Serious philosophical quips:
- The problem is that we cannot use x as the utility of s unless we equate it with the expectation, which requires further analysis. My finesse of this problem will be to use x defeasibly. (SEP, 1988)
- We cannot say that L is incorrect. All we can really say is
that for us, the use of a particular language L is cumbersome,
or that despite our best efforts, we cannot quite learn how to use it
properly, or at least, that the payoff of trying to use this language
in those few glowing successes does not exceed the cost of the multitudinous
errors in our unsuccessful attempts to use it. Having said this, we should
perhaps decide what we would regard as progress in the design of ...
inference systems. ... Are we inventing better systems of inference, or
are we just imposing new constraints on translations?
... Rationality cannot be guaranteed simply by demanding that some
privileged language be spoken. (TARK, 1988)
- Deriving entailments merely expands shorthand. ... Construing inference as an association between a set of sentences and their entailment simply reduces the range of language. ... What I am disputing is the claim that there is an interesting sense in which induction is ampliative, while deduction is not. ... Conclusions are nonmonotonic in computation. This is what essentially frees them to reach beyond explicit and implicit representation. ... Nonmonotonicity in computation is internal temporal credal nonmonotonicity. ... Ampliative inference is the result of rational nondeterministic nonmonotonic computation. ... In order to conceive of an inference that is not merely a rewriting of the agent's specified commitment, we have to turn our thoughts to computation. (Philosophy & AI, 1990)
- Why invent beautiful formalism if complex constraint can dissipate in insipid interpretation? Because we want to construct attitudes, not just have them. (J. Philosophy, 1991)
- Dialectic is ... a fair control procedure under limited computation.
(Minds and Machines, 1991)
- Sometimes there is proof; mostly there are arguments. (Opening line in "Process & Policy," tech report, 1992, rejected from publication until Computational Intelligence printed it in 1998; mimics the first line of JM Keynes' thesis)
- A player who chooses not to respond to a concession with his own
concession can instead try to defeat all arguments that are proposed in
defense of the original proposal. (Computational Dialectics, 1995)
- Now we have a better representation for mental state than the real-valued utility, and it is time to invade economics. ... AI possesses the powerful idea, the Mongol's stirrup, the longbow: the ability to model computation. The social sciences are in a position to accept our terms. (Strategic Directions in AI, Computing Surveys, 1995)
- This daubery of Hart will not stand. ... It does not suffice to
refute Hart by citing examples wherein ascription of action is possible
without ascription of responsibility. The principality of ascription must
be refuted, not the necessity of ascription. ... But defeat based on
effective counterargument is "something quite different," and Hart could
be no more explicit. ... Hart shows here, as elsewhere, his awareness
that the procedural aspects of argument and counter-argument are what
distinguish his semantical view ... . This observation makes Baker's
purported resuscitation (1977) and Feinberg's alleged clarification (1965)
merely expoundings of what is already in Hart. ... Hart does not confuse
punishment for an action with the doing of an action here. He does not
confuse defeat of a putative action ascription, or excusing, with
mitigation, or reduction of its penal implications. He simply used an
unfortunate word, "reduction" (which he duly placed in quotes) ...
Careful reading shows that Hart had a clear understanding of the logical
character of defeasible reasoning. Hart explains of a "reduction" of a
claim, "that only a weaker claim can be sustained." ... Geach's criticism
is outrageous. ... Geach does not actually give one of [his] hundreds of
good examples, much less a "balanced diet." I have found it difficult to
supply the missing example, but perhaps times have become solemn, or at
least more litigious. Geach discusses "to fall" in a different context.
"To fall" seems so predominantly accidental that Geach could be right;
still, I have heard dancers and athletes both praised and excoriated for
falling. ... Here is a Geach fallacy. Hart is at best wondering whether
responsibilities (such as what might lead to the death penalty) are always
attached to ascriptions of acts (such as voluntary homicide), not whether
there might be responsibilities attached to involuntary events. Hart
gives examples where involuntariness defeats ascription of responsibility;
he does not require that involuntariness universally preclude
responsibility. ... Geach's use of the word "hence" is purely rhetorical,
since non-assertorial predications might still be explained in some other
way (e.g., Hare, 1970). ... The conflation of epistemology and aspects of
logical notation are Geach's aim. As more logical systems ascend to the
meta-language, fewer philosophers of logic adhere to such a naive
epistemology. The traditional logicians ask how to reconcile object-level
assertion and meta-language predication of defeasibility, but their own
answer (Tarski's truth schemata) is woefully stipulatory: "p" is true if
and only if p. ... I see nowhere in Hart that the ascription of
responsibility is entailed necessarily, not even in principal uses of an
action sentence. In all contemporary work on conversational implicature,
the implication is taken to be defeasible. Defeasibility is after all the
source of Hart's ascriptivism. Hart says only that "the connection is not
necessarily vinculum juris" ... . Pitcher should at least replace "and"
with "when". ... Hart should not consider these aspects to be errors ...
since the earlier paper neither excludes descriptive status to action
sentences nor assigns them as conclusions of law. It simply draws an
analogy to law and makes a claim about primary use. Hart simply declined
to "press the view." ... Hart is concerned with epistemology, hence, with
criteria, and with rules that express criteria in language. Mackie is
concerned with ontological questions ... . Feinberg is wrong to disparage
the analogy between quasi-legal contexts and the contexts in which
defeasible reasoning has found application, [i.e.,] contexts in
which conventions of conversational implicature are in force. ... If both
exhibit defeasibility, then Baker's point is useless. ... Why did he
never return to "defeasible concepts"? I believe Hart found that there
were more important things to do than to debate closed-minded logicians.
Hart decided to be a philosopher of law, a great philosopher of law,
rather than be a footnote beneath the footprint of deductivist dogma.
... To follow Hart, we should all aspire to elucidate great themes,
rather than to participate in small rows. I am guilty here of analyzing
text and nuance, which merely invites more analysis, more quibbling about
utterances: what was Hart's meaning? what was Hart's force? ... Every
time we formalize another aspect of representation and reasoning, we learn
how right Hart was. (ICAIL, 1995)
- See Toulmin, "The topic of exceptions or conditions of rebuttal has
been discussed by Prof. H. L. A. Hart under the title of `defeasibility'
." (Toulmin, 1958, p. 142; compare Toulmin, 1950, which cites only Ross).
Also, David Gauthier's dissertation submitted in 1961: "practical
principles are defeasible. (I take the term from Professor H. L. A.
Hart.)" (Gauthier, 1963, p. 159) There is universal agreement that
Hart's concept is in line with W. D. Ross's prima facie duty (Ross, 1930,
p. 18ff; Loewer and Belzer quote Nell's interpretation of Kant as
precursor to Ross; Causey gives Wittgenstein; Melden, 1959, p. 18 gives
Ewing, 1947, p. 33, and Frankena, 1952, p. 196, and 1955, p. 231 as
references to the use of prima facie, to which we must add Prior, 1949, p.
viii and Barry, 1965 (his 1958 dissertation supervised by Hart), pp.
32-4). Hart's defeasibility clearly augments Ross's prima facie duties.
By the time Chisholm uses the term in 1964, "These questions concern: the
`defeasibility' of moral requirements," (Chisholm, 1964a, p. 147) the
attribution of its original use seems no longer necessary. Chisholm says
that his colleague, John Ladd, "who was quite taken by Hart," had used the
term (personal communication), and Ladd cites Hart, Ross, and Chisholm
liberally in his early work. Ladd later adds, "I, at one time, held a
view close to that of Hart, but the study of Navaho ethics convinced me
that it was untenable [Ladd cites a section of the book that was
unprinted]." (Ladd, 1957, p. 462) Chisholm reviews Hart that same year,
"Similarly for those facts which would defeat the ascription of killing."
(Chisholm, 1964b, p. 614) Also Firth, "Their warrant is not derived from
coherence nor defeasible." (Firth, 1964, p. 552) My bibliography's
references to Swain, Sosa, and Lehrer-Paxson show how the term became
mainstream in epistemology in the mid-1970's, especially with Pollock
(although Pollock communicated to me that he came to defeasibility by
expounding Wittgenstein, as did Hart), and also main-stream in the
philosophy of practical reasoning (e.g., Searle, Nozick, Raz, and Audi).
From there it was imported to AI work on non-monotonic reasoning (see
Doyle, 1980, Nute, (1985) 1988, Loui, 1987, Causey, 1991; Causey cites
Belzer, who cites Nozick). (Part of first footnote from ICAIL 95)
- Alchourrón left us at this same point: a choice between two paths, neither of which he could ascertain to have a priori superiority.
Alchourrón could only hope that we would choose well. I however, am able to believe consistently that inquiry will dignify other conclusions. (Defeasible Deontic Logic, 1997)
- AI's models consider that agents use language in various ways, that
agents use and convey knowledge, that agents plan, search, focus, and
argue. Agents can choose their language, apply their knowledge, change
their plans, continue their search, shift their focus, and rebut another's
arguments. ... AI's model may be more helpful than game theory's models in
framing situations. AI offers a broader picture of the phenomenon of
negotiated agreement, and is descriptive of cooperative phases of
settlement. ... Protocols for negotiation are rarely so well defined that
responses to such locutions or rewards for such locutions are compulsory.
But the possibility exists in artificial societies to advance such
protocols, and conformity to strict social norms can be observed in many
negotiations. It may even be useful to characterize the degree of
cooperation achieved in many phases of negotiation by the severity of the
linguistic and social regulations that the parties observe. ... Agents
who are inventing sidepayments are employing default models of the other
party. ... As proposed solutions to the problems are offered and rejected, the
proposer's model must increasingly accomodate the other party as an
unusual person, as someone who is an exception to the default, as a person
who has a complaint in response to every reasonable accomodation. At some
point, the proposer simply quits accomodating. ... Acceptance of the
sidepayment is almost impossible, since it is probably irrelevant to the
actual preferences of the deceiving agent. The deceiver must ...
endure the problem-solving subdialogue, continuing to deceive, at the risk
of appearing particular, uncooperative, stubborn, stupid, or
recalcitrant. ... This restriction to subspaces is one example of a
larger class of search patterns that help to induce agreement for normal
agents who employ non-pathological search procedures. This is because
agents normally focus their personal deliberative search where the
dialogue has focused its attention. ... If both parties search to improve
their views in this area of likely agreement, the valley pushes upward as
utilities are revised. ... The effect on utilities is like the meeting
of landmasses, a geological process that can create mountains. ... The
agent might ... be pathetically ignorant; that ignorance might yet be
heterogeneous. ... Negotiators can enter and exit language games,
sometimes explicitly and at other times implicitly. ... Negotiation
includes more important phenomena: dialogue, argument, planning, thinking,
focusing, and reformulating. ... What should impress us is ... faithful
exhibition of a complex range of behaviors through the proper processing
of the important data. ... An AI model consists of a collection of
fables. Each fable has its own simplicity, and each is used ... in a
larger narrative. The aim is not to prove surprising theorems, but to fit
the pieces together ... .(Unpublished tech report, Dialogue and
Deliberation, 1997)
- Computation is whatever results from intentional and teleological rule-following upon symbols. ...
The existence of a program is the test for computation, not the existence of an algorithm. (SEP, 1998)
- Case-based reasoning is a postponement of induction. The advantage
of the raw cases, on this view, is that revision of the rule-base can
better be performed because the original cases remain available; they have
not been discarded in favor of the rules that summarize them. ...
Induction derives its power from the aggregation of cases, from the
attempt to represent what tends to make one case like or unlike
another. Case-based reasoning derives its power from the attempt to
represent what suffices to make one case like or unlike another.
[It] emphasizes the structural aspects of theory-formation, not the
statistical aspects of data. (MIT Encyclopedia of Cognitive
Science, 1998)
- There is no such thing as perfectly precise agreement. ... Thus, the
whole idea of "settlement" is a fiction ... . The most interesting thing
about imprecise settlement is that it is not just an agreement to exclude
rows and columns from the game. Agreement, even if it is vague, is a way
of passing control from the players (who have their threats) to the
institution (which has its traditions). Thus, a contract is negotiated:
of course it can be renegotiated at a later time when players' strategic
positions have changed; but a contract can also be enforced, by appeal to
institutions that recognize and respect the covenant produced in the
earlier strategic situation. (Unsubmitted draft, Designing a better
negotiation game, 1998)
-
Instead of viewing negotiation as a game with a solution, this paper
conceives of negotiation as a process. ... Fair-split effects are easily
modeled prior to the process of negotiation by transforming the payoff
functions. Here we are modeling the procedural effect of altering objects
of value based on how they are reached through negotiation, not just what
they represent to an agent in relation to other potential settlements.
... An expectation-driven negotiating agent is contemplated. The
insistence on probabilities that are objective and the estimation of a
probability of breakdown ... force the agent to make rational concessions,
so long as the expectation is falling due to empirically-based pessimism.
The compensatory force [is] the willingness to punish based on resentment
born of the procedural utility attached to unilateral breakdown on
noncooperative negotiating partners. ... The picture of negotiation
becomes a picture of path-selection among various laissez-faire paths, at
least for a pair of pessimism-punishment agents negotiating with each
other. The problem is not one-shot coordination within the set of
possible settlements, e.g., the identification of a unique equilibrium.
The problem is to identify and assess the acceptability of the path that
the process of negotiation is taking, and to decide whether it is worth
controlling that path with alternately motivated or alternately justified
action. (U of R Kyburg Probability as a Guide for Life Symposium,
2004, appearing in a book, 2006)
- Bureaucracy is more frightening than AI, and they are two of the same
kind. ... Mostly what makes AI is the careful embedding of a
computational device in the world so that its computations have meanings
to the rest of us. ... Worrying about self-conscious AI, in a networked
world where Paypal account theft ... is our real problem, is like worrying
that the sun is exhausting itself in a world where a ground-impact
asteroid tomorrow could make the explosion of the ancient island kingdom
Thera look like a happy day in Minoa. ... [Still,] I think AI is
shocking, and it provides us with knowledge of ourselves that must cause
us to rethink ourselves fundamentally. I think AI makes us look at our
rule-governed lives in a different way. It tells us what is precious
about us and what we should strive to preserve and protect. (ITEST
Part I, 2004)
- Physics also requires a physicist, to ascribe physical states and
identify physical processes; it, too, is a symbol-processing and
symbol-interpreting activity. ... Where are the processes that tend to
maximize entropy? I get to define entropy, so they are whereever I choose
to see them. Where are the frictionless surfaces? I get to choose my
idealizations. One simply has to be a bit less dogmatic about one's
metaphysical commitments in order to understand that physics does not
provide the fundamental ontology. One day there are leptons, the next
day, there aren't. One day there is mass, the next day there is force
divided by acceleration. ... To be nomological is really what we have in
mind when we say something is part of the physical environment. It's not
the physicist's nor chemist's ontology that's immutable; it's the
inescapable regularity in our experience of the world.
(ITEST Footnote, 2004)
- A two-register machine with 8-bit addition and one level of address
indirection is also a machine that can be simulated on a piece of paper.
There is nothing special about Turing's machine that makes it the only
disembodied, abstract, mathematical calculating device, the only computer
that does not need to be plugged into an electrical outlet. ... It is
the great discovery of the past decades that the electronic computer can
store large and complex programs, process data that has considerable
representational complexity, and proceed through a causal chain of clocked
events that permits computations to proceed inexorably from start to
finish with no human intervention and no mechanical human motivation. We
should not underestimate the importance of this discovery, as electronic
computers are far better than steam-driven computers might have been.
... One cannot pause the [analog] computation at an intermediate step,
sit down for an espresso, then resume the computation. ... There is
nothing in the definition of the c language that requires
subtraction to take about the same time as addition. ... Programmers are
specific only about the things they care about ... . The average
code-monkey sitting in a cubicle, being mocked by Dilbert and Dogbert, is
as likely to be specifying a protocol for the interaction of information
processors as to be writing imperatives for a commandable automaton. ...
A Chinese arithmetic savant who calculates the sums of increasing pairs of
numbers, and who does so compulsively, and who does not recognize the
results of the calculations to be sums, is a busy Chinaman indeed. ...
If ... we are all living on this planet in order to provide simulation
runs to an extraterrestrial intelligence that is using Earth as its
computer, then we are computing only insofar as someone can interpret what
it is that we are being used to compute. ... Programmers who abandon
their programs, but leave them in place to continue their interaction with
the physical environment, are ... special kinds of creators. There may be
a theological point to be made here. Meanwhile ... there may be a
question
... of why the mouse in the garden cannot provide the requisite
intentionality for symbol systems and computations. ... If I litter your
world with rule-governed artifacts that are so well situated that you find
it convenient to treat them in your reasoning about the world as if they
had agency, and if you feel that intentionality can arise wherever there
is an agent, then indeed, my programming has made a mess of your world.
(ITEST Part II, 2004)
- Computation arises from elective (intentional) and purposive
(teleological) rule-following on symbols. ... We, in fact, are the
machine. We are also the programmers, though some of us do a lot more
rule-making, and others of us have to do a lot of rule-following. ... At
root in this conception of distributed program and social machine is the
idea that game playing is a kind of computation. ... When they were
looking at the person who was looking back, they could sit down. Perhaps
this was a bit cruel, but it was very time and space efficient. No one
doubts that I programmed this asynchronous interaction, or that the result
was a partition on the class membership into pairs. ... Tournaments and
elections ... are strategies for conferring legitimacy on asymmetric
pronouncements by starting with symmetric opportunities and investing
players in the outcomes. ... Ex-post non-exchangeability is legitimized
by ex-ante exchangeability and the proper stake in a defensible
transforming process. ... The right games played the right way make a
right society. ... Society prescribes computations. ... A corporation is
easy to see as a socially emergent AI which leeches onto our social
computations, which is faceless, sphexish, incessant, emotionless, and
more than likely evil. ... Lots of [rule-governed actors] even have
initials like the IRS, FBI, INS, EPA, NIH, FJC, and EEOC. ... If you are
trying to hold on to land in the Delmar-Washington or Pershing-Skinker
area, Washington University might seem a lot like AGENT SMITH in the real
estate investment, land use and rezoning virtual reality game. If you sue
Washington University, you will again find yourself toe-to-toe with
someone who punches very hard in the virtual world that is called the
Missouri civil court system. ... I recently realized that I have spent
the past couple of years being mostly unhappy with our Dean, the position,
not our Dean, the man. ... Not all socially emergent AI is evil. ... It
used to be that the emperor's guard would arrive with a writ every few
generations asking for an ox or two. Maybe depression-era families had to
type a letter a couple of times a month. ... The proliferation of
automated social actors has become a lot more frightening than the robot
on the floor of the Honda plant in Ohio that paints minivan doors with
four coats of burgundy before passing it along for further assembly.
... I am not so worried that the automatic teller machine can replace the
human teller. I am worried that there at one time popped into existence a
job called "human teller" a large part of which was so easily replaced by
eighty lines of a computer program. ... The same people who created
organizations with information-processing roles for human automatons, the
bureaucrats and the bureaucracies of the fifties, are now unleashing upon
us actor upon actor of automated social role. Every job has a title,
every job has a description, every organization has a mission, every
office has automation, and every social AI has just enough standing to
engage you in an information-processing dialogue that will bring death to
your creative life with their hundreds of thousands of cuts. ... It seems
that an average of six F-words will gain access to the call-taker's
supervisor. ... We have to avoid using forms when a blank sheet of paper
is just as good. We have to invent words when we are being poetic. We
have to go places where the computer cannot go. We have to be inefficient
when we don't believe in the process. ... Not every potential Neo is The
One. Some people are just spoon-bending pretenders. ... Frankly, people
programmed to be social AI should learn when they should get out of our
way. ... Looking at it another way, why should we be talking about
licensing professional programmers or holding the authors of AI
consultative programs liable for errors when we are not holding
legislators, bureaucrats, ministaires, and fonctionnaires to any
standard? ... The really sad realization is that Microsoft Windows
doesn't suck because Microsoft programmers are inept. It is a mess
because of some people's faulty business plan [that] was imposed on
Microsoft's otherwise capable software production army. ... [My dog]
doesn't know about ebay and easements, Anglo-American queueing norms, or
even marriage certificates. ... The world she occupies is the real world,
with dirt and apples, and some lower-case a's that smell differently from
other lower-case a's. Our world unfortunately encroaches on her world
when we put cars on roads with signal lights that she can't interpret. I
think that's a crime ... .
(ITEST Part III, 2004)
- Legal theory formation is as likely to occur in day-to-day reasoning
during the management of conflicting heuristics, the interpretation of
discourse, the interpretation of open-textured predicates that appear in
explicit, linguistically-specified rules, the interplay of principles and
rules, and the formulation of coherent bureaucratic standards. ... Legal
theory formation is almost certainly not what lawyers normally do,
and may only be a rough parody of what judges, legislators, and solicitors
do. Scientific theory formation is almost certainly not what
scientists do. ... The improvement of the scientific
reasoning model (e.g., Kyburg) pays close attention to error and
predictiveness, where probability is the main tool, while the improvement
of the legal reasoning model pays close attention to improper
over-generalization and improper under-generalization, where defeasible
argument is the controlling idea. In law, the strict attitude toward
error is made possible by defeasibility. In science, the quantification
of error (and predictiveness) is made possible by probability. ... It has
long been suggested that the more the defeasibility, the less simple or
less cogent the overall set of rules. ... I propose that a legal theory
is more cogent than another if its rules [align better] with broader social
principles. ... Scientific theory formation gives primacy to the rules,
and uses cases to identify optimal rules. Rules would be simple, few, and
aesthetically pleasing if not for the empirical evidence, which causes
asymmetries, non-linearities, and hidden variables. ... Optimal fit is a
heuristic for uncovering that reality. Legal theory formation takes cases
(actual or hypothetical) to be deontically primary, and induces rules only
as a practical compilation of edict. It does so for the purposes of
prescriptive clarity, structural discovery, or linguistic brevity. Rules
are as numerous and as complex as they need to be, in order to depict the
principles that justifiably differentiate circumstances. ... Modern
machine learning attempts to erase all distinctions between various kinds
of induction from data, as if all learning from cases can be expressed as
the selection of features followed by the determination of an appropriate
set of separating hyperplanes in that feature space. ... Few things could
be more historically or intellectually insulting. ... The intellectual
vacancy we see in machine learning is the direct result of an
unwillingness to study the phenomena first, before doing the mathematical
modeling. (CMSRA, 2005, denied journal publication in J.
Applied Logic with its peer conference papers -- can you guess the background of the objecting referee?)
- I have been approached by many people wanting to discuss the Toulmin
diagram, by scholars from every continent, for dozens of purposes, whether
they use the diagram formally or informally, whether they are
technologists or Luddites, whether they know a lot about Stephen Toulmin
or nothing at all. ... After a handful of colossal ideas: paradigm
shifts and methods, fuzzy logic, illocutionary force, the
analytic-synthetic distinction, supervenience and deductive-nomological
explanation; Toulmin diagrams must be mentioned next. This has to be
impressive and surprising to any intellectual historian. ... Toulmin
provided a lone outpost of resistance, a single place where the fire
burned during a long winter, where dialectical travelers of the logical
landscape could stop for a rest. ... It is a method [that] we can all find
slightly comical in its simplicity. ... In time, argumentation should
overcome even fuzzy predication and fuzzy connectives as the most
important nondeductive development in the history of logic.
(Argumentation, 2005, appearing in a book, 2006)
-
A genius is someone whose persona is tied to acts of genius, who does the
act naturally, frequently, incessantly, inevitably, involuntarily, as mere
personal expression. ... There is no genius without struggle. Geniuses
must be creative and brilliant, but they must also be a bit rebellious.
We find more geniuses among the latter-borns, the homosexuals and other
procreationally frustrated, maternally unconsummated and paternally
unperfected, the battered children, the immigrants and minorities, the
excommunicated, unwashed and unaccepted, the disenfranchised, and the
nationally uprising. ... If you want to honor Alan Turing's genius, don't
look down the list of Turing Award winners and adhere to their paradigms.
Don't even aspire to win the Turing Award yourself. Alan Turing wouldn't
have wanted a Turing Award. If you want to honor Turing's genius, write a
clever new program. Commit an act of genius. ... Genius expresses itself
in multiple arts; it is not confined to narrow talents. The savant is the
narrow virtuoso. The genius is a natural polymath. ... The geniuses we
know were able to chart a course between the enmity of their
contemporaries and the vulcanism of their spirit. My advice for the young
genius: be true to yourself, but also learn how and when to keep your
mouth shut. Otherwise society will try to crush you, because genius and
society are natural adversaries. (unpublished essay on Genius, 2005)