Illinois Genetic Algorithms LaboratoryDepartment of General EngineeringUniversity of Illinois at Urbana-ChampaignUrbana, IL 61801.
Searle, Intentionality, and theFuture of Classifier Systems
David E. GoldbergIllinois Genetic Algorithms Laboratory
University of Illinois at Urbana-ChampaignUrbana, IL 61801
1980 v. Now
Remember thinking how cool LCSswere.
Just apply them to gas pipelinesand voila, all AI problems ofWestern Civilization would besolved.
Started to ask John for examplesof successful application.
Found out that I was in themiddle of an interesting idea, nota working computer program.
John H. Holland (b. 1929)
Are we happy with LCSs? What’s Searle got to do with it. Revisiting the Chinese room. Art Burkes had it right. Designing a conscious computer. Searlean program for LCSs:
– Computational consciousness not impossible.– From consciousness to intentionality.– Intentionality and beyond.
What are we missing? What should we do?
Are We Happy With LCSs?
Have made progress:– Increasingly competent, solve hard problems
quickly reliably and accurately.– Principled manner.
But don’t seem very intelligent:– Do what we tell them.– Not autonomous in any serious sense.– Our discussions are largely technical.– Are we focused on right problems?
What’s Searle Got to Do With It?
Mill Prof of Philosophy of Berkeley. Philosopher of language and mind. Early work took off from Austin’s work on
speech acts. Searle is Darth Vader of artificial
intelligence. Notorious Chinese Room argument has
always puzzled me. In many ways, Searle is high philosophical
priest of emergence. Rejects dualism & materialism. Couldn’t understand how he could miss
possibility of more than mere systacticaltranslation.
John R. Searle (b. 1932)
The Chinese Room Argument
Strong AI is not possible on a computer. Monolingual English speaker in a room with
– Chinese writing (story)– 2nd Chinese symbols (questions)– Instructions in English for relating first set of symbols
to second.– 3rd set of Chinese symbols (answers)
English speaker does not understand Chinese evenif answers are indistinguishable from those ofChinese speaker.
Cracks in the Chinese Room
Mind, Language & Society,Basic Books, 1998, p. 53.
“When I say that the brainis a biological organ andconsciousness a biologicalprocess, I do not, of course,say or imply that it wouldbe impossible to produce anartificial brain out ofnonbiological materials.”
“The heart is also a biological organ, and thepumping of blood a biological process, but it ispossible to build an artificial heart that pumpsblood. There is no reason, in principle, why wecould not similarly make an artificial brain thatcauses consciousness.”
Searle was complaining about direct approach tointelligence.
Without consciousness and intentionality therecannot be intelligence.
How do we create an intelligent, conscious being?
Arthur Burks Had Interesting Take
Robots and FreeMinds, University ofMichigan, 1986.
“Tonight I willadvocate the thesis: AFINITEDETERMINISTICAUTOMATON CANPERFORM ALLNATURAL HUMANFUNCTIONS.”
Chapter 5: Evolution and Intentionality
“The course of biological evolution from cellsto Homo sapiens has been a gradualdevelopment of intentional systems fromdirect-response systems.”
“The [intentional] system contains a model ofits present status in relation to its goal andregularly updates that model on the basis ofthe information it receives. Finally, it decideswhat to do after consulting a strategy, whichhas value assessments attached in to variousalternative courses of action.”
CS-1 Had Bio/Psycho Roots
CS-1 had reservoirs forhunger and thirst (Holland& Reitman, 1978).
Schemata processorspaper had reservoirs, too(Holland, 1971).
CS-1 worked in mazerunning task.
But design was Lockean. Tabula rasa for everything
except rule firing,apportionment of credit,and rule discovery.
Is this enough? Thesis: Can’t take shortcut
around consciousness andintentionality.
So You Want a Conscious Computer
What does this mean? Consciousness is complex, emergent
phenomenon. How can we design it? Don’t throw pieces together and hope for
the best. My experience: Emergent phenomena
emerge when (a) key elements are presentand (b) system tuned properly.
Consider more Searle.
Shooting for C Not Crazy
Shooting for GA competence was crazy. Have accomplished. How:
– Considered essential elements.– Built qual/quant theories of how they worked.– Designed until limits of performance achieved.
Can do the same forconsciousness/intentionality!!
Searle’s Greatest Hits
Mind as biological phenomenon. Function of consciousness. Features of consciousness. How the mind works: Intentionality. The good stuff comes from intentionality:
Language & other institutional fact. What are we missing?
Mind as Biology
Consciousness is the primary feature ofminds.
3 features of consciousness:– Inner: in body and in sequence of events.– Qualitative: certain way they feel.– Subjective: first person ontology (does not
preclude objective epistemology).
Enormous variety of consciousness: smell arose, worry about income taxes, suddenrage about driver, etc.
Functions of Consciousness
What does it do? What is survival value? What doesn’t it do for our species? Consciousness is central to our survival. All actions a result of conscious thought
followed by action.
Consciousness, Intentionality, & Causation
Represent world, and act on representations. Intentional causation: Not billiard ball causation. Not all consciousness intentionally causal, but much
is. Should be best understood; are we not in touch
with it always? Descartes’s error. Yet difficult to describe: Can describe objects,
moods, thoughts, but not C itself. Problems:
– Not itself an object of observation (consciousnessobserves but is not observed).
– Tradition of separating mind/body: dualism.
Features of C
1. Ontological subjectivity.2. C comes in unified form. Thinking and feeling go on
at same time in same field of C: Vertical & horizontal.3. C connects us to world (Tie to intentionality).4. C states come in moods.5. Always structured.6. Varying degrees of attention.7. C is situated.8. Varying degrees of familiarity.9. Refer to other things10. Always pleasurable or unpleasurable
How the Mind Works: Intentionality
Primary evolutionary role of C is to relateus to environment.
Cannot eliminate intentionality of mind byappealing to language; alreadyintentionality of the mind.
Searle: Urge to reduce it to something elseis faulty.
DEG: As designers we need to reduce it tosomething and then find conditions ofemergence among those things.
Intentionality as Biology
Thirst, hunger as basic, causing desire todrink or eat.
Once this granted, camel nose under thetent, intentions based on other sensory.
Isn’t reality “confirmed” by our “success” inachieving intentional goals over and overagain.
Structure of Intentional States
Intentionality as way mental states are directed atobjects & states of affairs.
Can be directed at things that don’t exist? How can this be? Distinguish between type of intentional state and
content. Content: rain; Types: hope, believe, fear rain. Structural features:
– Direction of fit– Conditions of satisfaction
Direction of Fit
Term: from Austin, foreshadowed by Wittgenstein,examples Anscombe.
Anscombe’s lists:– Shopping list: Beer, butter, bacon. Husband matches
world to list.– Detective’s list: Follows shopper, “beer, butter,
bacon,” matches list to world.
Not all intentional states like this: e.g. when youare sorry, assume match between mind and world.Intentional state is null.
Conditions of Satisfaction
Beliefs can be true or false. Goals can be achieved or not. Easier to understand in terms of speech acts. Have 5 illocutionary points or types:
– Assertive: commit to the truth.– Directive: direct hearer to do something.– Commissive: speaker promises to do something.– Expressive: speaker expresses opinion about state of
the world.– Declarations: speaker creates something with
Intend to move body body moves:Example of intentional causation.
Differs from billiard ball or Humean causation. Self-referential: intend to move body, body moves
because I intended then intentional causation. Critical to distinguishing natural versus social
sciences. Intentional explanations not deterministic: Could
have done otherwise gap is free will.
Good Stuff from Intentionality
Searle goes on to talk about language andinstitutional facts (money, college degrees,etc.).
Disappointment with LCS is it can’t get tothe good stuff.
Can’t do language. Can’t form contracts. Can’t create new institutional fact.
Construction of Social Reality
Need to clarify observer-independent &observer-dependent features of the world.
Need 3 new elements:– Collective intentionality.– Assignment of function.– Constitutive rules
Observer Independent v. Dependent
Many features of the world independent ofour observations of them: observerindependence.
Many observer dependent: Something acharacteristic because of observerjudgment, but not relative to others.
OI vs. OD more important than mind-body. DEG aside: Isn’t it dualism in the back door
Need the notion of “we intend together.” Attempts to reduce to individual intention are
complex. Existence of biological organisms with collective
intentionality suggests CI is a primitive. DEG aside: Are social insects intentional in Searlean
sense? Could be that social affiliation is primitive,certain behaviors hard wired. Then, CI results from(a) naming the group, (b) attributing intention to it(as-if intentionality), and (c) treating the as-if asreal.
Assignment of Function
Use of objects as tools:– Monkey uses stick to get banana.– Man sits on rock.
Physical existence facilitates function, butfunction is observer relative.
All function assignment is observer relative.
How to distinguish between brute facts andinstitutional facts.
Types of rules:– Some rules regulate: “Drive on right side of road.”– Some rules regulate and constitute: Rules of chess
both regulate conduct of game and create it.
Constitutive rules have form: X counts as Y in C. “Move two and over one” counts as a knight’s
move in Chess.”
Simple Model of Construction of Social Reality
Strong thesis: All institutional reality explained by 3things:– Collective intentionality.– Assigned function wall keeps people out
physically, but low fence or boundary marker keepspeople out by convention.
– Constitutive rules. Money example: Evolution from valuable
commodity to fiat currency. Institutional reality powerful: X counts as Y in C can
be iterated and stacked forming powerful networkof institutional facts.
What Are We Missing?
Do not have C-machines. Searle’s 10:
1. Ontological subjectivity.2. C comes in unified form.3. C connects us to world.4. C states come in moods.5. Always structured.6. Varying degrees of attention.7. C is situated.8. Varying degrees of familiarity.9. Refer to other things10. Always pleasurable or unpleasurable
Can argue that we have vertical unity inmessage board.
Do not have horizontal unity. My first proposal recommended
modifications to permit time series. Modifications to rules. Modification to the boards.
Moods & Pleasant/Unpleasant Missing
This is big. Emotions are “engagement with the world”
(Solomon). Necessary for judgment and values. Don’t want a simulation. Emotions:
– Physiological component– Judgmental component
Other Things Missing
Attention Gestalt structure Situatedness & familiarity Refer to other things (may have this)
What Should We Do?
Stuff we’ve gotten right: Sensors, association,models (anticipation), learning
Can’t continue to work on same thing. No serious architectural changes proposed to LCS.
– Emotions: As judgments, source of values, andarbiter of attention.
– Multiple boards: As source of difference andsimilarity. Main hope of quality of consciousness &unity.
– Center of intention rooted in “biological needs.”
How Do We Break This Down?
Tough problem. If C is complex building block,
what are minimal essentialelements to achieve.
How do we know we’ve achievedit (first person ontology, thirdperson epistomology)?
Sets of tests and experiments. What theory needed to set
parameters of C? Not unlike approach that cracked
Summary & Conclusions
Have accomplished quite a bit in classifiersystems.
Many of our questions are technical. Deeper questions about whether we’re
attacking the right questions. Need consciousness and intention to get
the “good stuff” of intelligent behavior. Wrestling with Searle’s categories not a bad
place to start.