Module 1 - What is Risk?

b. Control-based and Modal-based accounts

There are two non-probabilistic ways of thinking about risk in the literature, that is, in terms of the absence of control or modal fragility. For example, Broncano-Berrocal (2015) discusses two kinds of risk: event-relative and agent-relative. A significant event is risky in the event-relative sense just in case it is sufficiently modally fragile. An event, E, is sufficiently modally fragile if and only if E occurs in the actual world but—when holding the relevant initial conditions for E constant—there is a large enough selection of nearby possible worlds in which E fails to occur. For example, buying a lottery ticket is a matter of risk because even if one’s ticket in the actual world is a winner, there is a dense selection of nearby possible worlds in which one buys a ticket but loses the lottery. It is a small change to reposition a few lottery balls.

In contrast, a significant event, E, is risky in the agent-relative sense just in case the agent lacks control over E, and an agent, S, “lacks control over E if and only if S lacks either effective control, dispositional tracking control, or both. Context and the type of event that E is make salient the form of control that is relevant” (Broncano-Berrocal 2015, p. 22). Broncano-Berrocal defines effective control as follows: 

A has effective control over B if and only if (i) it is A’s aim that B is in a certain state S, (ii) A has a disposition to cause B to be in S, and (iii) B is in S because of A’s disposition to cause B to be in S. (p. 19) 

A classic example of effective control is when a driver safely drives her car to a desired location. Broncano-Berrocal defines dispositional tracking control as follows:

A has tracking control over B if and only if A monitors B. Monitoring has two components. When A monitors B, A keeps track of, registers, or gathers information about B. This is the epistemic or informational component of monitoring. In addition, the information that A registers about B enables A to initiate, stop, or continue some performance or action that contributes to the achievement of the relevant goal; in a sense, the information compiled disposes or puts A in a position to perform goal-directed actions. (p. 20)

An example of tracking control is when a doctor monitors a patient’s blood pressure via regular testing. If the patient’s blood pressure is either too high or low, the doctor will administer drugs to ensure that the patient’s blood pressure is within a healthy range. Thus, regardless of whether the doctor administers drugs, there is a sense in which the doctor has control over the patient’s blood pressure.

  
According to Broncano-Berrocal, there are two kinds of non-probabilistic risk. A significant event may be risky to the extent that it is modally fragile, and an agent may be at risk with respect to a significant event just in case the agent lacks effective or tracking control over the event. These distinctions lead to four possible types of cases. Let us follow Broncano-Berrocal in labelling them A, B, C, and D-cases.

A-cases involve both a lack of control and modal fragility. For example, one lacks control over one’s fair lottery win, and this event is also modally fragile. 
 

B-cases involve a lack of agential control but are modally robust. For example, if unbeknownst to you, your friends rig a lottery in your favor such that you are certain to win, then, although you lack control over the result of the lottery, your lottery win is an event that will occur in nearly all nearby possible worlds. 

C-cases involve neither sense of risk. For example, when you successfully rig a lottery in your own favor, such an event is under your control and, depending on how skilled you are, could be modally robust. 

D-cases are under an agent’s control but are not modally robust. Examples might include impulsive decisions or decisions made on a whim. If on a whim, I decide to skip work and fly to Paris, then such a decision is under my control but could have easily not occurred, that is, is modally fragile.

Control theorists hold that risk necessarily involves the absence of agential control; hence A and B cases are risky. Modal theorists define risk in terms of modal fragility; hence A and D cases are risky. Who is correct? Both views have advantages and disadvantages.

An advantage of control-based accounts of risk is that agents often feel at risk regarding events that they have no control over. Furthermore, control is an important concept in many other related philosophical debates, for example, in discussions concerning luck and moral responsibility. 

A disadvantage of control-based accounts of risk is that it seems rather odd to describe many B cases as risky. Consider that no one has any level of effective or tracking control over whether the sun will rise tomorrow, but this event doesn’t seem to involve much risk. Consider further that if today is Friday, then tomorrow is Saturday. No one has effective or tracking control over this fact, but this event doesn’t seem to be risky. 

One advantage of modal accounts is that people often overlook counterfactual possibilities when making important decisions. But properly assessing what if scenarios can help people better manage risk. For example, consider the space shuttle Challenger disaster. Before the disaster, engineers at NASA knew that during previous flights foam debris had detached from the shuttle. However, this debris never hit a sensitive part of the shuttle, and these flights were thus viewed as successes instead of the near misses they actually were. Had higher-ups at NASA realized that the debris could have easily hit a different part of the shuttle and that this could lead to the shuttle exploding, then this disaster could have been avoided.

However, there are several problems with modal accounts. For one, we often don’t have access to counterfactual facts about alternative scenarios. Descriptions about what would have occurred had the situation been different seem to have truth values, but it’s difficult to say what these values are. But without such information, it can be difficult if not impossible to apply the modal account of risk. Second, which alternative scenarios or possible worlds are suitable for comparison with the actual world? Some description of the event and set of initial conditions to hold constant across possible scenarios is necessary to make counterfactual comparisons. But how should we set these conditions? It’s unclear, and this leads to modal accounts suffering from a kind of reference class problem that is similar to the reference class problem that troubles frequentist accounts of probability (Hill 2022).

For an event to be a matter of risk, it seems that there must be something at stake. The outcome of a risky wager must in some way make a subject or subjects better or worse off. In other words, the event in question must be significant. Below are three arguments in favor of this conclusion. 

First, it seems odd or strange to think of genuinely chancy but non-significant events as risky. Consider the following passage from Neil Levy:

An event or state of affairs can count as [… risky] only if it is significant. It may be genuinely chancy how many hairs fall out of my head today. It may even be metaphysically chancy ... whether I have an odd or an even number of hairs on my head at 12 noon. Nevertheless, it would be strange to say that it is a matter of [risk] … It seems to me odd, to say the least, to say that [… it was risky] that the coin I just drew out of my pocket came out tail first. This event would be [… risky], it seems to me, only if something rode on it (I had a bet with someone, or what have you). (2011, p. 13)

Levy’s reasoning is as follows. First, come up with an example that is chancy but not significant. Is such an event risky? Intuitively, such events are non-risky. It would be rather odd or strange and a violation of our folk conception about risk to claim that non-significant events can be risky. However, if you add to the case that the event is significant, we no longer withhold from describing the case as risky. Thus, an event is a matter of risk only if it is significant.