View of /trunk/chap_concept.tex

\chapter{Promises, Obligations and Intentions}\label{chap_intro}

\section{Examining promises}

Our experience of promises can be both positive and negative; for many
in today's society, promises are things that are made lightly and which are never kept.
This is not a problem: a theory of promises taking the view that promises
were always kept would not be a very good theory, and would suffer
from the same problems that have beset studies of modal
logic\endnote{Modal logic often begins with the logics of necessity,
  i.e. `it is necessary that' and move on to deontic logic, `it is
  permitted that'. Even after 50 years of widespread attention, these
  logics fail to make unambiguous progress.}.

Below are examples of the kinds of statements we shall refer to as
promises. Let us begin with everyday statements and progress gradually
to the kinds of abstract promises that we would like to use in a
variety of technical scenarios.

\begin{itemize}
\item I promise you that I will walk the dog.
\item I promise you that I fed your cat while you were away.
\item We promise to accept cash payments.
\item We promise to accept validated credit cards.
\item I'll lock the door when I leave.
\item I promise not to lock the door when I leave.
\item We'll definitely wash our hands before touching the food.
\end{itemize}

These examples are quite uncontroversial. They are easily found in
every day life, spoken by humans or posted on signs.  We now want to
argue that it is useful to extend the notion of promises to inanimate
objects that have been designed or programmed to behave in a certain
manner. This is not a very large step, but it is easier to make with
some examples.


\section{Extended notions of a promise}

Consider the following promises that might be made in the world
of Information Technology:

\begin{itemize}
\item The Internet Service Provider promises to deliver broadband Internet at a fixed for a fixed monthly payment.
\item The security officer promises that the system will conform to security requirements.
\item The support personnel promise to be available by pager 24 hours a day.
\item Support staff promises to reply to queries within 24 hours.
\end{itemize}
Again these are straightforward promises, which could be described
further to be more specific.  The final promise could also be
restated in more abstract terms, transferring the promise to an
abstract entity: ``the help desk'':

\begin{itemize}
\item The company help-desk promises to reply to service requests within 24 hours.
\item The weather promises to be fine.
\end{itemize}

This latter example illustrates the way that we transfer the
intentions of promises to `entities' that we consider to be
responsible by association. It is a small step from this transference
to a more general assignment of promises to individual components in a
piece of technology.  For example, we can document the properties of
the following tools and technologies in the spirit of this argument:

\begin{itemize}
\item I am a meat knife and promise to cut more efficiently through meat.
\item I am a logic gate and promise to transform a {\tt TRUE} signal into a {\tt FALSE} signal and vice versa.
\item I am a variable that promises to represent the value 17 of type integer.
\item I am a command line interpreter and promise to accept input and execute commands from the user.
\item I am a router and promise to accept packets from a list of authorized IP addresses.
\item I am a compliance monitor and promise to verify and automatically repair the state of the system based on
this description of system configuration and policy.
\item I am a high availability server and I promise you service delivery with 99.9999\%
availability.
\end{itemize}

From these example we see that the essence of promises is quite
general. Indeed such promises are all around us in everyday life, both
in mundane clothing as well as in technical disciplines. Statements
about engineering specifications can also profitably be considered as
promises, even though we might not ordinarily think of them in this
way.

When an electronics engineer looks in a component catalogue and sees
`resistors' for sale promising to have resistance of 500 Ohms to
within a tolerance of 5\%, we do not argue about who made this promise
or whether the resistor is capable of independent thought. The
coloured bands on the component are a sufficient expression of this
promise, and we accept it by association. By this reasoning, we
propose that the concept of a promise should be formulated in way
which allows for all of these uses.

The value of this association is that promises are things that we use
to form {\em expectations} of the behaviour of all manner of
things. Such expectations contribute to reducing our {\em uncertainty}
about their behaviour, and this can apply as much to technology as to
humans. We therefore take it as given that the concept of a promise is
a useful one and consider next how one can formalize promises in the
simplest and least assuming way.


\section{The promise literature}

Because of their overriding ubiquity, and practical importance, one
would like to have an account of promises that captures their key
properties and explains related concepts such as commitment,
obligation and intention.
Although the concept or its
relatives have been mentioned in such diverse areas from logic, law
and philosophy to economics, information science and computing, there
is no agreement on what constitutes the semantic content of the terms
or if there is even more than a tacit relationship between promise,
commitment, obligation etc. The most attention has been given to the
concept of {\em obligations} especially in the area of deontic logic.
We believe on the other hand that the philosophical implications of
promises are far wider than is generally assumed and that there is
both a need and a practical importance to clarify them once and for
all. Indeed, we will show that the concept of a promise is far simpler
than that of an obligation.


Atiyah \cite{atiyah1} suggests that any promise leads to an obligation
to keep that promise that is motivated by the threat of tit for tat
reprisals.  Reciprocation is thus coupled to the idea of promises
immediately, which seems to hop over fundamental definitions directly
to a discussion of the economics of keeping promises. The obligations
are to avoid injury and to reciprocate goodwill. It might be discussed
whether incentives are the same as obligations. Atiyah points out
however that promising something cannot be necessarily used to create
obligation at will.  Promises might cause obligations but they can
also represent obligations that already exist, i.e. to show commitment
to an existing obligation to pay the price of something.  e.g. I
promise to pay the bearer the sum of 1 pound (in gold). This is only
an existing admission of moral obligation.  Atiyah maintains,
plausibly, that the motivation for promising has changed throughout
history.  When people make promises, their intentions are culturally
bound. Thus a Victorian gentleman's conception of a promise might not
fit with that of a present-day child who promises to be home in time for
dinner.

Cartwright takes Atiyah's view and asks what might be the point of
promises if not to generate the assumed obligation\cite{cartwright1}.
Why do people bother to make promises about things to which they are
already obliged? His answer includes the idea that it is a face-saving
measure: to mitigate their humility, suggesting that an obligation is
interpreted as a kind of attack or levy of force? Alternatively, perhaps the
obligation to keep one's promises weighs heavier than the original
obligation (I promise you my word as a gentleman not to kill you, even
though the law says I am forbidden).
Referring to Fried\cite{fried1}, Cartwright points out that the
economics of contractual tit-for-tat suggested by Atiyah is tied
to promises and not to the obligations they might confer.

The idea that promises are an economic driver of contracts or
agreements as bilateral exchanges of promises is continued in the work
of Gilbert \cite{gilbert1}.  Then Carrillo and Dewatripont have argued
that promises can best be understood as a market mechanism for
reducing the uncertainty in a moral-hazard game \cite{carrilo1}.  This
work does not seem to have been pursued. Does a promise increase the
likelihood of voluntary cooperation?
A number of other works mention the concept of promises in the context
of game theory also. In these, the concept of a promise is tacitly assumed to
be related to the probability of choosing a particular game strategy.

More recently, a different motivation for promises was introduced by
Burgess in the context of distributed
management \cite{burgessdsom2005}. Rather than focusing on morals or
even economics as the principal motivator, Burgess uses the promise as
a measure of `voluntary cooperation' as a way of resolving fundamental
problems with logics of obligation for determining system behaviour.
Voluntary cooperation is seen as a way of simplifying constraints and avoiding
many-worlds paradoxes.
He pursues the argument further by emphasizing the role of autonomy of
the parts, and argues for a `promise theory' in which every component
in a system that can have unique information or independent action
should be viewed as axiomatically autonomous \cite{rosegarden}. Any
cooperation or even subordination of the parts that comes about in an
organized system must then be understood as the result of `honouring' purely
voluntary promises to do so. Burgess argues that no matter what one
believes about the power to oblige (even soldiers can refuse to follow
orders), voluntary cooperation can be used as a pragmatic
engineering methodology for mapping out the complexity of a control
problem in a way that is invariant with respect to centralization or
decentralization of systems.

In computer science, particularly the field of Multi-Agent Systems the
concept of {\em commitments} has been used for some time \cite{commit1,agents}. It has been
suggested that promises and commitments are the same. However, we
shall show that this is not the case. More
seriously, the sense in which the term commitment is used in such
discussions is more stylized than purposely considered and can only
benefit from the discussion in this paper.


\section{A model of the structure of a promise}\label{concept_scope}

Consider the following intuitive idea of what a promise might be:
{\em A promise is an announcement of fact or behaviour by a promiser
to a promisee, observed by a number of witnesses (referred to as the
scope of the promise), whose outcome has yet to be assessed.}

The promiser and promisee are both assumed to be `agents', i.e. humans
or inanimate objects to which we attach identity in the story of promises.
This general description fits the examples that we have already given
and gives some clues as to the constitution of a promise, but it also
opens up a number of questions that need answering.  Already we can
see that this apparently basic definition rests on a number of
assumptions: that we can observe the outcomes of behaviours and that
the outcome of a promise is clear at some single moment of time in the
future, to be measured and verified by an observer. A full account of
this might include a theory of measurement, but we wish to avoid this
level of detail as it binds us to too many details that have nothing
directly to do with the issue. Let us instead try to understand
to essential characteristics for promises and consider what
distinguishes a promise from related matters, such as obligations, commitments and other terms.

The model world in which we formulate promises must have the following
characteristics.

\begin{itemize}
\item There must be agents in order for promises to exist.
\item There must be a promiser (or source agent).
\item There must be a promisee (or recipient agent) which might be the same as the source.
\item There must be a body which describes the nature of the promise.
\end{itemize}
We might summarize these attributes with a notation as in ref. \cite{burgessdsom2005}:
\beq
promiser \promise{body} promisee
\eeq
\begin{itemize}
\item We can leave the body unspecified, but it
must consist of a quality (a type, topic or subject for the
promise) and a quantifier (which indicates how much of the realm of
possibility for that subject is being promised). For example: promise quality: ``travel to work'', promise quantity
``on Monday and Friday each week''.
\end{itemize}
Finally, what is implicit in the above is that a promise requires the
transmission of a message, or at least documentation in some kind of
physical form, e.g. a speech act, or a written statement, else it
cannot be made known to anyone except the promiser.  A promise must
therefore have documentation that is made intentionally or
otherwise. 

What then is a promise before we write it down? We shall refer to this
as an {\em intention}.  An intention is the basic formulation of a
course of behaviour, which is made internally by (or on behalf of) an
agent.  When an intention is made public, it becomes a promise.  If an
intention is documented or leaked in some way then anyone has a right
to assume it is a promise\endnote{We use the term `intention' in its common meaning. In Philosophy, one speaks of `intentionality', which is a much broader idea about mental awareness, that deals with the extent to which something can be `about' something else.}.

We take it as given that there has to be a source for every promise. A
promiser does not have to reveal its identity of course, so witnesses
to the promise might not know its source e.g. consider the anonymous
threat.  There is no reason to deny the existence of a source
however.  The lack of such information about a promiser is simply a defect in
the knowledge of the receiving agent, but one would normally prefer to
assume a consistent picture of promises and infer the existence of an
anonymous promiser. This justifies our postulating the source.

\section{Promises are documented intentions}

A key characteristic of a promise is that it documents an intention, so
let us explore the idea of intentions in more detail. Intentions turn out
to be a lowest common denominator for all of the concepts discussed in this paper
and thus have a special importance.

Since promises involve communication we require a notion of the spread
of information amongst the agents. We use the term {\em scope} for
this.  A scope is simply defined to be a set of agents.  For example,
the scope of a promise would typically refer to the promiser and a
list of witnesses to the promise, e.g. those who heard to utterance or
those who saw the written document.

\subsection{Intentions}

The realm of all possible formulations about behaviour is covered by
the concept of intentions.

\begin{definition}[Intention]
A description of possible behaviour that can be contemplated by an
agent and be brought to realization.
\end{definition}

The components of an intention are as follows: a source agent who formulates the
intention, a target agent if the intention is directed at a potential
subject, and a body which explains the quality and quantity of the
intention (see fig. \ref{intention}). Only the source of an intention
knows about the intention, i.e. the scope of an intention is the source only.
There are no witnesses.

Now we must be careful: the set of all possible intentions should be distinguished from actual
instances of intentions selected by an particular agent. We shall
sometimes use the phrase ``possible intentions'' to mean this full set
of abstract entities to emphasize when we wish to signify a general
description of behaviour rather than an individual
agent's decision.

An intention is not announced by the agent holding it to any other agents. 
Indeed, we may now \emph{define} any intention
that is announced to be a promise. Conversely  we notice that any promise that has
not been announced is merely an intention.
Some intentions are desirable while others are absolutely undesirable
and an agent might never choose them, yet they are possible intentions
nevertheless. The fact that such behaviours can be intended
is enough for them to qualify as intentions (``possible intentions'').

Intentions must always be thought of as belonging to a specific agent. Those
intentions which are actual plans of the agent are called its commitments. Other
phrases for a commitment that we may use are: intended intention, or real intention.

Due to the static nature of our account we pay no attention to the
process by which an intention might become a commitment or vice versa.

%Whether we choose to talk about the intention behind a promise or not,
%for every promise there is a corresponding intention XX. this is different from necessarily having this intention. A lie,
%for example, is a promise with an intention that is different to the
%one signalled by the promise.
%

\subsection{Promises}

A promise is the physical publication of an intention within a certain
scope. This suggests that there must be some agent to observe the
promise and its outcome which in turn requires the expensive notion of a
theory of observation so we shall tackle this issue separately (see
section \ref{assessment}).

\begin{definition}[Promise]
A promise is an intention that has been documented within a scope that goes beyond
the promiser.
\end{definition}

Promises thus have scope. Formally intentions also have a scope, but
the scope of an intention held by an agent is by definition limited to
the agent (source) itself. An intention could be leaked deliberately
(e.g. to the press, in order to influence someone). This might be a
form of leverage, or an attempt to impose an obligation on some party
in scope. However, at the instant an intention expands in scope to
encompass more agents it becomes a promise. A so-called letter of
intent, for example, is a promise rather than merely an intention.

%The notion of scope is useful, as with promises we have intended scope
%and actual scope. Again the notion of intention is primary and underlies
%many aspects of promises.

\begin{figure}[ht]
\begin{center}
\includegraphics[width=12cm]{figs/intention-parts}
\caption{\small Intentions and their structure.\label{intention}}
\end{center}
\end{figure}

\begin{figure}[ht]
\begin{center}
\includegraphics[width=10cm]{figs/promise-parts}
\caption{\small Promises are intentions with documentation and an audience.\label{promise}}
\end{center}
\end{figure}

The time aspect of promises presents further challenges.  Intentions
can become outdated by events.  An event which is found to fulfill and
intention documents the implicit promise, since one must admit to the
intention in a wider scope.  Conversely, the documentation for a
promise does not have to last for ever; if documentation of a promise
disappears completely, it reverts to being an intention.  A promise to
oneself is merely an intention, unless it is documented.

The distinction between the promisee and the scope of the promise is key
to understanding promises.  Suppose someone intends to arrange a
surprise birthday party for their friend. Initially this is an
intention. The intention is written in a diary or mentioned to a third
party and it becomes a promise. The target is not in the scope of the
promise however, so the promise remains unknown to the jubilant.
However, suppose that before telling anyone else, the promiser
destroys all evidence of the promise by tearing out the page of the
diary, effectively withdrawing knowledge of the promise, then the
promise reverts to being just an intention. But as long as knowledge
of the promise remains ``out there'' in the world, it remains a
promise that has been made.



\subsection{Obligations}

Having explained intentions and promises, let us now try to
describe the notion of obligations in the same manner. The intuitive
notion of an obligation seems straightforward, but it proves to have difficult
properties.
We might try to think of obligations in a straightforward way, for instance:
{\em an obligation is an intention that is perceived to be necessary by an agent}.
This certainly captures some of the characteristics that we understand by
the term, but it also leaves many questions unanswered: is the feeling of the necessity
voluntary or forced, a matter of survival or simply an authoritarian convention?

Unlike a promise, an obligation might be self-imposed or externally
imposed.  An obligation falls into the category of (possible)
intentions, so it must have source, a target and a body, and the body must
have a quality and a quantity. The source and target are now somewhat
difficult to understand however.

Beyond this, we shall not attempt to define obligations more carefully
in this paper. We shall merely state some assumptions about them.

An obligation can be imposed by external conditions, e.g. by the
expected behaviour of external agents, by laws threatening sanctions
etc, or it can be self-imposed by codes of personal behaviour which an
agent holds to be {\em necessary}. But this imposition suggests the
action of a force which attempts to induce a commitment in another
agent (or itself).  An obligation is an intention (``possible
intention'') which may or may not have the status of a commitment.  In
any case the agent is aware of any compelling reasons to include the
intention in the portfolio of commitments, either from within itself
or without due to external forces.

It seems natural then to refer to the source and target of the induced
intention as being the agent in which one attempts to induce the
intention, and the recipient of the intention respectively. However,
the source of the obligation itself might not be an agent at all, but
merely a set of external conditions, norms, experiences or other
information acquired by the agent that lead to a perceived priority.

Note again that even `forced' behaviour can be classified under the
realms of (possible) intentions since all behaviour can be
intended. Again, we emphasize that this does not imply that a coerced
agent holds the intention that is being forced upon it. Nor does it
say anything about whether the agent is able to resist the force or
not, or whether it matters if an obligation is self-imposed or
externally imposed.

The notion of an obligation immediately seems far more complicated
than an intention or a promise and does not seem to be close to the
notion of either promises or intentions.
%What we can say is this:
%\begin{itemize}
%\item An obligation is `observed' if the agent that perceives an obligation
%forms the intention to comply with the obligation.
%
%\item An obligation is `met' if the body of the obligation is achieved, with or
%without the compliance of the agent.
%\end{itemize}
%THESE STATEMENTS ARE NOT STATIC AND HAVE BEEN 
%REMOVED FOR THAT REASON. MEETING OBLIGATIONS WILL BE 
%CONNECTED TO ASSESSMENTS
We hold that obligations are far from being a reliable tool for
ensuring compliance. If a law-giver wanted to ensure the compliance of
an agent, a better strategy would be to obtain a promise from the
agent, and to convince it to view the intention as a {\em commitment}
since the law-giver could never know whether the agent had indeed
committed to the body of the obligation.

\begin{figure}[ht]
\begin{center}
\includegraphics[width=8cm]{figs/universe}
%\epsfig{file=figs/universe.eps,width=8cm}
\caption{\small The voluntary cooperation universe and the ranking of intention. Externally applied
obligations can overlap with any of the categories of formulatable
intentions.\label{uni}}
\end{center}
\end{figure}
To study the idea of compliance further let us return now once again
to promises. It is clear that promises and obligations cannot be
simply related (as some promises might be deceptions) so let us explore
deceptions in more detail.

\subsection{Deceptions -- non-intended promises}


Understanding deceptions (or lies) is also an important step in
clarifying the relationship between intentions and promises, because
it is possible for an agent to have two different intentions in play
at the same time: a commitment and an announced intention (i.e. a
promise) which are not compatible. Incompatibility means that striving
for both intentions simultaneously is fruitless because their
realizations cannot be combined.

In a deception, there is always a source and always a target and the
target cannot be the same as the source, as an agent cannot
(intentionally) deceive itself. Furthermore, we maintain that the target of a deception
must be in scope, so there must be a physical documentation and hence
a deception necessarily involves a promise and not merely an intention.

\begin{definition}[Deception]\label{concept_deception}
A deception consists of two intentions: a documented intention (i.e. a promise) and
a non-documented commitment, which are incompatible. 
\end{definition}
The non-documented commitment will also be called the hidden intention.

In a deception the hidden intention is more important than the
witnessed one one and we might refer to it as the dominant
intention. This simply expresses that it is a commitment while the
promise contains merely a ``possible intention''. It is the {\em real}
intention of the agent (``intended intention''), while the intention
in the promise can merely be described as {\em non-real}.  If the
dominant intention should be rescinded, a deception will revert to
being a promise, but this is only known to the source.



\subsection{``Non-deceptions''}

A deception is the augmentation of a promise with an {\em incompatible}
intention. We should like a name for the augmentation of a promise
with a positive intention.  We might call this a promised commitment, or
intended promise.
%Such an object would be the the complement of a deception (a non-deception),
%or the documentation of an intended intention. We shall refer to an ``intended intention''
%below as a commitment.
%
From these slightly strained terms, we can now appreciate why the
concept of a promise is in fact so important. A promise is simply a
promise (the documentation of an intention), regardless of what lies
behind it.  Any internal priorities or considerations are hidden from
the view of other agents and cannot be observed.  Thus, promises are
an independently important concept because we can (indeed must) talk
about promises without discussing the basis on which they are made.

When a promise is made, we are neither required nor able to confront
the truth or falsity of the promise.  Indeed, as soon as we ask such
questions, new issues such as trust and a plethora of other subjective
issues come into play.  Such issues are probably un-resolvable in a
logical sense. However, what we assume is here that no matter how
trustworthy a promise might be, it can increase or decrease our
certainty of a promised outcome and thus it bears an {\em influence}.
The matter of assessing the promise can be very complicated and
uncertain and we shall not attempt to discuss this here in any depth.


\section{Ranking of intentions}

The foregoing discussion of deceptions suggests the existence of a
ranking function which induces a partial ordering onto the intentions
that are referred to or in play at any given moment. There are
intentions one intends to invoke (i.e. that one commits to),
intentions one prefers, intentions one feels obliged to intend, and
finally deceptions which one intends to not honour the intention
documented for a wider audience.

\subsection{Commitment}

%I HAVE CHANGED THIS A BIT - TO MAKE COMMITMENT INTENDED INTENTION

To commit to something is to make a decision in favour of it.  The
issue of commitments is therefore about the favourization of
intentions. Commitment is a personal decision and has nothing to do
with physical representation or communication, thus the issue of
commitment precedes any discussion of promises.  A commitment has a source, a
target and a body, i.e. it is an intention. Like an intention it
has no physical representation and does not have a non-trivial scope.

Once an intention becomes a commitment we often assume that some point
of no return has been passed in the act of committing (deciding) about
the particular intention.  i.e. adding the intention to a list of
commitments.  For example, in a game of chicken in which two cars
drive towards each other to see which one will swerve off first, a
driver has committed to not swerving when the decision to not back
down has been made\cite{schelling1}. This might have certain
irreversible consequences, but it is difficult to generalize the idea
of irreversibility in examples of this kind. What commitment
essentially boils down to is the elevation of some intention beyond an
arbitrary threshold. In other words, in the universe of intentions
there is a subset of these which we may call commitments. 
%Conversely, we can have intentions to which we are not committed.
%THIS IS MYSTERIOUS: as commitments = real intentions.


\begin{definition}[Commitment]
Commitments are intentions that we are committed to. We may call them
intended intentions, or equivalently real intentions, intentions that
we hold, or committed intentions.
\end{definition}


\subsection{A partial ordering}
%
%To decide that certain intentions are in play, and others are latent,
%we have introduced a value function or partial ordering on the list of
%intentions. When we commit, we select the importance of a particular
%intention, or make the intentions active. We can also make intentions
%latent or remove them altogether.
%
%
%
%Commitments are at the top of this partial ordering, then come perhaps
%voluntary obligations (which we have not committed to). A preference
%is an intention we would like to do. - but maybe not commit or feel
%obliged. Next are intentions which we don't care much about, and
%finally there are deceptions intentions that are advertised in
%deceptions 
A given agent at a certain moment of time ranks intentions by applying
to them a {\em partial ordering}. An intention is considered higher in
this ordering if it is closer to a commitment. Commitments are at the
top of this ranking; at the bottom are those intentions which are
incompatible with commitments. The latter are termed negative
intentions or contrary intentions.  In between these extrema lie
voluntary obligations, and involuntary obligations, and these will in
general intersect with all other categories(see fig. \ref{uni}).

\section{Assessments}\label{assessment}

The notion of whether promises are kept or not is central to their
sustained usefulness in language, thus we need to make mention of how
this comes about in a theory of promises. It would be easy to go
overboard and delve into the complexities of observation and
measurement to provide a satisfactory answer but that is not in the
spirit of this paper. We seek instead a simpler notion which is at the same
level of abstraction as the concepts of promise and intention that we
have introduced thus far. We call this the concept of {\em
assessment}.

\begin{definition}[Assessment]
An assessment is a subjective statement made by an agent about whether
the intentions of itself or of another agent were fulfilled. 
\end{definition}
Our notion of an assessment is more generic and less quantitative than
a verification. It is both subjective and not {\em a priori}
linked to observation. In a static theory of promises and intentions the existence
of intentions as well as the value of assessments is linked to state parameters like time.
Thus, for an intention of agent $A$, in existence at time $t$, it may be the case that agent $B$'s 
assessment, made at time $t'>t$ is positive (or negative).

At this level of description, we need not say any more about it than
this.  The reality of whether promises have been kept through specific
actions is neither here nor there in the world of politics and to some
extent economics. What is important is how a witness to the promise
assesses the outcome of the promise. Such an assessor may or may not
feel obliged to assess an outcome in a particular way, might promise
to conform to certain criteria, and so on. What matters is only the
assessment, which might or might not be rationally obtained.  We
believe that this is a fair model of the world in which we live.

An assessment is a supposed outcome relative to some method of
assessment.  Assessment involves a variety of possible routes to
inference, i.e. there are different kinds of assessment. This is a
subjective issue, but this should not be viewed as a weakness of our
theory: it is a true feature of the subjective nature of individual
assessment.

Some example assessments are shown in table \ref{tab}.  We see that
assessments are quite sensitive to physical representation of the
promise.  Once again the notion of representation (or documentation)
is a key to the importance of a promise as a concept.

\begin{table}
\small
\begin{tabular}{l|l|l|l}
Promise       & Representation & Scope & Assessment \\
\hline\hline
Fed the cat            & Speech act  & Cat owner             & Either did or did not.\\
Credit card accepted   & Action      & Customer visitors     & Either did or did not.\\
Response in 24 hours   & Contract    & Signing parties       & Reply in time or not.\\
$var = value$          & Source code & Programmer / compiler & Syntax ok, value in range\\
$var = value$          & Object code & Execution engine      & Valid instruction/exception.\\
\end{tabular}
\normalsize
\vspace{0.2cm}
\caption{\small Promise assessments\label{tab}}
\end{table}




\section{The value of promises -- relativity}

Promises are valuable to agents, because they help reduce uncertainty
and because their outcomes could be beneficial if they become certain.
Because certainty is key, a promise is worth nothing unless there is
trust. Zero trust makes promises worthless. Trust might be based on a
history of keeping promises or, in our terminology, on a history of
positive assessments about a succession of promises. So there is
a symmetry between trust and promises that must be broken to solve the
dilemma.

If there is trust, a promise about future behaviour does not need to
be perceived as an obligation on the promiser but as an indication
that best effort will be respected.  If a given agent $X$ does not
trust the promiser however, it might assume that the promise implies
an obligation on the promiser. This perception of obligation by $X$
does not of course imply an obligation perceived by the promiser.
There is a fundamental subjectivity in these perceptions.

The value of a promise is an expectation of the eventual
benefit. Suppose, then that $A$ promises $B$ 400 dollars per year. $B$
promises to wash $A$'s windows at this price.  Both are satisfied with
the value they get from this arrangement and prefer not to question it
too much as this could unleash all kinds of consequences.  Observer
$C$ can see that the values are quite mismatched, or that $A$ is
getting a poor deal by its judgement, but $C$ also cannot deny that
the relationship is stable because both $A$ and $B$ are happy.

The value of promises may be questioned by those who consider promises
as a concept secondary to obligations. If one views obligations as the
primary concept, the value or importance of promises unavoidably
shrinks. We shall now survey advantages and disadvantages of
obligations as an alternative cornerstone of a theory of promises. The
discussion will be somewhat asymmetric because we will not base our
comparison on a proper definition of the concept of an obligation
(which we cannot fathom). Suffice it to say that for some agent $A$ an
obligation is an intention (``possible intention''), which has been
elevated to the status of an obligation, whatever the consequences of
this status may be.

\section{For and against the primacy of obligations}

Obligations are discussed extensively in the literature whereas
promises have been ignored. By the sheer weight of tradition,
obligations dominate discussions of behaviour.

\subsection{In favour of obligations}

\begin{enumerate}
\item Some people might think that a promise is an obligation because it seems
to create one, and might therefore be considered equivalent to that obligation.

\item Obligations are a well known concept from deontic logic. There is an advantage to to reducing
the less well-known concept of promises to one that has been studied
for more than fifty years.

\item Obligations have a formal status in state laws and regulations. There is no such
public body of promises.

\item Many obligations give rise to promises which occur in the process of fulfilling
an obligation. e.g. the cat must get fed while owner is on holiday,
the owner is obliged to get the cat fed (by law forbidding cruelty to
animals). A friend promises to help in the fulfillment of the obligation.
\end{enumerate}

\subsection{Against obligations}

\begin{enumerate}
\item If a future promise (e.g. the promise to feed the cat in the future) is
in fact a deception then this is falsified the necessity of a
relationship between promises and obligations.  In other words, all
promises cannot be obligations because some promises can be deceptions
and these cannot be understood as such.

\item All descriptions of deontic logic are fraught with logical difficulties.

\item Not all promises are about future actions, so there cannot be an implied
obligation for all promises. e.g. I promise that the cat got fed. Indeed the owner
might actually be displeased that the cat was fed if it was supposed to be dieting.

\item In law, it is true that there is a dissimilarity between promises and obligations. They are
quite different entities.  Obligations may cause promises and promises
may cause obligations, but promises have a physical representation in
space and time, whereas obligations do not.  Obligations are at a
different level of abstraction altogether. Moreover, the international
monetary system can be viewed as an example of a de-facto standard
promise -- the promise to redeem the value of money on a voluntary
basis. It is also well known by law-makers that laws are only observed
about issues that most people will basically keep to voluntarily, thus
the power of obligation may be a fictitious one, based on de-facto promises.

\item Promises are made on a voluntary basis. For obligation however, the
concept of voluntarity is almost irrational. In any case it might be voluntary to
imply an obligation on someone else, but engaging in a promise you may face an
involuntary obligation or a voluntary one. Voluntary is therefore natural for
promises but is quite problematic for obligations.

\item Promises announce positive extensions of user behaviour, obligations
are a negative constraint on the degrees of freedom of the obliged party.

\end{enumerate}

If one would choose between promises and obligations, it seems
abundantly clear that promises are the simpler concept. Moreover, the
concept of a promise seems more natural in the technological world:
since computers cannot feel ethical responsibility, the reduction to
promises to obligations seems to be neither philosophically
satisfactory nor technically correct.
Moreover, there are some behaviours one cannot oblige (empty the ocean
with a sieve). These can be promised, even if the promises are clearly
deceptions.


So far we have argued that promises are different, simpler and can be
analyzed independently of obligations. There is one more point that is
of principal practical importance.  Promises are {\em local}
constructions, whereas obligations are {\em non-local}. The source of
a promise is localized in a single entity that has all of the
information and self-control to be available to resolve conflicts and
problems with multiple promises.  The sources of obligations however
are distributed amongst many individuals and the obliged party does
not have the access to resolve the conflicts without maintaining a
voluntary dialogue with all of these multiple parties.

From a practical viewpoint, obligations are simply less effective at
reducing uncertainty because they tend to increase uncertainty not
reduce it. Indeed, obligations can be inconsistent, but promises
cannot. More precisely: consistency of promises is a matter that can
be verfied at the level of sources only. Promises made by different
agents cannot be inconsistent.

Preferably then one would not use obligation as a coordination
principle if a mechanism based on promises can be used
instead. Promises are simply more trustworthy. A collaboration based
on promises works better if one has trust. In a world of obligations
however, trust is meaningless because one has only a presumed outcome.


\section{Summary}

Without attempting to suggest applications in any field, we have
argued for the usefulness of promises as an independent and practical
concept, whether in philosophy, economics or technology. We have
compared promises to the more usual idea of obligations and have
concluded unequivocally that promises are a simpler theoretical notion
and a more practical tool than obligations in the reduction of
an agent's uncertainty about the behaviour of other agents.

We show that intentions, promises and commitments can be explained in
the absence of an understanding of obligations. Furthermore, although
it seems to be a common view that obligations are a more primitive
concept than promises, our paper suggests the contrary.  Promises need
not be viewed merely as proxies for obligations; if promises give rise
to obligations, this can in fact be studied purely in an exposition
based on promises, intentions and commitments.  Indeed more often than
not promises are made by agents who would not be able to explain the
extent to which their promises might lead to obligations or not, or to
what extent such obligations would be more significant than the
promises from which they arose.