artificial intelligensia and the search for meaning khurshid ahmad, chair of computer science...

Artificial Intelligensia and the Search for Meaning

Khurshid Ahmad,Chair of Computer Science

Trinity College, Dublin, IRELAND14th June 2013

Knowledge and Meaning

The key challenge in artificial intelligence, and indeed across computing, is to be able to understand the subjective use of language and other semiotic systems in areas of business, governance, heritage, science and technology.

Subjective use of language has been studied under various umbrellas: knowledge acquisition in decision support, sentiment analysis in finance, digital heritage, requirements analysis.

In computing a rational approach to problem solving, exemplified in formal methods, for example, has had to encounter the same problem as in other areas like compliance in governance, behaviour of financial markets, uncertainty in medical decision making and others.


My search for meaning is a random walk through various areas of human enterprise:

1. decision making in life critical services (water); 2. assessment of affective contents of texts;

3. indexing of large collections of images (art collections, biological cells, forensics).

This I have achieved through being involved in building systems for:

a. information extraction systems (terms, ontology)b. image classification systems based on neural netsforecasting systems based on fuzzy aggregation


My search for meaning is a random walk through various areas of human enterprise with the help of a large number of colleagues including Carl Vogel, Arthur Hughes, Tim Fernando and Rozenn Dahyot of SCSS; Dermot Kelleher (now at Imperial College) and Tony Davies of the Trinity Medical School; Colm Kearney (now at Monash Uni) and Brian Lucey of the Business School, and Andrea Zemankova (Slovak Academy of Sciences).

My colleagues, Daniel Isemann, Dara Javaherian, Stephen Kelly, Xiubo Zhang and Aaron Gerow, have been helpful with their intellectual skills and programming knowledge.


Language is prolific: in a space of 500 years we have moved from hot metal presses to a world of instant communication as in digital librarires, blogs, twitter..

The subjective use of language, incorporating hedging and qualifiers, metaphors and similes, creativity and oxymorons, selective use of information, censoring and advocacy, helps and hinders.

Language expands to incorporate new realities, alternative behaviours, new explanations, exorcising myths and fears, …

Language is highly parsimonious and ambiguous, when compared to other semes


Knowledge about and of persons, places, things, and events is disseminated in language on a consensual basis.

This consensus is reflected in the preferential usage of words in a language, preferential usage of visual signs in images/pictures/scenes.

The consensus is recorded in the archives of a specialist domain

The archive has to be randomly sampled to find statistically significant usage of key symbols.

These symbols are then used to generate higher level knowledge atoms: interpretation of texts, images/scenes, time series.

Knowledge and Meaning: The number seme

Human beings have developed complex systems of communications and it is the interaction between these systems that appears to help in the creation, storage, usage and deletion of knowledge.

Number systems have been involving over many millennia: The Nilitic measurments of land and water, The Hindu numerals (with a ZERO), rational numbers, fractional numbers, imaginary and transcendental numbers. Then we had ‘ordered’ collections of numbers – time series invented in Japanese rice markets (candlesticks) in the 14th century, and now econometrics.

The description of numbers in words and pictures is key to an understanding of our world;

Knowledge and Meaning: The number seme

The description of numbers in words and pictures is key to an understanding of our world. APPLE Inc., share trading on NYSE together with the volatility in prices. Japanese candlestick patterns show when the opening price was less than closing price (RED) and vice versa (GREEN). TEN minute trading sequences shown for June 2013;

Knowledge and Meaning: Visual Semes

N.Holmes, C.Spence (2005)Multisensory Integration: Space, Time and SuperadditivityCurrent Biology, Volume 15, Issue 18, Pages R762-R764

SMASH! Your head twists aroundat speed, your eyes catch splinters of porcelain skidding across the floor, and you see a furry black tail disappearing under the cupboard. Your cat just spent one of its lives, and cost you your favourite vase. For the rapid orienting behaviour that allowed you to catch a glimpse of the culprit red-pawed, you can thank your superior colliculi.

Interaction in visual processing

Dirk B. Walther, Eamon Caddigan, and Li Fei-Fei, and Diane M. Beck (2009) Natural Scene Categories Revealed in Distributed Patterns ofActivity in the Human Brain. The Journal of Neuroscience, August 26, 2009 • Vol. 29(34) pp 10573–10581

Natural Scene Categories Revealed in Distributed Patterns ofActivity in the Human Brain –within and across the visual system

Human subjects are extremely efficient at categorizing natural scenes, despite the fact that different classes of natural scenes often sharesimilar image statistics. (Walther et al 2009:10573)

Interaction in visual processing

Dirk B. Walther, Eamon Caddigan, and Li Fei-Fei, and Diane M. Beck (2009) Natural Scene Categories Revealed in Distributed Patterns ofActivity in the Human Brain. The Journal of Neuroscience, August 26, 2009 • Vol. 29(34) pp 10573–10581

Natural Scene Categories Revealed in Distributed Patterns ofActivity in the Human Brain


http://en.wikipedia.org/wiki/Philosophy_of_language

What do we do with the various semes (linguistic, visual, tactile)?

How do the semes relate to the mind and/or brain

What is the nature of meaning?

How are a sequence of semes (words, shapes/ colours/textures) organised into meaningful whole (sentences, pictures)?

What are the meanings of the parts of sentences, pictures, sculptures?

Knowledge and Meaning:A multi-sensory world

Our brain, and perhaps that of a collective of people, generates its own projections of a given ‘reality’: accentuating, (censuring, ignoring) what we (do not) want to see/hear/know

Multisensory Processing

Emergence of knowledge and modes of

communication


AgencementProblems in finance and business are amongst the hardest problems to be solved on computer systems:

Economic actors can be viewed as nodes in a socio-technical network or agencements – a network comprising human beings, computer systems, including algorithms, heuristics, communication devices. Hardie, Iain and Mackenzie, Donald. (2007) Assembling an Economic Actor: The Agencement of a Hedge Fund,” Sociological Review Vol 55 (No. 1), pp 57-80.

Teubert, Wolfgang (2003). Writing, hermenutics and corpus linguistics. Logos and Language Vol.IV (no. 2) pp 1-17.

Words, Works and Worlds: Hermeneutics and LSP

Interpertament

Texts are responses to previous texts and the texts are then responded to in turn and the cycle continues


Continual regulatory change of financial institutions is the order of the day and will be so for some time to come.

Post 2008 regulatory change has resulted in a deluge of legislation, procedures, compliance schemes, from a host of organisations.

All change is accompanied by a change in ontology – what there is – and language is a key device to rub out the old and bring in the new.

We have been building tools for dealing with ontological change in a range of disciplines, from anthropology to environmental engineering, and from bacteriology and immunology to nuclear physics and nano technology. And, now finance

Language for General Purposes

Language for Special Purposes

Terminology

Knowledge and Meaning:Terminology, slang and jargon, special language, a

‘slither’ of general language

Knowledge and Meaning:Terminology, slang and jargon, special language, a

‘slither’ of general language

Language for General Purposes(Exploratory work carried out by

computer scientists)

Language for Special Purposes(Experimental Ontologies, exploring logic and formal

reasoning in science)

Terminology(Medical

Ontologies)


Rocksteady: A scaling affect analysis system that can be used, in conjunction with quantitative information, like equity/commodity prices, polling results, can be used to analyse and forecast changes in the prices/polls.

The systems acquires text through RSS feeds, collates a corpus and time stamps documents, and then carries out text analysis.

It uses an ontologically organised dictionary that comprises affect on the one hand and domain specific terms on the other.

The results can be aggregated at different time scales.


Project Slándáil aims to build and test a prototype system managing disaster emergencies by fusing information available in different modalities in social media with due regard to ethical and factual data provenance. The prototype will cover three major EU languages: English comprehensively, and German and Italian to a limited extent..


The Ùraigh Project strategy is to design a software system which can store and process big data (c. 10 TB per week) coming out of a noisy data stream. Work involves cell biologists, grid computing, statistics.


dark liquidity pools

Secretive actions conducted by large brokerages to make transactions outside of the markets so that the trades will not adversely affect the price of large sections of stock. Institutional investors commonly use dark liquidity pools as a method of concealing their investors or investments. Because the prices of these transactions are concealed, investors are typically able to receive a lower price.

http://www.investorwords.com/84/action.html

http://www.investorwords.com/585/brokerage.html

http://www.investorwords.com/5046/transaction.html

http://www.investorwords.com/2962/market.html

http://www.investorwords.com/5014/trade.html

http://www.investorwords.com/3807/price.html

http://www.businessdictionary.com/definition/section.html

http://www.investorwords.com/4725/stock.html

http://www.investorwords.com/2504/institutional_investor.html

http://www.investorwords.com/2837/liquidity.html

http://www.investorwords.com/3735/pool.html

http://www.businessdictionary.com/definition/method.html

http://www.investorwords.com/2630/investor.html

http://www.investorwords.com/2599/investment.html

http://www.businessdictionary.com/definition/able.html

http://www.businessdictionary.com/definition/receive.html

The emergence of knowledge

action The price movement and volume of a stock or overall market.

brokerage Used interchangeably with broker when referring to a firm rather than an individual. also called brokerage house or brokerage firm.

transactions (a) An agreement between a buyer and a seller to exchange an asset for payment.(b) In accounting, any event or condition recorded in the book of accounts.

market A public place where buyers and sellers make transactions, directly or via intermediaries. Also sometimes means the stock market.

trades (a) A transaction of a security or commodity.(b) The exchange of products and/or services without the use of money. also called barter.

price Cost, usually expressed in monetary terms.

http://www.investorwords.com/3807/price.html

http://www.investorwords.com/3149/movement.html

http://www.investorwords.com/5258/volume.html

http://www.investorwords.com/4725/stock.html

http://www.investorwords.com/2962/market.html

http://www.investorwords.com/584/broker.html

http://www.investorwords.com/1967/firm.html

http://www.businessdictionary.com/definition/individual.html

http://www.investorwords.com/588/brokerage_house.html

http://www.investorwords.com/587/brokerage_firm.html

http://www.businessdictionary.com/definition/agreement.html

http://www.businessdictionary.com/definition/buyer.html

http://www.businessdictionary.com/definition/seller.html

http://www.investorwords.com/1797/exchange.html

http://www.investorwords.com/273/asset.html

http://www.investorwords.com/3634/payment.html

http://www.investorwords.com/48/accounting.html

http://www.businessdictionary.com/definition/event.html

http://www.investorwords.com/6456/condition.html

http://www.investorwords.com/42/account.html

http://www.investorwords.com/3930/public.html

http://www.businessdictionary.com/definition/buyer.html

http://www.businessdictionary.com/definition/seller.html


http://www.businessdictionary.com/definition/via.html

http://www.investorwords.com/2556/intermediary.html

http://www.businessdictionary.com/definition/mean.html

http://www.investorwords.com/4743/stock_market.html


http://www.investorwords.com/4446/security.html

http://www.investorwords.com/975/commodity.html

http://www.investorwords.com/1797/exchange.html

http://www.investorwords.com/3874/product.html

http://www.investorwords.com/6664/service.html

http://www.investorwords.com/3100/money.html

http://www.investorwords.com/426/barter.html

http://www.investorwords.com/1148/cost.html

http://www.investorwords.com/3095/monetary.html

http://www.investorwords.com/4950/terms.html

The emergence of knowledge

Pre 19th Century

MATTER Gas Liquid Solids

Accessible through senses

Post-Modern Era

MATTERMolecule Atom

Atom Electron+NucleusNucleus

Protons+Neutrons+Mesons (Japanese

Electrons) +500 ‘Elementary Particles’

Accessible indirectly through sense: light spectrometers, electron microscopes, ‘mass’ spectrometers, particle accelerators (atom smashing machines)

1960’s MATTERElementary Particles

Leptons +Quarks

1970’s Leptons (e.g. electrons) Accessible indirectly

1980’s Quarks PERMANENTLY CONFINED: My particle is in a jail forever!!

The emergence of knowledge: Fact and

FictionMurray Gell-Mann found that the eightfold way could really best be explained by a particle, undiscovered as yet, that had three parts (hadrons), each holding a fraction of a charge. He called them "quarks" with a nod to James Joyce, whose novel Finnegan's Wake contains the passage: "Three quarks for Muster Mark!"

Fractional charge seemed an outrageous suggestion at first, but proof came for his theoretical quarks in 1974.

Knowledge and Meaning: Fact and Fiction

"Three quarks for Muster Mark!"

Fractional charge seemed an outrageous suggestion at first, but proof came for his theoretical quarks in 1974.

The European Council for Nuclear Research is currently spending 3,320 Million Swiss Francs ( 1.32 CH FR= 1 US $) for building the Large Hadron Collider to look for the elusive, the permanently confined quarks. The European Investment Bank is investing Euro 300 Million in the ‘enterprise’

We spoke and wrote slowly- c. 1930

We spoke and wrote slowly- c. 2008?

July 2008: The positive emotion accompanying sub-prime mortgage

http://upload.wikimedia.org/wikipedia/commons/thumb/d/d5/Subprime_Mortgage_Offer.jpeg/300px-Subprime_Mortgage_Offer.jpeg


http://bioportal.bioontology.org/ontologies/42280/?p=terms&conceptid=10025650http://bioportal.bioontology.org/ontologies/42280/?p=terms&conceptid=10025650

Knowledge and Meaning: Roots of Rationalism

Wittgenstein, Ludwig (1922/1971). Tractatus Logico-Philosophicus. London: Routledge & Kegan Paul Ltd.

What can be said at all can be said clearly; and whereof one cannot speak thereof one must be silent.

Ideal Language1.1 One name for every simple

1.1.1 A name is simple symbol in the sense that it has no parts which are

symbols themselves1.1.2 Nothing which is not simple will have

a simple symbol1.2 Never the same name for two simples1.3 The symbol for the whole will be “complex” containing the symbol for the parts1.4 What is complex in the world is a fact.

Our ever changing world – Do theories change and, if so, why?

The history of any discipline shows major changes in the discipline over a period of time. The underpinning theories in a discipline appear to change as well. In physics, we have moved from an indivisible atom (c. 1900) to a divisible atom (c. 1920) comprising elementary particles (protons and neutrons, c. 1935 ). The elementary particles, it turns out, are in themselves comprise quarks (c. 1970’s) ……

There are two major theories of this change: First, new theories appear through a process of iterative refinement – a gradual process. Second, theories appear when suddenly anomalies in existing theories are discovered and are discarded.

Our ever changing world – Do theories change and, if so, why? Iterative

RefinementKarl Popper tried to build a purely deductive approach to science [and econometrics]. For Popper ‘all scientific discussions start with a problem (P1), to which we offer some sort of tentative solution – a tentative theory (TT); this theory is then criticized, in an attempt at error elimination (EE); and as in the case of dialectic, this process renews itself: the theory and its critical revision to new problems (P2)’ (Redman 1994:69).

P1 TT EE P2

It is possible, suggested Karl Popper, that science could start anywhere.

Popper has influenced the development of econometrics.

Redman, Deborah, A. (1994). Karl Popper’s Theory of Science and Econometrics: The Rise and Decline of Social Engineering. Journal of Economic Issues. Vol 28 (No. 1)., pp 67-99

Austrian born philosopher and logician; held chair at LSE. Born 1902, died 1994.

Paradigm Shifts: What is paradigm shift anyway?

A research paradigm (Kuhn 1970) was defined originally by Kuhn to 'suggest that some accepted example of actual scientific practice - examples which include law, theory, application and instrumentation together - provide models from which spring particular coherent traditions of scientific research' (1970: 10).

KUHN, T. S.(1970).The Structure of Scientific Revolutions. Chicago: Chicago Univ. Press.

American philosopher and sociologist of science; held chairs at Princeton and MIT. Born 1922, died 1996

Our ever changing world – Do theories change and, if so, why? Paradigm Shifts


Theories are refined incrementally and in some instances there is a paradigm shift of revolutionary proportions.

Knowledge and Meaning:Language and other

disciplines

Bertrand Russell’s Introduction to Wittgenstein, Ludwig (1922/1971). Tractatus Logico-Philosophicus. London: Routledge & Kegan Paul Ltd.


disciplines

Ahmad, Khurshid. (2013). MEANING AND ONTLOGICAL COMMITMENT: A SURVEY OF THE USE OF THE TERM ‘SEMANTIC PRIMITIVE’ (In preparation).

Knowledge and Meaning:Limits of Rationality

Bechtel, William. (1988). Philosophy of Mind - An Overview for Cognitive Science. Hillsdale (NJ): Lawrence Erlbaum Associates


disciplines

Bechtel, William. (1988). Philosophy of Mind - An Overview for Cognitive Science. Hillsdale (NJ): Lawrence Erlbaum Associates

Language and Meaning

What you see/hear/touch is what you get? Perception of sound/orthography & cognition of language!

• Language can be viewed as 'a communicative process based on knowledge. Generally when humans use language, the producer and comprehender are processing information, making use of their knowledge of the language and of the topics of conversation. Language is a process of communication between intelligent active processors, in which both the producer and the comprehender(s) perform complex cognitive tasks.

Winograd, Terry. (1983). Language as a Cognitive Process. Wokingham: Addison-Wesley Inc.,



Producer

Current Goals

Cognitive Processing

Knowledge Base

Knowledge of the language

Knowledge of the

situation

Knowledge of the world

Comprehender

Understood Meaning

Cognitive Processing

Knowledge Base

Knowledge of the language

Knowledge of the

situation

Knowledge of the world

Medium

Speech

or

Writing



Phonlogical Rules

Morphological Rules

Dictionary (items)

Grammar Rules

Dictionary Definitions

Semantic Rules

Deductive Rules

Inferential Rules

Phonological

Morphological

Lexical

Syntactic (Parsing)

Semantic

Reasoning

Sounds

Phonemes

Morphemes

Words

Syntactic Structures

Representation Structures

Stored Knowledge

Processes Assigned Structures



Information Asymmetry in Human Transactions: Financial Trading

Information invariably comprises an element of surprise.

Information flows when there is an asymmetry – the writer knows more than the reader/listener.

The providers of information can present information that blends facts and opinion.

It is important to understand the attitude of the writer and how the attitude is expressed.

Information Asymmetry: Noise Traders and Informed

Traders

TradersNoise

Pessimistic Optimistic

Informed Pessimistic Herding Short-sell

Optimistic Buy Herding

Assume that there are two kinds of traders only in a market: informed traders and noise traders. The noise trader fails to ascertain the true value of an asset and relies on guesswork, heuristics, imitation of the informed trader, or prayer. The noise trader misprices and the informed trader should see this as an opportunity to create a margin through arbitrage. This arbitrage is not always possible and worse still the informed tries to follow the noise trader.

Behaviour and Financial Markets

Experimental economists have reported mixed results on rationality: people are often better (e.g. in two-person anonymous interactions), in agreement with (e.g. in flow supply and demand markets), or worse (e.g. in asset trading), in achieving gains for themselves and others than is predicted by rational analysis.

Patterns in these contradictions and confirmations provide important clues to the implicit rules or norms that people may follow, and can motivate new theoretical hypotheses for examination in both the field and the laboratory.

http://nobelprize.org/nobel_prizes/economics/laureates/2002/smith-lecture.pdf

Affect, Emotion and Mood

Differentiation of Meaning and Affect Dimensions and Scales

AFFECT CATEGORY SCALE/VALUESPutting a qualitative value on entities

Positive/Negative

Showing how the depth or shallowness of affect

Strong/Weak

Indicating the persistence of affect

Active/Passive

Implying a moral judgement Virtue/Vice

Sentiment Analysis

Sentiment analysis: a systematic, computer-based analysis of written text and speech excerpts, for determining the attitude of the author or speaker respectively in relation to a specific topic.

Khurshid Ahmad (2011) (Ed.) Affective Computing and Sentiment Analysis: Metaphor, Ontology, Affect and Terminology. Heidelberg: Springer.

Sentiment Analysis

Sentiment analysis systems were used first by political scientists to understand the orientation of political parties – through an analysis of party manifestos

Scholars in economics and finance have used opinion columns in financial newspapers to estimate the impact of opinions and opinion makers on prices and traded volumes of equities

Khurshid Ahmad (2011) (Ed.) Affective Computing and Sentiment Analysis: Metaphor, Ontology, Affect and Terminology. Heidelberg: Springer.

Behaviour and Commodity Markets

• Oil is an important commodity and oil fields and refineries are not where the consumers are.

• Changes in oil price is critical to everybody

The oil markets are chaotic places

Oil traders receive news and rumours from a variety of sources.

2240 news items on oil 2.2 Million words


134 affect categories including sentiment, orientation, activity;

Selectable inputs (news sources, social media)

Automatic report generation

Real-time results

Self-learning lexicon

Drill-down capability at every level

RockSteady An affective computing system developed and tested at Trinity College, Dublin






March 2011 – Arab spring and Euro market crises






Oil companies are powerful and important

Behaviour and Politics

• Irish Election Results: The repetition of names of persons, places, ideas and events, is used sometimes for emphasizing the importance of what a name represents. Similarly, the repetition of sentiment-bearing words and phrases is used by writers and speakers to articulate their feelings and attitudes. During an election campaign, this repetition may have a bearing on the electability of politicians and on the reputation of political parties. Using Rocksteady, we analyzed a sample of the news published by Irish media between 27th Dec. 2010 and 21st Feb. 2011. We had analysed 3,024 news items, comprising 1.86 million words, published within Ireland.

• We looked at 11 national and 30 regional news papers and TV stations

Behaviour and Politics

User-configurable dictionaries

Corpus linguistic methods for the automatic extraction of a candidate terminology of a specialist domain of knowledge.

Collocation analysis of the candidate terms leads to some insight into the ontological commitment of the domain community or collective. Examine use of metaphors

The candidate terminology and ontology can be easily verified and validated and subsequently may be used in the construction of information extraction systems and of knowledge-based systems

User-configurable dictionaries

Otherwise ‘crude oil’ will be NEGATIVE kind of oil;

And a HEDGE FUND will be something to do with sustaining countryside

Regulatory Change & The Text Deluge

Patent being applied for the CiCui System

Regulatory Change & The Text Deluge


Content providers are building systems for dealing with the deluge of laws, regulations, and guidelines:

Thomson-Reuters Accelus Compliance Manager and Wolter Kulwer’s Compliance Resource Network are systems that provide access to the shifting mass of regulation, often with multi-jurisdictional requirements. FINRA’s guidance to the finance industry on the use of website and social networking websites (2010), for promotion of goods and services, refers to 18 or so rules and regulations issued by a number of different organisations (including NASD, SEA, NYSE, and FINRA); any changes to these rules will have to be incorporated in compliance procedures related to the websites.

Ontology Data Bases –


Lexical AnalysisStatistical Analysis * Disambiguation

Validation via April O’Neal and Experts

Ontology PopulationTerminology Alignment * Hierarchies

* Part-Whole Relations

Semantic Analysis

Collocation * Colligation

Keyw

ord

Lis

ts

On

tolo

gy E

ncod

ing

Candidate Taxonomy/ Candidate Ontology

Terminology and Ontology Extraction System (CiCui) being tested with input from a major news vendors on a continuous loop;

Identification of key events and people in text; disambiguation, risk modelling.

Text categorisation

Lexical AnalysisStatistical Analysis * Disambiguation

Validation via April O’Neal and Experts

Ontology PopulationTerminology Alignment * Hierarchies * Part-Whole

Relations

Semantic AnalysisCollocation * Colligation

Ontology Data Bases – Candidates for deep ontology CiCui System

Keyw

ord

L

ists

On

tolo

gy

En

cod

ing


Candidate Taxonomy/ Candidate Ontology

artificial intelligensia and the search for meaning khurshid ahmad, chair of computer science...

Documents

subjective use of language

knowledge acquisition

programming knowledge

areas of business

semiotic systems

building systems

higher level knowledge

meaningmy search