Following Casey Luskin’s introductions (here and here), my co-authors and I are discussing our recent paper in the journal BIO-Complexity detailing how intelligent design can serve as a heuristic to guide scientific inquiry. In this article, I will describe how a theory may be related to evidence by what the theory leads us to expect concerning the data we observe.
Predictions, Retrodictions, and Expectations
There are two principal ways by which a theory may be related to evidence. A prediction is when a theory anticipates a data point before the data point is discovered, whereas a retrodiction is when a theory explains a data point that was already known when the theory was developed. A broader term covering both cases is expectations, which refers to what a theory leads us to expect given its causal and explanatory power.
Scientific models succeed or fail primarily by their ability to explain the relevant data, irrespective of whether the theory was developed before or after the data were known. We therefore prefer the word “expectation” since this best captures the relevant epistemic relationship irrespective of the chronology of theory development and observational discovery.
For example, consider the example of a gorge that has been known about for centuries or millennia. A causal model may be developed to account for the gorge on the basis of the observation that flowing water possesses the causal power to erode rock and carve channels. We may then draw an inference that the river-erosion model best explains the origins of the gorge, even though the existence of the gorge was known before the development of the model.
In a similar way, intelligent design argues on the basis of empirically grounded knowledge of what intelligent agents can produce. In particular, in our ordinary experience, information-rich, language-based codes are habitually associated with intelligent causes. This knowledge then generates expectations regarding what we might find in biological systems on the supposition that an intelligent agent was causally responsible. This theory can be explanatory even when the relevant biological feature was discovered before the development of modern intelligent design theory.
This chronology issue is not unique to intelligent design. Indeed, when On the Origin of Species was published in 1859, Darwin was largely using his theory to explain observations and biological phenomena that were already known. Thus, accounting for previously known evidence is a normal feature of scientific theory-building.
Expectations and Bayesian Reasoning
As a Bayesian, I conceive of evidence in terms of a likelihood ratio of the probability of the data given the truth of a hypothesis against the probability of that same data given the falsity of the hypothesis. A Bayesian framework therefore asks how expected a data point is under competing hypotheses.
An important point that is often misunderstood is that a data point need not be certain or inevitable, or even more probable than not, under a hypothesis in order to confer evidential support. What matters is the ratio of the two probabilities. For example, even if the hypothesis renders the probability of the data as low as 1 percent, if the probability of those same data is much lower than 1 percent, the overall likelihood ratio is top-heavy and therefore confers evidential support upon the hypothesis under investigation. Applied to intelligent design, the inference has the following structure: Various features of living organisms and of the universe are not especially surprising given the hypothesis of design, whereas they become wildly surprising supposing the falsehood of intelligent design. Therefore, these features are strongly confirmatory of the design hypothesis.
A Kettle in the Cabin
Consider the following analogy, which is offered by my colleague philosopher Dr. Tim McGrew. Imagine that you are hiking in a remote area far from civilization and you encounter a cabin in the midst of a forest. Wondering whether the cabin is inhabited, you decide to inspect. As you open the door, you encounter a tea kettle steeping on the stove. On the hypothesis that the cabin is inhabited, this observation is not especially probable. Indeed, the hypothesis is compatible with various observations — the cabin may contain tea, coffee, no hot drink, or be empty. The observation is therefore not strongly predicted by the hypothesis. Nonetheless, the observation is far more probable given the hypothesis than on the falsity of the hypothesis. In view of the top-heaviness of this likelihood ratio, the observation is strongly confirmatory of the hypothesis.
For readers who would like to do a deeper dive into Bayesian formulations of design arguments, I would recommend this podcast I recorded a couple of years ago with Tim McGrew for ID the Future. See also this series of three articles I wrote in 2020 articulating a Bayesian approach to intelligent design (part 1, part 2, part 3).