which consists of a set of central theses that are made “irrefutable by methodological fiat”; they are irrefutable in the sense that the hard core does not in and of itself have any deducible empirical consequences. In order to derive testable predictions from the hard core, a theory has to rely on certain auxiliary hypotheses which allow for refutation of said theory, since the auxiliary hypotheses are (in contrast to the hard core) often considered falsified when evidence is presented that is inconsistent with the theory’s predictions. Indeed, when a scientist working within a certain research program is presented with evidence that is inconsistent with the predictions of the theory that the scientist endorses, the natural (and rational) reaction is to retain the hard core but modify the auxiliary hypotheses in accordance with the “partially articulated set of suggestions” constituting the program’s positive heuristic (Lakatos, 1978a, p. 50). That way, the scientist ends up endorsing a new theory, while nevertheless working within the same research program as before. This is how Lakatos aptly puts it:

The negative heuristic of the programme forbids us to direct the modus tollens at this “hard core”. Instead, we must use our ingenuity to articulate or even invent “auxiliary hypotheses”, which form a protective belt around this core, and we must redirect the modus tollens to these. It is the protective belt of auxiliary hypotheses which has to bear the brunt of tests and get adjusted and re-adjusted, or even completely replaced, to defend the thus-hardened core. (Lakatos, 1978a, p. 48).

Given what has been said thus far, it is reasonable to think of theories of independent intelligences as a Lakatosian research program. These theories are indeed committed to the same hard core, which can be formulated as follows:

Hard core

Human intelligence has a multidimensional structure, consisting of independent cognitive abilities. Moreover, human intelligence is not characterized by any general ability that is of greater practical importance, or that has greater predictive validity, than other, more specialized cognitive abilities.

Moreover, the theories differ insofar as they rely on different auxiliary hypotheses protecting the hard core. Consider some prominent examples. Edward Thorndike and Godfrey Thomson’s sampling theory argues that g does not reflect the existence of a singular cognitive ability that influences one’s performance on all cognitive tasks, but rather that it arises due to the overlap between different and uncorrelated neural “elements” (or cognitive abilities) (Thomson, 1951). The reason for the positive manifold is in other words not that human intelligence is characterized by a general cognitive ability, but instead that current IQ tests are not able measure any single element or cognitive ability without also sampling other, uncorrelated abilities. The sampling theory thus acknowledges that factor analyses of IQ tests produce g as a statistical construct, but argues that g nevertheless is illusory in the sense that it does not adequately represent the myriad of uncorrelated intelligences that really exist in the human mind.

Another interesting theory is Joy Paul Guilford’s (1967) structure-of-intellect (SOI) model, which states that there only exist specialized cognitive abilities; not only isn’t there a g factor according to the SOI model, but it does not even allow for the existence of broad, second-order abilities. More specifically, the SOI model argues that there are three cognitive ability facets (not factors)—namely, contents (visual, auditory, symbolic, semantic, behavioral), products (units, classes, relations, systems, transformations, implications), and operations (cognition, memory, divergent production, convergent production, and evaluation)—each of which can be represented as one dimension of a rectangular prism containing 150 (5 × 6 × 5) cells. Each cell represents a type of intelligence that is uncorrelated with any of the other cells.

In his Frames of Mind: The Theory of Multiple Intelligences (2011), originally published in 1983, Howard Gardner develops what arguably is the most influential theory of independent intelligences—the multiple intelligences theory (MI). Originally, Garner argued that there were seven “relatively independent” kinds of intelligence: Linguistic intelligence, logical-mathematical intelligence, musical intelligence, spatial intelligence, bodily-kinesthetic intelligence, interpersonal intelligence, and intrapersonal intelligence. Later, he has added other kinds of intelligence to the list, such as naturalistic intelligence and existential intelligence (Gardner, 1999). For present purposes, it does not really matter what differentiates these intelligences and makes them unique. Rather, what’s important is Gardner’s contention that g is an artifact of the way in which IQ tests are constructed and, furthermore, that his seven (or nine) intelligences are conceived as modular, independent, and grounded in different biological subsystems.

The last example I want to discuss is Robert J. Sternberg’s triarchic theory of intelligence. The triarchic theory argues that there are three different types of intelligence or components of information processing: analytical/componential, creative/experiential, and practical/contextual (Sternberg, 1985; Sternberg et al., 2000; Wagner, 1985). Whereas analytical (or componential) intelligence resembles g insofar as it is highly relevant to scholastic achievement and problem solving, the other types of intelligence involve different components of information processing and are typically unrelated to analytical ability. Notably, Sternberg argues that practical intelligence is not the same as g, and that it therefore is not unusual to find people that are “book smart” without being “street smart”, or vice versa.

Now although these theories rely on different auxiliary hypotheses that differentiate them and make sure that they entail different empirical predictions, the auxiliary hypotheses all function to protect the same hard core of theoretical commitments—which means that they ultimately belong to the same research program. Continuing, the next section will make use of Lakatos’s evaluative standard in order to determine the quality and productiveness of the independent intelligences research program.

3 Progress and Degeneration in Human Intelligence Research

According to Lakatos, the quality of a research program depends on a couple of properties which it may or may not have. A research program is considered progressive if it has the properties in question. However, if it does not have the properties, then it is considered degenerating.

Any progressive research program necessarily satisfies two important conditions: theoretical progressiveness and empirical progressiveness. A research program is theoretically progressive just in case “each new theory has some excess empirical content over its predecessor, that is, [. .] it predicts some novel hitherto unexpected fact” (Lakatos, 1978a, p. 33). To say that a sequence of theories constituting a research program is theoretically progressive means that the later theories entail novel predictions that are neither derivable from nor contained in earlier theories in the same program.Footnote 6 A research program is empirically progressive just in case “some of this excess empirical content is also corroborated, that is, [. .] each new theory leads to the discovery of some new fact” (Lakatos, 1978a, p. 34). To say that a sequence of theories constituting a research program is empirically progressive means that some of the novel predictions entailed by the program’s later theories must be corroborated by empirical evidence. A research program is progressive if, and only if, it satisfies both of these conditions—otherwise it is degenerating (Lakatos, 1978a, p. 34).

Human intelligence research is characterized by a number of different research programs, and Lakatos’s conceptual framework has been used several times before in order to evaluate some of them. In a paper written with encouragement from Lakatos (see Lakatos & Feyerabend 1999, pp. 348–349), Urbach (1974) relied on Lakatos’s so-called “methodology” (i.e., his conceptual framework) in order to judge the relative quality of the different positions that were being debated in the aftermath of the publication of Jensen’s (1969) famous article on IQ and scholastic achievement. Following Urbach, Rushton & Jensen (2005) used the technical notions of progress and degeneration (as defined above) in order to compare the merits and demerits of hereditarianism and environmentalism. Moreover, another example is provided by te Nijenhuis et al., (2017), who have argued that the Lakatosian anomalies engendered by testing of certain permutations of Spearman’s hypothesis using the method of correlated vectors can be suggestive of new and progressive ways of developing relevant theory.Footnote 7 And another recent example is Egeland’s (2022) evaluation of what he calls the “postindustrial selection research program”, with the co-occurrence model as its latest and most sophisticated constituent theory, as scientifically progressive. Continuing, for the remainder of this section it will be argued that the independent intelligences research program is degenerating, since it has not lead “to the discovery of hitherto unknown novel facts” (Lakatos, 1978b).

Consider first Thorndike and Thomson’s sampling theory, which argues that g does not reflect the existence of a singular cognitive ability that influences one’s performance on all cognitive tasks, but rather that it arises due to the overlap between different and uncorrelated neural “elements” (or cognitive abilities). The sampling theory’s greatest weakness is that it is not theoretically progressive; it does not entail any novel prediction that is not contained in, or derivable from, earlier theories belonging to the same research program. More specifically, the sampling theory successfully demonstrates that the positive manifold does not by itself entail the existence of g as a general cognitive ability, since the existence of all-positive correlations between different cognitive tests is consistent with the claim that there are multiple independent cognitive abilities that invariably are sampled together in the tests, but the theory itself does not provide any new empirical content. As one psychometrician has put it:

The sampling theory hardly qualifies as a true theory, for it does not make any assertion to which evidence is relevant. Perhaps the large number of adherents to this view is due to the fact that no one has offered evidence against it. But until the view is defined more sharply, one cannot even conceive of the possibility of contrary evidence, nor, for that matter, confirmatory evidence. A statement about the human mind which can be neither supported nor refuted by any facts, known or conceivable, is certainly useless (Loevinger, 1951, pp. 594–595).

Furthermore, the sampling theory also does not satisfy the condition of empirical progressiveness. Not just does it not provide novel predictions, at least some of which are corroborated by the empirical evidence, but it also does not cohere with all the evidence that we already have. For example, given that the sampling theory were true, one would not expect that cognitive tests that appear to sample quite different neural “elements”—such as tests with radically different culture loadings—to correlate as highly with each other as they in fact do. Several examples to this effect are provided by Jensen (1998, pp. 117–122). (See also Eysenck 1987, for more on the empirical problems with the sampling theory.)

Next, we have Guilford’s SOI model. The SOI model claims that there are 150 specialized cognitive abilities, all of which should be measurable and uncorrelated. Although the model has some theoretical oddities (if not outright problems)—such as determining the number of cognitive abilities on the basis of the ways in which certain postulated cognitive ability facets are thought to combine—it is without a doubt theoretically progressive. As Lakatos would put it, the model has “excess empirical content over its predecessor”. However, it is equally clear that the SOI model is not empirically progressive. When Guildford’s U.S. Air Force personnel data was reanalyzed and corrections were made for attenuation and range restriction, the correlations between his tests had a mean value of + 0.45, and all the correlations were positive (Alliger, 1988; cf. Carroll, 1993, pp. 57–60).

But what about Gardner’s famous MI theory? Gardner’s theory that there are seven (or nine) “relatively independent” intelligences is also theoretically progressive, since it predicts that the aforementioned intelligences should be measurable, uncorrelated, and have separate biological foundations. Nevertheless, it should be mentioned that it also has several theoretical problems: It is too vague, since it does not specify how independent the intelligences are. Are they only weakly correlated or not correlated at all, or are they perhaps even negatively correlated? Gardner does not say. Moreover, he does not specify how his intelligences can work together when performing complex tasks that require the individual to use of more than one of them at the same time. The theory has also been alleged to rely on circular reasoning, and it does not offer adequate concrete suggestions for how it can be tested. Concerning this latter point, testing the theory in accordance with Gardner’s suggestions would either be too time and resource consuming for practical purposes, or result in relatively unreliable data with little or no practical value. (For helpful and more detailed expostulations of the theoretical problems facing Gardner’s theory, see: Brody 1992, p. 40; Hunt 2011; Jensen, 1998, pp. 128–132; Warne, 2020, pp. 57–59).

Moreover, the MI theory has grave empirical problems, the biggest of which is that systematic and well-designed studies attempting to determine whether there exist multiple intelligences invariably discover a g factor that explains the positive correlations between the tests (e.g., Castejon et al., 2010; Pyryt, 2000; Visser et al., 2006).Footnote 8 Other serious problems include the fact that Gardner (2011) appears to have a rather strong confirmation bias, choosing to favor and focus on evidence that appears to confirm his theory, while ignoring or simply dismissing contradicting evidence (e.g., Bouchard Jr, 1984; Messick, 1992; Scarr, 1985). For example, he occasionally outright rejects not just results that conflict with the MI theory, but even the gathering of evidence that might threaten its perceived veracity as “incendiary” or “pseudo-science” (e.g., Gardner 2006, p. 239).Footnote 9 A related issue concerns Gardner’s reliance on anecdotal data on outliers and eminent individuals, rather than representative samples of subjects taking tests that are supposed to measure his different intelligences. By its very nature, anecdotal evidence has extremely limited generalizability and cannot be used to provide empirical support to theories about the psychological or behavioral structure of all (or most) people’s abilities. Moreover, Gardner has reportedly estimated that one would need an IQ of at least 120 to be included in his list of eminent individuals (Jensen, 1998, p. 128). But if high IQ is required for intellectual success or eminence, any plausible explanation of why (e.g.) Picasso or T. S. Elliot achieved such a high degree of eminence must reference their high level of g, since most of the predictive and explanatory power of IQ comes from g (Ackerman et al., 2013; Ganzach & Patel, 2018; Kell et al., 2013; Ree & Carretta, 2022; Zaboski et al., 2018). All of this means that the MI theory is not empirically progressive.

Lastly, we have Sternberg’s triarchic theory of intelligence, which argues that there are practical and creative intelligences, that they are different from g, and that they are just as important for achieving important real-life outcomes. Moreover, it should be noted that Sternberg (1982) does not in any way deny either the existence or importance of g; he just believes that there are other, equally (if not more) important intelligences. The triarchic theory is theoretically progressive since it offers novel predictions. For example, it predicts that practical and creative intelligence do not correlate with g, and that these intelligences should be just as important as g for achieving one’s goals.

However, one problem with Sternberg’s theory is that he appears to have inconsistent attitudes toward the claim that practical intelligence is context-dependent. Sometimes he (1985) argues that the ability to learn tacit, non-academic knowledge is context-invariant, meaning that the non-academic, non-g skills needed to become successful in one environment will enable one to achieve success in other environments. However, at other times Sternberg seems to fall back on the commonsense position that high practical ability in one context does not guarantee high practical ability in another. For example, a test of practical intelligence for Kenyan children involved a question about plausible causes of a regular stomachache, and the correct answers included that the person with the stomachache “is affected by the evil eye (sihoho)” (Sternberg et al., 2001, p. 408). Now it is not just doubtful that believing in superstitions that might be dangerous to one’s health is helpful for achieving one’s goals in rural Kenyan villages, it is pretty much certain that it is not going to help you in a modern Western environment.Footnote 10 Sternberg’s inconsistent attitudes toward the context-dependent nature of high practical ability manifests what one commentator has called a “heads-I-win-tails-you-lose” strategy (Gottfredson, 2003).

Moreover, just as with Guilford and Gardner’s theories, the triarchic theory’s biggest problem is that it is not empirically progressive. Indeed, when Sternberg’s data are factor-analyzed, his three intelligences are not just found be correlated and, hence, produce a g factor (Brody, 2003; Chooi et al., 2014; Hunt, 2008; Jukes et al., 2006; Messick, 1992; Stemler, Chamvu, et al., 2009; Stemler, Grigorenko, et al., 2009), but no one (at least to the present author’s knowledge) has been able to demonstrate that this general factor is not more important than either practical or creative ability alone. After all, the proposition that it is g that really matters for achieving positive outcomes in different environments coheres more with the available evidence (Ackerman et al., 2013; Ganzach & Patel, 2018; Kell et al., 2013; Ree & Carretta, 2022; Zaboski et al., 2018).

In sum, none of the most prominent theories of different and independent types of intelligence are both theoretically and empirically progressive. What is remarkable about the independent intelligences research program is not just that it isn’t the case that all of the theoretical developments within the program—the so-called problem shifts (Lakatos, 1978a, p. 37)—are theoretically and empirically progressive, which must be so if the program is to be considered progressive in general, but that none of the theoretical developments within the program appear to be theoretically and empirically progressive. The independent intelligences research program must therefore be considered degenerating.

4 From Psychological Muck to Philosophical Gold: Illustrating some Problems with Lakatos’s Conceptual Framework

When judging in accordance with the Lakatosian standard, whereby the quality of a sequence of scientific theories sharing the same hard core is a function of its theoretical and empirical progressiveness, the independent intelligences research program is clearly degenerating. However, despite its flaws as a scientific view of the nature of human intelligence, the independent intelligences research program is nevertheless philosophically useful since it illustrates some of the problems with Lakatos’s conceptual framework. Although Lakatos’s methodology has a lot going for it and certainly constitutes an improvement over earlier falsificationist philosophies, Musgrave (1976) has plausibly argued that it faces three particularly pressing problems (cf. Laudan, 1977, pp. 77–78). Using Lakatos’s own examples—namely, Newton, Prout and Bohr’s research programs—he shows that Lakatos’s conceptions of the negative heuristic, the positive heuristic, and the relationship between scientific appraisal and advice all are vulnerable to objections. Let’s consider the negative heuristic first.

Lakatos tells us that the hard core of a research program is rationally made “irrefutable by methodological fiat”, and that the negative heuristic commands us to direct the modus tollens at the auxiliary hypotheses instead. However, as Musgrave (1976, pp. 459–467) shows by focusing on Lakatos’s own examples, it is not true that scientists always consider the hard core immune to empirical refutation, nor is it the case that they always should do so. Moreover, this is a point that the independent intelligences research program can help us illustrate. Although I don’t know of anyone who has attempted to modify or alter its hard core, it would clearly be irrational to retain it regardless of the number of anomalies that accumulate with observation or experimentation.

When Gardner (2011, p. xxxix) outright dismisses evidence that threatens the hard core, or when he states that

even if at the end of the day, the bad guys [i.e., those who work within the alternative, g-centered research program] turn out to be more correct scientifically than I am, life is short. And we have to make choices about how we spend our time, and that’s where I think the multiple intelligences way of thinking about things will continue to be useful, even if the scientific evidence isn’t supportive (Gardner, 2009, 0:45:11 − 0:45:32).

he has rightly been criticized for having an overly dogmatic attitude toward both the hard core and the auxiliary hypotheses of the MI theory (Cofnas, 2016, pp. 483–486; Warne, 2020, pp. 52–61). And the reason is that retaining the hard core at all costs, or making it “irrefutable by methodological fiat”, is not always rational when anomalies abound and there is no clear way forward for saving the research program.

This brings us to Lakatos’s second problem, which is that the positive heuristic cannot always anticipate empirical refutations and provide guidelines for how the auxiliary hypotheses (i.e., the protective belt) are to be modified in response to them (Musgrave, 1976, pp. 467–473). According to Lakatos, the positive heuristic “consists of a partially articulated set of suggestions or hints on how to change, develop the ‘refutable variants’ of the research programme, how to modify, sophisticate the ‘refutable’ protective belt”, and it “saves the scientist from becoming confused by the ocean of anomalies [… since] their existence is fully expected” (Lakatos, 1978a, pp. 50–51). However, when studying actual research programs, we find that the positive heuristic often is developed over time along with other parts of the program, and that it therefore cannot always provide guidelines for handling anomalies or empirical refutations before they present themselves. Indeed, the independent intelligences research program is a case in point, as its positive heuristic has not included good enough “hints” to develop new auxiliary hypotheses that enable the research program to “digest” the g factor that invariably is produced when cognitive tests are submitted to factor analysis. Not just that, as we have seen above, scientists working within the independent intelligences research program do not always “expect” that their independent intelligences will correlate, even though they pretty much always do.

Lastly, we have the third problem facing Lakatos’s methodology. The problem is that it is not clear what the relationship between scientific appraisal and advice is. We know that a research program is of high quality when it is theoretically and empirically progressive, and that it is considered degenerating when it does not have these properties. But what should a scientist do with this information? Should, for example, the scientist working within a degenerating research program abandon it in favor of a progressive one? As Musgrave (1976, pp. 473–483) explains, Lakatos argues that a research program will only be abandoned when there exists an alternative program for the scientist to work on—regardless of whether, or to what extent, the program is confronted with anomalies. Moreover, Lakatos sometimes claims that his methodology does not issue any advice to scientists concerning which research program they should work within, other than that they should be honest about the relative quality of the various programs that exist (Lakatos, 1971b, pp. 174, 178). However, at other times Lakatos seems to suggest that there isn’t a clear distinction between appraisal and advice, and that resources, time, and energy should be focused on progressive research programs (unless there only exist degenerating ones) (Lakatos, 1968, p. 343; 1971a, p. 100; 1971b, p. 174; 1978a, pp. 89–90). So, in other words, the problem is that it is not clear what scientists should do when presented with relevant information about the relative merits and demerits of the research programs on the market.

Moreover, this is an issue that intelligence researchers cannot remain agnostic about, and it is especially relevant when it comes to the independent intelligences research program. Given that said program is degenerating—as it indeed has been argued above—should its proponents continue their work on it, or should they perhaps abandon it in favor of the g-centered research program? There does not appear to be a forthcoming answer. On the one hand, it does not seem reasonable that the scientific community should devote just as much energy to the independent intelligences research program as its more meritorious rival. But, on the other hand, one would not want everyone working within the independent intelligences research program to abandon it—after all, there is never any a priori guarantee that fortunes won’t change, or that what has worked in the past will continue to do so in the future. A reasonable compromise might be that the scientific community should devote most, but not all, of its resources and energy to progressive research programs (cf. Musgrave, 1976).

5 Conclusion

This paper has argued that theories of different and independent types of intelligence constitute a degenerating Lakatosian research program. However, despite its scientific weaknesses, the program has philosophical value insofar as it provides an illustrative example of the biggest problems facing Lakatos’s methodology. Although no final conclusion is reached regarding what the way forward should be, it is recommended that intelligence researchers, and the wider society too, should give priority to progressive, rather than degenerating, research programs—but with conscious awareness of the fact that the quality of any such program may change with time.