When people ask what Fundamental Uncertainty is about, I usually say it’s a book about epistemology. If they want to know more, I say it’s a book arguing that truth is grounded not in observation or more truth, but in usefulness, and because what’s useful depends on what we care about, truth is grounded in care. And this is a reasonable way to present the book because this is, in fact, its core claim.
But there’s another way to frame the book, and that’s to say it explains why epistemic uncertainty exists.
My impression is that many and maybe most people treat uncertainty as a contingent feature of our attempts to know the truth. This is certainly a standard Bayesian perspective on uncertainty, where if we weren’t flawed creatures with numerous limitations, then perhaps we could know what’s true with complete certainty. And sometimes, in limited domains, it seems like we can.
But the argument I make in the book is that uncertainty is a fundamental feature of knowing truth, or, to state it more precisely, it’s a feature of holding testable beliefs that aim to accurately predict and explain our observations. Uncertainty arises because attempts to know testable truths encounter the Problem of the Criterion, the only resolution to the Problem of the Criterion is to make unjustifiable assumptions, and the making of these assumptions creates uncertainty.
For what it’s worth, Karl Popper makes a similar argument, though he builds his argument around the trilemma of justification (a choice between dogmatism, infinite regress, and circularity) rather than the Problem of the Criterion. Where Popper and I differ more is our analysis of how unjustifiable assumptions are chosen. Popper focuses on the practical methods by which scientists pick assumptions and doesn’t overly concern himself with the ultimate grounding that motivates their choices. I look more deeply to ask “why do we care about what we care about?”, find that our care is rooted in the evolutionarily-instilled goals of survival and reproduction, and use this as the basis to explain why we make the assumptions we do when aiming to know the truth.
Popper and I have greater differences over what this means for the Bayesian project. Popper seems to reject Bayesianism out-of-hand for relying on induction. I make no such move, and think Bayesianism is great on practical grounds. My only substantive disagreement with standard Bayesianism is with its account of the cause of uncertainty.
We can sort Bayesian sources of uncertainty into two sources: marginal uncertainty from limited observational accuracy and finiteness of computation, and ultimate uncertainty from the problem of choosing priors. Bayesian methods offer no way to choose priors within the theory, so the choice is “arbitrary” because it relies on making a choice not constrained by Bayes’ Theorem. But this framing of the choice of priors as arbitrary is not quite right, or so I argue, because we choose priors that are useful. There may be reasonable disagreements about what is useful, and those disagreements persist because different people care about different things, but this is a far cry from the arbitrariness Bayesianism would ground uncertainty in, since care is not arbitrary or unconstrained-by-anything, but contingent on our being as we are.
Reminder: I’m offering $2000 in prizes for the best essays about the themes covered in Fundamental Uncertainty. Deadline to submit your essay is August 1st.
When people ask what Fundamental Uncertainty is about, I usually say it’s a book about epistemology. If they want to know more, I say it’s a book arguing that truth is grounded not in observation or more truth, but in usefulness, and because what’s useful depends on what we care about, truth is grounded in care. And this is a reasonable way to present the book because this is, in fact, its core claim.
But there’s another way to frame the book, and that’s to say it explains why epistemic uncertainty exists.
My impression is that many and maybe most people treat uncertainty as a contingent feature of our attempts to know the truth. This is certainly a standard Bayesian perspective on uncertainty, where if we weren’t flawed creatures with numerous limitations, then perhaps we could know what’s true with complete certainty. And sometimes, in limited domains, it seems like we can.
But the argument I make in the book is that uncertainty is a fundamental feature of knowing truth, or, to state it more precisely, it’s a feature of holding testable beliefs that aim to accurately predict and explain our observations. Uncertainty arises because attempts to know testable truths encounter the Problem of the Criterion, the only resolution to the Problem of the Criterion is to make unjustifiable assumptions, and the making of these assumptions creates uncertainty.
For what it’s worth, Karl Popper makes a similar argument, though he builds his argument around the trilemma of justification (a choice between dogmatism, infinite regress, and circularity) rather than the Problem of the Criterion. Where Popper and I differ more is our analysis of how unjustifiable assumptions are chosen. Popper focuses on the practical methods by which scientists pick assumptions and doesn’t overly concern himself with the ultimate grounding that motivates their choices. I look more deeply to ask “why do we care about what we care about?”, find that our care is rooted in the evolutionarily-instilled goals of survival and reproduction, and use this as the basis to explain why we make the assumptions we do when aiming to know the truth.
Popper and I have greater differences over what this means for the Bayesian project. Popper seems to reject Bayesianism out-of-hand for relying on induction. I make no such move, and think Bayesianism is great on practical grounds. My only substantive disagreement with standard Bayesianism is with its account of the cause of uncertainty.
We can sort Bayesian sources of uncertainty into two sources: marginal uncertainty from limited observational accuracy and finiteness of computation, and ultimate uncertainty from the problem of choosing priors. Bayesian methods offer no way to choose priors within the theory, so the choice is “arbitrary” because it relies on making a choice not constrained by Bayes’ Theorem. But this framing of the choice of priors as arbitrary is not quite right, or so I argue, because we choose priors that are useful. There may be reasonable disagreements about what is useful, and those disagreements persist because different people care about different things, but this is a far cry from the arbitrariness Bayesianism would ground uncertainty in, since care is not arbitrary or unconstrained-by-anything, but contingent on our being as we are.
Reminder: I’m offering $2000 in prizes for the best essays about the themes covered in Fundamental Uncertainty. Deadline to submit your essay is August 1st.