How has [or will] the prevalence of “big data” – the exploding plethora of information and computing power to classify, categorize and correlate it, combined with Artificial General Intelligence, or strong” AI impact[ed] two of what are arguably the 20th C's most important contributions to the philosophy of science: Quine’s underdetermination of theory by data, and the holism exemplified by “our statements about the external world face the tribunal of the external world not individually but as a corporate body.”

For instance, would the philosophical conundrums revealed/identified by these questions be considered by, or “strike,” AGI as quaint/nonsensical, or would they likely be [re]solved [put to rest, implicitly or explicity] by entities whose computational algorithms [thought processes(?)] we may, or certainly will, never come to understand?

ADDENDUM EDIT: It seems to me obvious how the above question differs from the referenced 2015 post: "How has [or will] the prevalence of “big data” – the exploding plethora of information and computing power to correlate it – impact[ed] (i) the scientific method’s theory/hypothesis formation, (ii) the underdetermination thesis, and (iii) the realist/relativist/ constructivist debate? The question might profitably be bifurcated in terms of the social sciences and the physical sciences."

First: The 2015 query made no reference to AGI. It did not, for instance, implicitly [that is. one way in which the query can be interpreted] ask for a human being's hypothesis about how a machine-that-learns [and makes decisions based upon strategies and criteria that we can hardly describe or understand] might approach, deal with, answer, arguably provocative and perplexing questions in the philosophy of science.

Second: No component of the earlier query made reference to semantic holism, which may be particularly salient [or not] to the processes involved in [one or another type] of AGI.

Third: The present query makes reference neither to the scientific method, nor to issues surrounding theory/hypothesis formation or to the [anti]realism/relativism/constructivism debate.

Fourth: We live in a different world than we did three years ago. Things are moving quickly, particularly in the realm of AI. The technological advances that have been made in that domain over the past three years are substantial; which implies that the scope information from which a relevant answer to this and the earlier question might be derived has expanded. (In this context, consider also that the 2015 post, despite having been answered numerous time, was shut down two weeks after it was posted.)

  • 2
    What impact would some future X have on the philosophical issue Y is not an SE answerable question. I am also not sure why underdetermination is a conundrum, it is pretty well understood, or why big data makes any difference, regardless of what one's position on underdetermination is.
    – Conifold
    Dec 1, 2018 at 6:40
  • 1
    Holist Underdetermination is the claim (due to Duhem) that theories or hypotheses can only be subjected to empirical testing in groups or collections, never in isolation." Dec 1, 2018 at 12:14
  • 3
    Thus, the issue is not about "quantity of data" : the hypothesis can never be "ultimately" tested not due to lack of data, vut because they are part of a complex network of theories and hypotheses taht is impossible to disentangle. Dec 1, 2018 at 12:16
  • @Mauro ALLEGRANZA "Impossible to disentangle" for entities with limitations such as ours not only as to data storage/availability capacity (compilation and retention), but capacity to determine the relevance/salience of known data available for use in particular domains of inquiry -- what data sets are salient/relevant to a particular issue. That is, you may be speaking of physical, not logical impossibility.
    – gonzo
    Dec 1, 2018 at 19:55

1 Answer 1


"Mathematicians wish to treat matters of perception mathematically, and make themselves ridiculous... the mind.. does it tacitly, naturally, and without technical rules." - Pascal, Pensées

Big data and narrow AI have minimal impact on the underdetermination thesis. In effect, the computational practices are a natural extension of essentially well-understood statistical methods applied to automated systems of Universal Turing Machines that function without developing their own semantics. As such both sub-disciplines can be likened to sophisticated calculators. They make mistakes and are forgiven, for they know not what they do. As such, they do not have an impact on genuine, autonomous scientific inference. They are extensions of our own minds, and are tools proper. If they have any applicability to holistic underdetermination, it might be in the form of expert systems which invoke logic to evaluate coherence among models in model theory, a field in mathematics which purports to give axiomatic reasoning it's full berth as an analogy to informal logic. In this way, let's say within the context of the Toulmin method, it might help to illustrate the limitations of non-semantically aware systems using deductive and inductive methods to approximate abduction by substituting human-encoded expertise in the form of knowledge bases and knowledge rules. But even in using multiple logics above and beyond mere FOPC, ZFC, and fuzzy logics, these processes on the whole are incapable of creating concepts, analyzing propositions for conceptual content, and extracting meaning from the systems to make pragmatic determinations about content corresponding to physical properties and among propositions. Thus, fundamentally even if they are designed to calculate correspondent, coherent, and pragmatic values of truth (no small task), they will rigidly do so within the parameters of the algorithms and databases they are instantiated by.

AGI, however, is a different beast entirely, because by definition, AGI must encode the non-deterministic capacities of the human being. As such, AGI, must meet several challenges, including the capacity through embodied intelligence to create processes of inference and categorization through an interplay with the environment approximately like the human body. This fundamental functionality gives rise to neurocomputational elements that allow physical causality to be converted to mental causality and back as supervenience expects. In addition, these neurocomputational states must embody the capacity for natural language, and together among language, concept, and inference must approximate the unconscious and intuitional aspects of abduction, primarily in the creation, assessment, and manipulation of meaning.

Thus AGI will unite connectionist and symbolic camps, kill Cartesian duality thoroughly and finally outside of the philosophy of the mind particularly (where it has no serious adherents) among philosophers and mathematicians generally, and eliminate the realist-anti-realist debate showing that the same category mistake that applies to mind-body dichotomy, software-hardware dichotomy, etc. is at work with realist and anti-realism sentiment. Ultimately it will bring to light the notion that just as descent with modification and selection are at play governing genes, so too do those genes govern metaphysics.

Underdetermination of scientific theory, in my understanding, is just a more sophisticated form of Hume's skepticism of inductive reasoning. Any form of underdetermination which attacks the methods of induction by highlighting that it is not deduction lost any value as a pragmatic razor with the death of logical positivism and the attempts to mechanize science, at least in regards to its interpretation of the rational aspect of theory building, reduction, and comparison. The broader form which attempts to show some form of epistemic weakness in regards to the relationship between propositions of theory and evidence does highlight the role society plays in refereeing scientific theory, but does so at the wrong level to claim the supremacy of some pure form of relativism in which truth is so malleable it is divorced from the physical; rather it only complements AGI insofar as it complements cognitive science in showing that the flexibility in mapping metaphysical propositions within a framework of being foundational to scientific methods is a near-universal property of human, biological intelligence. As such, it will emphasize the role of neurocomputational units that comprise the unconscious properties that govern the conscious rational structures that participate in science. It is the neurons which ultimately govern the formation of concepts and provide the basic structures of inference which ultimately ground the scientific endeavor in the biological fountainhead of meaning.

(I can provide references for these statements, but it's late.)

  • Thank you for this answer. Please describe the category mistake here: "[AGI] will eliminate the realist-anti-realist debate showing that the same category mistake that applies to mind-body dichotomy, software-hardware dichotomy, etc. is at work with realist and anti-realism."
    – gonzo
    Oct 1, 2019 at 15:54
  • Also, you refer to "a framework of being foundational to scientific methods." How would you describe that framework. Finally, How will AGI simulate embodiment, enabling it to "create processes of inference and categorization through an interplay with the environment." Or are you proposing, for instance, that only a fleet of cloud computing "robots" moving through actual environments (something like a fleet of driverless cars, each learning from both the environment it frequents and the environments frequented by its cohorts) will be found to actually be capable of instantiating AGI?
    – gonzo
    Oct 1, 2019 at 16:33
  • I'm at work, but I'll tackle the first one. The realism-anti-realism dichotomy presupposes the dichotomy of correspondence-coherence which in reality are two categories which presuppose a mind-body duality. All arguments of in-here-out-there thinking are rational, logical structures which use the conceptual metaphor of "Containment" (which are neurocomputational mappings) and presume that human beings are like containers (obviously from our conceptual predisposition to extensionality from our sensori-motor capacities). Those two theories of truth are aspects of a single process, hence mistake.
    – J D
    Oct 1, 2019 at 16:57
  • Truth is better understood in terms of a multi-layered neurocomputational architecture in which correspondence, coherence, and pragmatism are different aspects of the unconscious foundation of conscious rational processes. It's the only way to account for the genuine lack of foundation among logics such as arithmetic, set theory, classical logic, etc. This is laid bare obviously by the Agrippan trilemma, but obfuscated by the mind-body duality and the notion of objectivity.
    – J D
    Oct 1, 2019 at 17:03
  • @gonzo It's a linguistic illusion that hides the obvious monism that all thought is embodied (primarily, but not exclusively in neurons). Right? Take Copenhagen vs. Bohmian interpretations to which underdetermination applies. Is it the "nature of reality" that is at stake, or just the "models of reality"? These are map-disputes more than they are territory-disputes. They should be had, but they are more like Leibnitz-Newtonian disputes of orthography than debates over the epsilon-delta definition. Anyway, that's what it looks like as a computer science guy.
    – J D
    Oct 1, 2019 at 17:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.