Memory and Commitment to Do
Thor Grünbaum (University of Copenhagen)
Imagine that you are assembling a complicated piece of Ikea furniture. At some point, you need to pause the activity to have lunch. At a later point in time, you resume the activity of assembling the piece. Or imagine that you decide to stop by the bookstore on your way home from work later in the day. At a later point in time, you stop at the bookstore to buy a book.
In these kinds of situations, it looks like agents are able to simply resume their postponed activity or engage in their planed action without also engaging in new deliberation and decision making. The agent can remember doing something without having to reconsider whether to do the action. And even if the agents do not repeat their deliberation and decision process, the later actions are genuine expressions of autonomous and rational agency.
You already settled on assembling the Ikea piece, so you can just resume the activity. You already decided to stop by the bookstore, so now you just engage in stopping by the bookstore. You do not have to reconsider reasons for doing the action and settle on doing it again.
But if that is the case, agents must have some way to retain and make occurrent their original commitment to act.
It is not enough that agents retain and retrieve some representation of what they are doing (assembling my new closet, stopping by the bookstore). Agent must also retain their commitment to perform these acts in the current context. Otherwise, when retrieving a representation of the action of assembling the piece, the agent still has to decide whether to engage in the action.
Retainment of one’s commitment to perform a specific action under specific circumstances seems like a form of memory. After all, agents can and often do sometimes forget to do the things they are committed to doing. But it is a strange form of memory.
Remembering my commitment to act is not like episodically remembering the happening of some event or semantically remembering some fact. When remembering my commitment to act, I do not represent my commitment. Rather, I am committed. Remembering is here the way in which my commitment is transformed from a standing to an occurrent commitment.
What would it be like to be an agent without this special form of long-term memory for commitment? This would be an agent that could only maintain their commitments to act in their working memory for projects that currently occupy their attention. As soon as the project is interrupted or the planed action is stored for later execution, the agent would only be able to represent engaging in past activity or facts about the activity.
As a clue to what this might be like, consider the fictional case of Leonard Shelby in Christopher Nolan’s movie Memento. Mr Shelby has lost the ability to form new long-term memories, and his practical life is trapped in the present moment.
Mr Shelby can write himself a note for the future and hope that, later, when reading the note, he might form the belief that his past self wanted him to do something. But since the later Mr Shelby has no long-term memory of the past and no persistence of his own earlier attitudes and commitments, he reads the note without being able to provide any context. He reads it as you would read an alien message in a bottle.
Mr Shelby is thus unable to know whether the message “Find him and kill him” is an expression of an intention, an intention already performed, or an order for someone else. Or maybe it is an abbreviated way to describe some fact. Just imagine the problems Mr Shelby must have interpreting the personal pronoun.
To be sure, Mr Shelby has lost more than just the ability to remember commitments. He has lost the ability to form new long-term representations. He cannot retain any episodic representations of past events (since after the incident). He therefore has to document everything in writing or by pictures.
Now imagine that we pushed this representational activity of documenting inside Mr Shelby’s brain. According to this “vehicle internalist” view, biological memory of past events can be iconic, propositional, or have whatever format you like, as long as it has a declarative form. This “internalist Shelby” is like an agent with a declarative long-term memory but no special memory for commitments.
Without any special long-term memory for the commitments, when “internalist Shelby” “internally” encodes declarative messages for the future self, it is just like leaving notes or other signals around for a future self to interpret. It is not sufficient for the stability of Mr Shelby’s intentions.
At the very least, receiving a message from the past in this way (say, “kill him” or “stop by bookstore”) should induce “internalist Shelby” to consider the import of the message before deciding what to do. Receiving an internal message or external note from a past self should induce Mr Shelby to deliberate about whether to perform the action.
Mr Shelby’s predicament is of course special. He has no way of verifying the information about his past self’s motivation, decisions, and goals. He constantly needs to reconsider his options. But in its extremity, the case tells us something about the normal situation.
Temporally extended actions like assembling an Ikea piece or planning an action in the future require stability of the agent’s commitments, intentions, and plans. This form of stability requires that agents are not constantly reconsidering whether to perform the actions they have decided to do.
The stability of one’s commitment to act requires that when an agent forms the commitment, the commitment is retained in some long-term memory format. This form of memory for commitments to act enables not only that commitments persist from formation to execution, but it also enables an agent to remember their intention to act when they plan other and potentially conflicting actions.
This form of long-term memory must enable agent to remember their commitment to act at various points in time without systematically inducing agent to reconsider their intention.
If we accept that too much reconsideration undermines stability of commitment, then memory for commitment cannot just be a way for the past self to leave (mental or “external”) declarative messages for the present self.
Interestingly, this point has implications for a radical extended mind thesis of memory for commitment to act. In the total absence of the relevant form of “internal” long-term memory for one’s intentions, plans, and future actions, using external resources such as electronic reminders and calendars seems insufficient for the stability of one’s intentions and plans.
Accepting these musings, there are at least two different ways to conceive of memory for one’s commitment to act.
First, we can conceive of it as a form of episodic memory where to remember one’s commitment to act is to believe a bunch of propositions about oneself. I remember having been engaged in assembling the Ikea piece, I remember saying to myself “at 2 pm, return to assembling the piece”, etc.
All these pieces of information might be a good reason to engage in the activity now, so I decide to do it now. But the information on its own does not amount to a commitment to act. The consequence is that agent engage in too much reconsideration of their decision to act. This would undermine the stability of commitments and plans over time.
Second, we can conceive of memory for commitment as the way in which a standing (non-operative, non-occurrent) long-term state becomes operative and occurrent in working memory.
In other writings, I have called this conception of one’s memory for one’s commitment to act the state-transition account. This is the type of memory we need to account for our amazing ability to assemble complicated pieces of Ikea furniture.
