Friday, October 31, 2008

australia's digital counter-revolution

From "digital revolution" to counter revolution in a mere ten months. Senator Stephen Conroy, Minister for Broadband, Communications and the Digital Economy is threatening to censor the internet of illegal material in Australia at ISP level

Here is Mark Newton's argument against, which leads to the conclusion that Conroy's proposal will lead to increased child abuse:
The online community's argument is a simple one:
  • there's no problem to solve because actual illegal material on the Internet is so rare that nobody ever finds it;
  • even if there was a problem to solve, there's no serious public demand to solve it;
  • even if there was a public demand to solve it, none of the solutions proposed by the ALP will be effective, and the Government has handily provided original research to decimate their own case;
  • even if they were effective, they'll slow down Internet access and reduce Internet reliability, as shown by the same original research released by the Minister on July 22;
  • even if the proposed solutions had perfect performance and reliability, none of them are affordable;
  • even if they were affordable, they'll be implemented terribly by the same underclass of bureaucrat that deemed Mohammad Haneef a terrorist, or Bill Henson a pornographer. The salivating of hangers-on like Family First and Nick Xenophon, lobbying to have the blacklist expanded before it's even in force, demonstrate perfectly how open the system will be to political manipulation and lobbying;
  • even if they were implemented perfectly by perfect administrators, the blacklists will inevitably leak, be published on the Internet, whereupon they'll fall into the hands of nefarious individuals and consequently enable child abuse all over the world, with the direct assistance of the Commonwealth of Australia; and
  • there's no possibility that the blacklists won't leak. Finland's list has already leaked, CyberPatrol's encrypted blacklist is cracked every six months or so. It's delusional to believe that Australia will be any better at securing its officially sanctioned list of Child Porn and Terrorism sites than anyone else. It might take a month, a year, five years, ten years, or two hours. But it will leak, secrets always do. Pressing it into service will be like setting a ticking time bomb, and when it explodes there'll be a thronging multitude of critics pointing at Senator Conroy and saying, "I told you so, you were warned, but you did it anyway".
This isn't a complicated argument. To justify the ALP's policy, cogent, successful arguments against each and every one of those independent points will need to be mounted.
- The perplexing internet debate by mark newton

Further links with brief commentary:

Filtering out the fury: how government tried to gag web censor critics
Documents how Conroy's office tried to silence Newton by directly contacting his employer, Internode

Closed Environment Testing of ISP-Level Internet Content Filtering: Report to the Minister for Broadband, Communications and the Digital Economy, June 2008 (pdf, 89pp)
I've only read the executive summary, which does claim that filtering technology has improved significantly since 2005 - less degradation, more filtering accuracy and more filtering options for ISP customers. I don't really believe this but it might explain why Conroy went ahead.

But amazingly this report goes onto say:
"ACMA (Australian Communications and Media Authority) was not asked, as part of the trial, to assess the capability of ISP-level filtering technologies that filter only illegal content. ACMA was also not asked to investigate the balance of costs and benefits associated with implementing ISP level filtering ..."
This is weird. Conroy is talking about filtering illegal material but did not even ask the ACMA to investigate this!

ISP-level content filtering won't work
"The leaders of three of Australia's largest internet service providers — Telstra Media's Justin Milne, iiNet's Michael Malone and Internode's Simon Hackett have, in video interviews with ZDNet.com.au over the past few months detailed technical, legal and ethical reasons why ISP-level filtering won't work."
Dale Clapperton, Electronic Frontiers Australia (video interview, some dumb questions -"Do you have children Dale?" but Dale rises above it. There was another question about how children might circumvent the free government software available at the family level. Dale then turned that around to point out that yes, that will happen at the ISP level, too.

Stop Australian Internet Censorship Petition. Points out that smaller ISPs will be driven out of business by the extra cost

The high price of internet filtering by Michael Meloni
  • 10,000 out of every one million innocent sites will be blocked, figures based on the government report cited above. This will lead to loss of income to businesses. What will be the appeal process?
  • ISPs will need more call centre staff to deal with angry customers
  • There are other more serious threats to which resources could be directed - cyber bullying, identity theft, online predators

Shelby Steele on Saturday Extra

This Saturday 7:30am
Radio National

Geraldine Doogue:
However, I'll be taking that up with one of the acknowledged luminaries in the field, Professor Shelby Steele, a leading African-American intellectual. Noel Pearson wrote a terrific review in The Monthly earlier this year of Steele's provocative book: "A Bound man: Why We're Excited About Barack Obama And Why He Can't Win". Steele's is a dense, fascinating analysis, which defies easy summary. Essentially, he (like Pearson) sees insidious dangers in individuals identifying as constant victims---Afro-Americans and Aboriginal people victimised by white colonial misdeeds---and seems to believe there is no easy, linear route to acceptance. We have been pursuing the very busy Professor Steele for many, many weeks and are absolutely thrilled that he's agreed to be interviewed the Saturday before the Presidential vote. At this stage, I don't know whether he abides by that verdict of his. Steele is a conservative by nature but I will be keen to test whether over the course of this gruelling campaign, which has of course included the financial meltdown, his views have shifted. If anyone has a particular question they'd like posed, please let me know by Friday. All perspectives, especially of those who've lived in the midst of the American melting-pot, most appreciated.
http://blogs.abc.net.au/saturdayextra/2008/10/almost-at-the-f.html
Noel Pearson's article about Steele's analysis of Obama, from May, is now online

update 1st November: If you missed the Radio National program or live overseas then it is available as an mp3 at the Saturday Extra Latest Programs page

I'm still too busy to comment properly on this but Steele's contention that Obama has yet to clearly define himself, either personally or politically, seems correct to me

Sunday, October 26, 2008

Basic SVG tutorial


SVG file not being displayed in your browser


I've created a webpage of basic SVG shapes and written some introductory tutorials (task 3) on the WHS xo wiki SVG page.

On the webpage, Chrome displays the stroke-array feature, which produces a dashed line, and a second conventional line, whereas Firefox does not. See my previous post for a brief discussion about SVG support

Reference:
Simple SVG Display: How to display SVG files in web pages

SVG Essentials by J. David Eisenberg O'Reilly, 2002
Main basis for the tutorials that I've written

Friday, October 24, 2008

SVG support

SVG = Scalable Vector Graphics

Some of my students complained about the ordinariness of the xo home screen. I pointed them to the Modifying Sugar page in the Sugar Manuals which suggests various ways to customise Sugar. In turn, this led me into teaching the class about SVG and how to make their own icons.

The modern browsers are pushing SVG forward and leaving MS and IE behind. Amazingly, MS and IE still offer ZERO support for SVG, despite it being a W3C recommendation since 2003. Tim Berners-Lee has recently diplomatically criticised MS for being behind in supporting scalable vector graphics

Adobes support for their SVG viewer will be discontinued after January 2009 but that doesn't matter because the modern browsers like Opera, Chrome, Firefox and browser plugins such as Batik continue to improve their support for SVG.

Interview, with Ted Gould:
JMZ: Do you think it possible that SVG maintains a presence as a web standard without any serious commerical sponsorship?

TG: I don't know that a single major corporate sponsor is required. I think that the "X factor" here is the different browser vendors. They've already pushed to make HTML 5 happen, I think they could go after SVG next. This would give them good in-browser implementations for things like animations that can only really be done with Adobe's Flash plugin today. I don't think that they like the idea of having the future of the web being implemented as a plugin.
Here is an excellent, informative chart (and commentary), from Jeff Schiller, of the results of running various SVG implementations (web browsers and browser plugins) through the official SVG Test Suite:



Reference:

SVG Essentials by J. David Eisenberg O'Reilly, 2002
I've had this book for some time. It explains the basics behind SVG and how to write XML directly to produce vector graphics.

Making SVG Icons for Sugar
As well as providing some good tips this OLPC page recognises that:
Unfortunately, no suitable vector editing tools are available as activities for the XO laptops at present. We acknowledge this omission, and hope to provide a vector-based "Draw" activity (to supplement the bitmap-based "Paint" activity) in order to provide the proper tools for icon generation on the laptops themselves in the future.
Inkscape: An Open Source vector graphics editor. I found that by saving with the Plain SVG option you do obtain XML similar to what you would get by handcrafting a simple icon

Chrome, Googles' new browser, is partly based on Web Kit (which receives high ratings from Jeff Schiller), so Chrome may overtake Firefox in the SVG implementation race. At any rate, I downloaded Chrome hoping it would support SVG animations but alas it did not. Neither does Firefox.

Batik looks great but I haven't played with it yet.

Thursday, October 23, 2008

breaking ranks

Labour back bencher Julia Irwin has broken ranks and spoken out against the phony Rudd / Gilliard "education revolution"

See Govt MP slams 'education revolution'

What she is saying is that the Rudd government has not addressed one central issue of disadvantage in Australian education - the disparity between wealthy Private Schools and struggling government schools, which creates the long tail of underachievement in australian schools. This disparity was exacerbated in the Howard years and nothing is being done about it by the new government.

An education revolution requires more than a photo-shoot with a laptop

Related:
maths education crisis in australia

curriculum reform will not improve education without quality teachers

Thursday, October 16, 2008

Infidel by Ayaan Hirsi Ali


Infidel by Ayaan Hirsi Ali

I've recently read this book. It made me think, cry and sometimes laugh. Her story is incredible, very well written and powerful.

It helped me understand Islam, the Middle East and Africa since Ayaan lived in and graphically describes Somalia, Saudi Arabia, Ethiopia and Kenya. She was raised as a Muslim and suffered many beatings (at one point her skull was fractured by her Quran instructor and she nearly died), genital mutilation, civil war (in Somalia) and an arranged marriage.

Her relationships with her extended family and clan are described in detail. The suffering of her mother and sister in particular are heart breaking. For me, the saddest part of the book was what happened to her sister, who eventually descended into madness. Nothing is spared in description as, after years of doubt and agonising, Ayaan finally rips off the burqa (and also the hijab) and becomes a liberated woman. She is a great writer of personal, intimate prose and she has an enthralling story to tell. One aspect of this book is you obtain the full sense of the journey from Ayaan's grandmother, a desert nomad, through all the painful stages to the emergence of a modern and liberated Ayaan.

She describes the rise of the Muslim Brotherhood as a reaction against widespread government corruption and the propagation of a fundamental, literal interpretation of the Quran

After escaping to the Netherlands (running away from the forced marriage), Ayaan completed her education in politics and history and worked as an interpreter. In that role she witnessed more horror in the way Muslim women refugees were treated by men. She witnessed and participated as the public debate broke out in Holland about tolerance and integration of Islamic refugees. In the aftermath of 9/11 and the murder of Pim Fortuyn she participated more in the public discussion, received death threats, ran for parliament and was elected. She subsequently helped Theo van Gogh make a film about Islamic culture (Submission). Because of this film van Gogh was murdered and a note addressed to Ayaan was stabbed into his body.

You can read this book as an amazing personal story, or, for one perspective of the role of Islam in the world today, or, to deepen your understanding of what happens to some people as they grow up in the countries mentioned above

Of course, the book has been criticised. It might be true that she doesn't distinguish clearly between moderate and fundamentalist or literal Islam. What she does is track the resurgence of Islamic fundamentalism and what that means for her lived experience, which applies to many others. You get the very strong impression that to live as a Muslim woman in certain countries does mean in practice following certain fundamental principles which enslave women psychologically, physically and socially - eg. when in Holland she campaigned for the number of honour killings in in that country to be recorded in the data base. And when it was, as a trial, everyone was shocked at how many there were.

So, how to distinguish between the lived experience of many, many women in a time of resurgence of fundamentalism and the counter point that there is some distinction between fundamental and "modern Islam"? I don't know.

Others would criticise Ayaan for her "right wing" affiliations. She currently works for the American Enterprise Institute as a Resident Fellow. I think the evolution of her political views is well described in the book. At one stage she was a researcher for the Labour Party in Holland but eventually left them because of their too tolerant stance towards the abuse of Islamic women. If the so-called left doesn't take a stand against honour killings, genital mutilation and the like then should we blame her for joining the right? Not in my view.

One of the reviewers says:
"Hirsi Ali has invited her critics to walk a mile in her shoes. Most wouldn't last a hundred yards" (Aminatta Forna, Evening Standard)

Thursday, October 09, 2008

minsky 9: the self

Overview of Chapter 9 of The Emotion Machine (summary, online draft, buy)

The Single Self idea keeps us from wasting time about difficult questions about our mind

How does vision work? "Your Self simply peers out through your eyes"
How does memory work? "Your Self knows how to recollect what might be relevant"

Sometimes physicists strive to construct a single model or a grand unified theory. Nevertheless, physics contains many different subjects and each has its own (useful) way of describing the world. Whenever a subject becomes important to us we tend to build multiple models. This diversity is a principle source of our resourcefulness

We each make multiple models of ourselves

Our subpersonalities will frequently need to compete for control of higher level processes

William James (1890) on recalling childhood:
"... that child is a foreign creature with which our present self is no more identified in feeling than it is with some stranger's living child today ..."
Daniel Dennett (1991) on Self:
"... like spider webs, our tales are spun by us; our human consciousness, and our narrative selfhood, is their product, not their source ... their effect on any audience ... is to encourage them to posit a unified agent whose words they are, about whom they are ... a 'centre of narrative gravity'"
Instead of asking about our Identity it is better to ask, "Which of my models of myself best serves my present purposes?"

Personality traits

We describe people as having character traits - disciplined, honest, attentive, friendly

Possible causes for personal traits:
Inborn, genetic -
Learned -
Investment principle - hard to displace tried and trusted methods that work
Archetypes and Self-Ideals - our cultural heroes and villains
Self control - to keep ourselves from constantly changing our goals and priorities

Nevertheless, the concept of traits can be treacherous, for example, the generalities of astrology influence many

Self Control

To achieve long range goals you need self control. But self control is hard. We use tricks to achieve self control, we threaten or bribe ourselves, "I'll be ashamed if I give in to this", "I'll be proud if I can accomplish this"

Why must we use devious tricks to control or Ways to Think, instead of just choosing to do what you want to do?

Directness would be too dangerous (see Chapter 3) We would probably die if one part of our mind could take over the rest. In emergencies our instincts need to take over

Many of us spend much of our lives seeking ways to make our minds behave

Dumbbell ideas and dispositions

People like two part distinctions. Minksy provides many examples, both of traits (eg. solitary v. sociable) and of alleged characteristics of right and left brain thinking (eg. rational v. intuitive). Many things seem to come in opposing pairs.

Minsky thinks two part distinctions are too simple:
"...it usually makes little sense to commit ourselves, for all future times, about which objects to like and dislike - or about which persons, places, goals or beliefs we should seek to avoid, or accept or reject - because all such decisions should also depend on contexts ....most dumbbell distinctions ... appear to be so simple and clear that they seem to be all that you need - and that tempts you to stop. Yet most of the novel ideas in this book came from finding that two parts are rarely enough - and eventually my rule became: when thinking about psychology, one should never start with less than three different parts or hypotheses!"

Why do we like the idea of a self?

What leads us to the strange idea that our thoughts cannot just proceed by themselves, but need something else to control them? We use words like "Me" and "I" to keep us from thinking about what we are!

Various ways in which the Single-Self concept is useful to us:
Localised body is consistent with Single-Self
Private mind - the idea (illusion?) that only you hold the keys to the strong closed box of your private mind
Explaining our minds - if we can think "I perceive the things that I see" then it keeps us from wasting time on questions about perception that we don't know the answer to
Moral responsibility - to justify our laws and moral codes we assume that Selves are responsible for intentional deeds
Centralised economy / Decisiveness - "Thats enough thinking, I've made my decision!"
Causal attribution - we like to attribute causes
Attention and focus - we often think we have a single stream of consciousness to which we attend
Social relations - others think of themselves as Single Selves, we might be seen as weird if we didn't play along!

Our minds are messy. We spend large parts of our lives tidying them up - selecting, suppressing, refining

What is pleasure and why do we like it?

Emotions are hard to describe because they seem hard to split into parts. Hence there seems to be nothing to use as pieces of explanation

Minsky argues that pleasure is a suitcase word for quite a few different processes:
  • Satisfaction - achieving an ambition
  • Exploration - a quest, the pleasure is not only at the end
  • Goal suppression - critics and other goals suppressed
  • Relief - if the goal was the elimination of an irritation
Pleasure and satisfaction refer to extensive networks of processes we don't yet understand. We tend to treat complex, hard to grasp things as single and indivisible.

The pleasure of exploration

Adventurousness is an antidote for exploring unfamiliar terrain, which can lead to pain and distress. Learning by small incremental positively rewarded steps is limited

When we are learning a new technique, we need to work harder with fewer rewards, while enduring the additional stress of being confused and disoriented. We may have to abandon older techniques which have served us well. We may arouse a sense of loss or grief and a temptation to quit. Such learners have trained themselves to enjoy discomfort

Exploring the contradiction of enjoying discomfort: Pleasure is not a basic all or nothing thing just as the Self is not a single thing. Some parts of the mind may be uncomfortable but other parts enjoy forcing those first parts to work for them: "Good, this is a chance to experience awkwardness and to discover new kinds of mistakes"

We can envisage pleasure as negative, in the way it can suppress competing goals

What makes feelings so hard to describe?

The alleged mysteries of "subjective experience" or "directness of experience" arises from the inability of our higher level processes to detect all the intermediate steps involved in these experiences, eg. touching, redness. Some philosophers (dualists) conclude that materialist explanations of such things is impossible. Minsky argues that they have not worked hard enough to imagine adequate models of those processes.

How important is "privileged access" to our own mind? Sometimes our self assessments are inept, our friends may have better ideas of our real state

The sense of having an experience

Perception or Sensations are not "basic" (see Chapter 5). More signals flow down to the sensory cortex than in the opposite direction, presumably to help us see what we expect to see. We frequently "see" things that do not exist, eg. this square:

How is a human mind organised?

Each normal child eventually learns to:
  • recognise, represent and reflect upon some of his own internal states
  • self reflect on some of his intentions and feeling
  • identify with aspects of how others behave
Summary of the kinds of structures would support these developments:

1) Deal with various situations by activating certain sets of resources, which are different Ways to Think
2) How would we determine which resources to select?
(a) for simple situations use use If--> Do rules
(b) for more versatile situations use Critic --> Selector schemes

3) The adult mind develops multiple levels of functioning and each level contains Critics and Selectors (See Chapters 5,6,7)

4) Various Ways to Think might also have levels of different symbolic expressiveness (See Chapter 8)
5) Our model needs to have room for answers to questions we haven't thought of asking yet. Envision the mind as a decentralised cloud of yet unimagined processes, interacting in still unspecified ways

Central and Peripheral Controls

We have various "alarmers" which interrupt higher level processes. Sometimes our thinking processes "break down". Some examples:
  • trouble recalling past events
  • trouble solving an urgent problem
  • cannot decide which action to take
  • lost track of what you were trying to do
  • a surprise happens
But we are capable of rapid recovery

Mental Bugs and Parasites

Examples of mental parasites include self reproducing sets of ideas (memes) which can displace competing ideas - doctrines, philosophies, faiths, beliefs

The dignity of complexity

Our brains have evolved in a process that has taken 30 million centuries

Some of our sources of human resourcefulness come from three vastly different time scales:
  • Genetic endowment
  • Cultural heritage
  • Individual experience
Most of our commonsense knowledge may be embodied as metaphors in the form of panalogies (See Chapter 6)

We have multiple descriptions of things - and can quickly switch among them
We make memory records of what we've done - so that later we can reflect on them
Whenever one of our Ways to Think fails, we can switch to another
We split hard problems into smaller parts, and keep track of them with our context stacks
We manage to control our minds with all sorts of bribes, incentives and threats

Our minds have bugs! For example, our powerful imagination can lead us to set out on extensive but futile quests!

Tuesday, October 07, 2008

minsky 8: resourcefulness

Overview of Chapter 8 of The Emotion Machine (summary, online draft, buy)

Before Alan Turing we did not know about a single machine that could emulate other machines. Turing opened the door to developing a machine which can have multiple Ways to Think

We have multiple ways of estimating distances - remembering typical sizes, overlaps, context, binocular vision, perceived speed. Each method is imperfect but taken together we can usually avoid serious mistakes. We effortlessly switch between methods choosing the more appropriate one for each different situation. This might also apply to how we think.

Panalogy (a word invented by Minsky meaning parallel analogy, also discussed in Chapter 6)

Panalogies are corresponding features of different meanings which are connected to the same parts of one larger structure. Minsky suggests that our brain architecture has evolved structures which make it easy to link knowledge fragments in this way.

Examples:
  • Whenever you think about your Self, you are reflecting about a panalogy of mental models of yourself
  • Sight is intertwined with memory, we fill in huge chunks from memory when "seeing" (without realising)
  • We rarely make entirely new ideas, instead we modify existing ideas
This creates both speed (swapping between multiple meanings) and also the potential for confusion and ambiguity. It might explain the importance of metaphor and analogy in our thinking. Hence ambiguity becomes a virtue and not a fault because much of our human resourcefulness comes from using analogies that result from this.

How do people learn so rapidly?

Sometimes we learn new tricks from a single exposure (whereas a dog may require hundreds of lessons). A difference engine could be converted into a copying machine, so the structure in long term memory becomes the same as the one in short term memory


Minsky thinks our minds are like computers in this respect. Short term memory is expensive and limited.
"... a blow to the head can cause a person to lose all memory of what happened before and including that accident ... transfer to long term memory may take a day or more and require sleep"
Other reasons why long term memories may require much time and processing:
Retrieval - it may have to be linked to an existing panalogy, otherwise how could it be retrieved?
Credit Assignment - to be useful it would need to be linked to other relevant panalogies
Real Estate problem - finding a place for new memories would not be simple (might involve destruction)
Copying complex descriptions - hard to think of plausible schemes for making complex, linked memories

Learning involved many varied skills, such as:
  • Adding new If --> Do --> Then rules
  • Changing low level connections
  • Making new subgoals for goals
  • Choosing better search techniques
  • Changing high level descriptions
  • Making new Suppressors and Censors
  • Making new Selectors and Critics
  • Linking older fragments of knowledge
  • Making new kinds of analogies
  • Making new models and virtual worlds

Credit Assignment

Behaviourism is limited. Learning complex things cannot be explained by reinforcement or if-do rules

Minsky speculates that we might use higher level processes to decide what to learn from each incident by reflecting on our recent thoughts. These processes could be used to make such "credit assignments":
  • choosing how to represent a situation will affect which future ones will seem similar
  • learn only the parts of your thinking that helped, and forgot those which were irrelevant
  • connect each new fragment of knowledge so that you can access it when it is relevant
The quality of our credit assignments might account for our "intelligence" (a suitcase word). The section about Poincare's unconscious processes (7-7) pointed out that this might take days. There is an incubation period.

We need more research about what kind of credit assignments infants can make, how children develop better techniques, how long such processes persist and the extent to which we can control them

Transfer of learning to other realms:

Some children seem to transfer their learning to other realms, while others don't

Transfer of learning might be superior for those who make better credit assignments. To gain more from each experience, it would not be wise for us to remember too many details but only those aspects that were relevant to our goals. Also what we learn from an experience might be more profound if we assign credit to the earlier choices we made that selected our winning strategy

Creativity and Genius

Genius consists of unusual combinations of otherwise common ingredients:
  • genetics
  • fortunate mental accidents
  • learn how to praise self internally
  • intense positive attention from parents
  • isolation from other children
  • mental management
  • enduring discomfort when replacing a Way to Think
  • selecting which new idea to develop

Memories and Representations

Minsky defines representation to mean any structure inside one's brain that one can use to answer some questions

What distinguishes us from other animals? It is our ability to treat ideas as though they were things, our ability to conceptualise. Minsky argues that there must be representation structures (networks) inside our brains. Knowledge fragments don't have meanings unless they are linked. He discusses various possible ways to represent knowledge:
  • Describing events as stories or scripts
  • Describing structures with semantic networks
  • Using trans-frames to represent actions
  • Using frames to embody commonsense knowledge
  • Learning by building "knowledge Lines"

Connectionist and Statistical Representations

He contrasts two different ways to represent an apple, through a semantic (symbolic) network and a connectionist network


Connectionist networks (based on numbers showing strength of associations) can learn to recognise many important types of patterns, without any need for a person to program them. But number based networks have limitations. Every relationship is reduced to a number or strength so there remains almost no trace of the evidence that led to it, eg. the number 12 could represent all sorts of things

Minsky:
I see the popularity (of Connectionist Networks). in recent years, as having retarded the search for higher level ideas about human psychological machinery... research on commonsense thinking kept advancing until about 1980, but then it was clearly recognised that further progress would need ways to acquire and organise millions of fragments of commonsense knowledge. That prospect seemed so daunting that most researchers decided to try, instead, to invent machines that could learn, by themselves, all the knowledge that they would need - in short, to invent new kinds of "baby machines" ...

Quite a few of these learning machines did indeed learn to do some useful things, but none of them went on to develop higher-level reflective Ways to Think - and I suspect that this was mainly because they tried to represent knowledge in numerical terms....

... I do not mean to suggest that such networks are not important ... it seems safe to assume that many of the low level processes in our brains must use some form of Connectionist Networks (pp. 289-91)
How do we learn new representations?

Kant 1787: ... experience and sensory knowledge is only part of knowledge ... cognition adds new knowledge

Minsky thinks we are born with primitive forms of structures like K-lines, Frames and Semantic Networks which are then built on to create representations

Which representations to use for which purposes?

A dialogue between different approaches to the best way to represent knowledge:

Mathematician: It is always best to express things with logic
Connectionist: No, logic is far too inflexible to represent commonsense knowledge. Instead, you ought to use Connectionist Networks
Linguist: No, because Connectionist Nets are even more rigid. They represent things in numerical ways that are hard to convert to useful abstractions. Instead, why not simply use everyday language - with its unrivaled expressiveness
Conceptualist: No, language is much too ambiguous. You should use Semantic Networks instead - in which ideas get connected by definite concepts!
Statistician: Those linkages are too definite and don't express the uncertainties we face, so you need to use probabilities
Mathematician: All such informal schemes are so unconstrained that they can be self contradictory. Only logic can ensure us against those circular inconsistencies

minsky 7: thinking

Overview of Chapter 7 of The Emotion Machine (summary, online draft, buy)

Our ability to think in different ways (new Ways to Think) distinguishes us from other animals. However, we rarely ask good questions about what thinking is or what chooses which subjects we think about. Thinking often just happens smoothly.

How do we explain things such as long term plans, reminding ourselves of things to do, choosing among conflicting goals and whether to quit or persist?

The Critic-Selector Model of Mind

Minsky discusses various possibilities about how to resolve conflicts when more than one Critic-Selector is aroused and compete for resources.

Central Problems for Human Psychology

Small group in depth studies (eg. Piaget) are more valuable than large group statistical studies to the development of psychology. In large group studies small but vital details are overlooked.

The research that Minsky thinks needs to be done:
  • What are the principal Problem Types that our mental Critics recognise?
  • What are the major Ways to Think that or mental Selectors engage?
  • How are our brains organised to manage all those processes?
What are Some Useful Ways to Think?

Knowing How
Searching Extensively

Reasoning by Analogy
Dividing and Conquering
Reformulating
Planning

First solve a different problem ->
Simplifying
Elevating
Changing the subject

Reflective ->
Wishful thinking
Self reflection
Impersonation

Others ->
Logical contradiction
Logical reasoning
External representation
Imagination

Social ->
Cry for help
Ask for help

Last resort ->
Resignation

What Are Some Useful Types of Critics?

These mirror the Levels of Mental Activities, from Chapter 5

Innate Reactions and Built-in Alarms - some alarms are hard to ignore, eg. a babies cry
Learned Reactive Critics - eg. moving to a quieter environment
Deliberative Critics - thinking about what went wrong
Reflective Critics - diagnosticians which verify progress or suggest alternatives
Self-Reflective Critics - various forms of self criticism
Self-Conscious Critics - these affect one's image of oneself, eg. I'm losing track of what I am doing (Confusion)

How Do We Learn New Selectors and Critics?

We can improve our Ways to Think by creating higher level Selectors and Critics that help to reduce the size of the searches we make. Most "theories of learning" do not address this

Poincare's Unconscious Processes

Minsky includes several great quotes from mathematician Henri Poincare (1913) about his unconscious learning process. These stages are described:
  • Preparation -
  • Incubation -
  • Revelation -
  • Evaluation -
Incubation and Revelation occur without our being aware of them. We don't really understand how they work.

How do we organise and change our collection of Critics and Ways to Think? We don't know. These issues should be recognised as central to the development of psychology.

Do We Normally Think "Bipolarly"?

Common sense thinking may consist of a brief "micro-manic" phase producing a few ideas, followed by a brief "micro-depressive" phase looking for flaws - all taking place so quickly the reflective systems don't notice it

Monday, October 06, 2008

minsky 6: common sense

Overview of Chapter 6 of The Emotion Machine (summary, online draft, buy)

Comparing what computers can do (play chess) with what humans can do which computers can't yet do (make a bed, read a book or babysit) provides us with insights about humans. Computer programs don't have commonsense knowledge, eg. when someone says, "a package is tied up with string", this includes "obvious" facts about the nature of string and packages (eg. with string you can pull but not push a thing)

Computer programs are not self aware of their goals - whether they are achieved or at what quality or cost. Computer programs are not as resourceful as humans, when "stuck", eg. they don't reason by analogies

What Do We Mean by Common Sense?

Minsky provides an extensive account of the common sense knowledge involved in answering a phone call. We aren't normally aware of how much we know

He introduces a new word, panalogy, which means parallel analogy

"Charles gave Joan the book"
  • Physical Realm - book moves from Charles to Joan
  • Social Realm - is Charles generous or hoping to ingratiate himself?
  • Dominion Realm - Joan now controls the book

Here we have three meanings of the word "give". Our brains may structurally connect analogous items of knowledge from different realms (points of view). This might explain how we can easily switch without even being conscious of it between these different meanings

Multiple meanings are sometimes seen as a defect because of their ambiguities. The panalogy concept reframes them as a strength

It is hard to categorise commonsense knowledge. The scheme favoured by Minsky is along the lines of the kinds of thinking that can be applied to the categories:
  • Positive expertise
  • Negative expertise
  • Debugging skills - knowing alternatives when usual methods fail
  • Adaptive skills - how to adapt old knowledge to new situations
Could we build a "baby-machine", a machine that will gradually learn more by itself? Such programs have been tried but have failed to develop good new ways to represent knowledge. A machine will fail to learn the right things from most of its experiences. Minsky argues that learning requires selectivity and appropriate "credit assignments". You cannot learn things that you can't represent

Building intelligent machines becomes stuck due to not having ways to overcome problems like:
The Optimisation Paradox: difficult to improve once you work well
The Investment Principle: reliance on existing processes makes it hard to develop alternatives
The Complexity Barrier: changing complex systems has unexpected side effects

At any rate, good new ways to represent knowledge are usually not quickly and widely adopted:
  • you need new skills to work with them efficiently
  • such skills take time and performance will probably worsen during the changeover period
Evolution is more about rejecting bad changes than selecting beneficial changes. Most species evolve to occupy narrow, specialised niches. Evolution can learn to avoid common mistakes but usually not uncommon mistakes - except by evolving language systems.

We first need to evolve ways to protect against changes that cause bad side effects. Excellent method here is to split system into parts that can evolve more independently. eg. organs

Remembering

Our Amnesia of Infancy leads us to develop simplistic views of what memories are and how they work. Minsky argues for goal based organisation (accomplishment) rather than descriptive organisation (data base and matches):

POSITIVES
  • What kinds of goals might this item serve?
  • In which situations might it be relevant?
  • How has it been applied in the past?

NEGATIVES
  • What are its most likely side effects?
  • How much will it cost to use it?
  • What are its common exceptions and bugs?

SOURCES AND LINKS
  • Was it learned from a reliable source?
  • Is it likely to be outdated soon?
  • Which other people are likely to know it?

Intentions and Goals

What is "self control", responsibility or intention? Moralists, Psychiatrists and Jurists argue about this.

Sometimes a goal can seem like a physical force, hard to resist, even though part of us does not want to do it. The goal may conflict with our high level values. There is no reason to expect that all our goals should be consistent.

Difference Engines

Psychology words don't meaningfully describe goals, they just pass the meaning onto another word that needs to be explained (want, motive, desire, purpose, aim, hope, aspire, yearn, crave). We need to talk about the underlying machinery:
"A system will seem to have a goal when it persists at applying different techniques until the present situation changes into a certain other condition"
Motives and goals could be explained as consisting of these three things:
Aim - description of a possible future situation
Resourcefulness - methods to reduce the difference between the present situation and the future situation
Persistence - keep applying those methods


When you hear a story you react most to how it differs from what you expected. Some names for this include accommodation, adaptation, acclimatization, habituation, becoming accustomed
Our eyes normally make small motions which helps maintain an image. Our systems mainly react to change.

Roger Schank, Tell Me a Story (1990) has conjectured that representing events as stories may be one of our principal ways to learn and remember

Making Decisions

When people say, "I used my free will to make that decision," this is roughly the same as saying, "some process stopped my deliberations and made me adopt what seemed best at the moment". "Free will" is not a process we use to make a decision, but one that we use to stop other processes! "My decison was free" is similar to "I don't want to know what decided me"

Reasoning by Analogy

Minsky identifies reasoning by analogy as one of the main methods by which we solve new problems. Why does analogy work so well? According to Douglas Lenet it is because there is a lot of common causality in the world.

Positive vs. Negative Expertise

We see things as positive because we censor or suppress other processes that would see them as unpleasant.

Minsky simulates a dialogue with a teacher who believes in positive reinforcement and small steps. Such an approach is not bad in itself but limited:
  • difficult tasks almost always involve episodes of distress and discomfort
  • reinforcement can lead to rigidity, lack of adaption
  • other processes may fail when the "normal way" is abandoned
  • the development of higher level managerial resources is put on hold
We must learn to "enjoy" some suffering when learning new things that need large scale changes in how we think. It's a mistake to make education too pleasant.

minsky 5: levels of mental activities

Overview of Chapter 5 of The Emotion Machine (summary, online draft, buy)

Minsky proposes a 6 level model of mind, which is described in some detail (for a little more detail see the wiki):


Psychologist: I find it hard to see the difference between your top most three levels
Student: No theory should have more parts than it needs

Minsky: The boundaries are indistinct but psychology is not like maths or science. When you know that your theory is incomplete then leave some room for other ideas you might need later!

Individualist: Where is the Self that makes our decisions? What decides which goals we'll pursue?

Minsky: It would be dangerous to locate all control in one single place because then all could be lost from a single mistake. Our minds use multiple ways to control themselves. Freud anticipated this with his ideas of superego, ego and id

Imagination:
"We don't see things as they are. We see things as we are" - Anais Nin

... most of what we think we see comes from our knowledge and our imagination
Abraham Lincoln (vague patches of darkness and light)

We don't know how our brains achieve this. "Seeing" seems simple because the rest of our minds are blind to the processes that do it for us. Perception is complex and invisible

More abstract, higher level descriptions are more powerful and efficient. Minsky is arguing for the importance of higher level semantics here as compared with inefficient processing of visual images

We internalise prediction machines. Nevertheless, such a system will never be very resourceful until it knows a great deal about the world it is in. This leads into the next chapter (Common Sense)

Related:
Dennett's Creatures is a bit similar to Minsky's model of mind
Skinnerian creatures ask themselves, "What do I do next?"
Popperian creatures ask themselves, "What do I think about next?"
Gregorian creatures ask themselves, "How can I learn to think better about what to think about next?"

minsky 4: consciousness

Overview of Chapter 4 of The Emotion Machine (summary, online draft, buy)

Consciousness is a suitcase word. It can mean such diverse things as a unifier, self awareness, identity, an animator of the mind, a provider of meaning or a detector of feelings. It refers to many different mental activities that don't have a single cause or origin

We need a way to divide the mind into parts that is more meaningful than crude folk psychology "dumbbell" (two part distinctions) such as conscious v. unconscious, premeditated v. impulsive, etc. For example, the "unconscious" state may represent various different states. Information may be inaccessible for a variety of reasons such as simple failure to retrieve, active censorship or "sublimated" into a form which can't be recognised (to borrow Freud's terminology)

Minsky uses the Plato / Socrates shadows on the cave allegory to discuss a possible structure of our minds.

Imagine we have an A-Brain and a B-Brain. The A-Brain receives signals from external world via organs such as eyes, ears, nose and skin - and can react to those signals by making our muscles move. The A-Brain has no sense of what the events mean.

The B-Brain receives and reacts to signals from A-Brain. However, the B-Brain has no direct connection to the outer world, so it is like the prisoners in Plato's cave, who see only shadows on the wall. The B-Brain mistakes A's descriptions for real things

For example, if B sees that A has got stuck at repeating itself, it might suffice for B to instruct A to change its strategy. To acquire its skills the B brain may need a C brain to help (eg. it's not always appropriate to stop repeating oneself, especially when crossing a road)

Student: Would not this raise increasingly difficult questions, because each higher level would need to be smarter and wiser?
Minsky: No, C-Brain could act as a "manager" who has no special expertise about particular jobs but could still give "general" guidance, like: "If B's descriptions seem too vague, C tells it to use more specific details", etc.

Minsky proposes six levels of processes:


The organism principle

Student:
Does your theory really need so many different levels? Are you sure that you can't make do with fewer of them? Indeed, why should we need any "levels" at all - instead of a single, big, cross connected network of resources?
The Organism Principle: When a system evolves to become more complex, this always involves a compromise. If its parts become too separate, then the system's abilities will be limited. But if there are too many interconnections, then each change in one part will disrupt many others

Hence, our bodies are composed of distinctive separate parts we call "organs". This also applies to the brain organ. New design is built on top of old design: "... large parts of our brains work mainly to correct mistakes that other parts make ..."

Psychology is hard because each "law of thought" has exceptions. It will never be like physics which has "unified theories" which work flawlessly.

Minsky's proposed solution to consciousness being a suitcase word: We must try to design - as opposed to define - machines that can do what human minds do. DESIGN not DEFINE

Consciousness seems mysterious because we exaggerate our perceptiveness. Most processes are hidden from us. We see things less as they are and more with a view to how they are used (eg. hammer, ball). Our minds did not evolve to serve as instruments for observing themselves

There are many suitcase words in psychology: attention, emotion, perception, consciousness, thinking, feeling, self, intelligence, pleasure, pain, happiness

Why do people, including scientists, look for a single concept, process or thing to explain multiple aspects of mind? They prefer one large problem rather than dozens or hundreds of smaller problems

Aaron Sloman:
"People are too impatient. They want a three-line definition of consciousness and a five-line proof that a computational system can or cannot have consciousness. And they want it today. They don't want to do the hard work of unraveling complex and muddled concepts that we already have, and exploring new variants that could emerge from precisely specified architectures for behaving systems"
How do we initiate what we call consciousness?

Most mental processes don't cause us to think or reflect about why or how. But when those low level processes don't function well or when they encounter obstacles the high-level activities start up with these properties: self models, serial processes, symbolic descriptions and recent memories. A trouble detecting critic (T) might operate as shown:


If you reverse the trouble detector, then you have a consciousness detector, ie. a part of our brain that sends signals to other parts including our language system which then invents words to describe this condition, such as: conscious, attentive, aware, alert, me, myself, deliberate, intentional, free will

Our higher level descriptions are mainly stable, they have been formed previously. Hence it's an illusion to think we live in the present moment!

The Immanence Illusion: For most of the questions you would otherwise ask, some answers will have already arrived before the higher levels of your mind have had enough time to ask for them. Our Critics may recognise a problem and start retrieving the knowledge you need before your other processes have had time to ask questions about it

Some philosphers regard explaining "subjective experience" as the hardest problem in psychology: the quality of deep blue, the sensation of middle C. Minsky argues that terms like experience or inner life refer to big suitcases of different phenomena. Our "insights" from inside our mind are frequently wrong. If consciousness means "awareness of our internal processes" then it doesn't live up to its reputation

Minsky uses the word model in this book to mean "a mental representation that can be used to answer some questions about some other, more complex thing or idea". We have multiple models: professional, political, beliefs about abilities, ideas about social roles, moral and ethical views. Our thinking depends on (a) quality of models; (b) how good our ways of choosing which model to use in different situations

"Free will" might mean "I have no model that explains how I made the choice I made"

The Cartesian Theater is the idea that our minds contain a central stage on which various actors perform while we (the self) watches and then makes decisions. This popular idea is analysed and debunked. The spatial metaphor is deeply held and hard to abandon

The idea that we live in the here and now, moving steadily into the future - is an illusion! "Real time" is a process of zigzagging through memories as we assess our progress on goals, hopes, plans and regrets!

Dennett and Kinsbourne (1992):
"... there is no single, definitive 'stream of consciousness', only a parallel stream of conflicting and continuously revised contents"
There are problems with thinking too much about how we think, to be too self aware would be very tedious! Minsky employs an amusing and enlightening dialogue with HAL to illustrate:
... interpreting those records is so tedious ... I often hear people say things like, "I'm trying to get in touch with myself." ... take my word for it, they would not like the result of accomplishing this

Saturday, October 04, 2008

minsky 3: from pain to suffering

Overview of Chapter 3 of The Emotion Machine (summary, online draft, buy)

"Emotions are different Ways to Think."

When I reread this chapter it seemed to have two separate parts:
Sections 3-1 to 3-4 is an extended discussion of whether pain and suffering is a mystery
Sections 3-5 to 3-8 is building on Freud's idea that our minds are battlegrounds between basic instincts and higher ideals

What is the connection between these two parts? I think Minsky is using pain and suffering as one example of basic instincts. This sets the scene for the contrast of the Freudian conflict later in the chapter.

Pain and pleasure have many similar qualities. They both constrict one's range of attention, both have connections with how we learn, both reduce the priorities of one's other goals. Pain protects our bodies but destroys our minds. In an evolutionary sense this might be a programming bug that evolved before our higher level intellects

The extended discussion on pain, suffering and grief is very interesting and punctuated with some great quotes (from Woody Allen, Dennett, Shakespeare and Oscar Wilde). Minsky is a great writer as well as having great ideas.

One deficiency of behaviourism is that it only observed the actions of what people do while ignoring questions about what people do not do. Negative expertise is a very large part of every person's precious collection of commonsense knowledge. Negative expertise might work through Critics, each of which learns to recognise some particular kind of potential mistake

Types of Critics:
  • Corrector - declares you are doing something dangerous
  • Suppressor - interrupts before you begin an action
  • Censor - prevents "incorrect" ideas occurring to you in certain situations
Freud was on the right track. The human mind is like a battleground, there are continual conflicts between our animal instincts and our acquired ideals


Human thinking does not proceed in any single, uniform way

On the issue of controlling our moods:

If you could switch all your Critics off then nothing would seem to have any faults ... everything now seems glorious

If you turned too many Critics on, you'd see imperfection everywhere ... ugliness ... if you found fault with your goals themselves, you'd feel no urge to straighten things out, or to respond to any encouragement

Sometimes we use one emotional state to combat another emotional state. For example, you might call up an image of a Challenger to use jealousy, anger or shame to combat sleep

Why do we need such fantasies, why aren't we more rational?
  • concept of "rational" itself is a kind of fantasy because our thinking is never based on just pure logic
  • directness would be too dangerous, if we could turn Hunger off we might starve; if we could turn Anger on we might fight all the time; if we could extinguish Sleep then we might wear out bodies out (these considerations shaped our evolution)

Friday, October 03, 2008

minsky 2: attachments and goals

Overview of Chapter 2 of The Emotion Machine (summary, online draft, buy)

This chapter is about how people choose which goals to pursue, through the strong self conscious feelings such as Pride and Shame. Pride or Shame (as distinct from not so strong emotions such as Pleasure or Dissatisfaction) play a unique role in determining our values, goals or ends (as distinct from learning methods of how to achieve a goal once we have it)

A sketch of a difference engine, which works to reduce the difference between your present situation and a goal, is introduced

Imprimers: a new word is introduced by Minsky (derived from imprinting):
"An imprimer is one of those persons to whom a child has become attached"
The "caregiver" word is not sufficient since attachments can form without physical care

Limits of behaviourism: The idea of learning by being "reinforced" by success or by "trial and error" does not explain how we develop completely new goals or "values" or "ideals". It would be potentially dangerous if strangers could easily alter our higher level goals.

Several different ways in which a child might change:
  • Positive experience
  • Negative experience
  • Aversion learning: when a stranger scolds ...
  • Attachment praise: imprimer praises
  • Attachment censure: imprimer scolds
  • Internal impriming
How could we elevate a goal? By moving it up the 6 level model, eg. from Deliberative thinking to Self-Conscious emotions



"The problem we faced" and "the action we took" are not simple objects that we can connect. Minsky argues that we also need to make structures which represent both external events and relevant internal mental events. We need reflective resources to choose which things to remember out of all the things we were doing when solving a particular problem. Minsky calls these "credit assignments"

Thursday, October 02, 2008

minsky 1: Falling in Love

Overview of Chapter 1 of The Emotion Machine (summary, online draft, buy)

This chapter introduces a framework to think about the mind.

An attempt is going to be made to explain things that we take for granted, such as, perception, our comprehension of words, our preference for certain feelings.

Many words used to describe emotions and psychology are suitcase words, which have vague or multiple meanings. At strategic moments in this book Minsky introduces new words of his own because our current vocabulary often unintentionally serves to obscure the real workings of the mind

Some mind myths are critiqued. We don't have a single "logical" or "rational" Way to Think. Logic says nothing about which assumptions we begin with. There are dozens of different Ways to Think. We don't have a single Self but multiple models of Self.

The purpose of Minsky's lampooning of love is to point out that with love and some other emotions it is as though a switch has been thrown and a different program has started to run. It's an illustration of one of our many Ways to Think.

Quite a lot of infant behaviour can be explained by IF-then-DO reaction rules


With deliberate, calculated vagueness Minsky conceives of the mind as a cloud of resources. Different resources are activated for different Ways to Think and / or different emotional states.

During our childhood years our brains go through multiple stages of growth. Minsky conjectures that at least six levels of mental procedures will summarise his main ideas about how the human mind is organised

These ideas are explained in more detail in subsequent chapters

total cost of ownership: XO versus computer lab

Tony Anderson published his calculations for total cost of ownership of an XO compared with a computer lab in a long and sometimes acrimonious thread on olpc news:

his results:
OLPC: 4.5 cents per hour
computer lab: $2 per computer access hour

If he's correct then that's 40 times more value for money roughly speaking :-)

Here are his calculations:
Surprise! TCO for XOs in schools is $437.50 (4.5 cents per hour)

I have attempted to recalculate the TCO based on a more realistic description of the OLPC model, but using the numbers in the report.

The TCO model in the report is based on teaching ICT in a public secondary school. The purpose of this training is to prepare university-bound students to enter a course of study leading to a position in the computer field. Assume that the school enrollment is 160 students. The study assumes 3.5 hours lab time per day, 5 days per week, for 39 weeks per year (10920 available hours per year). Shared among 160 students for an average of 68.25 hours per year per student or 1.75 computer access hours per week.

The OLPC model is based on each child having an XO throughout the class day and at home on evenings, weekends and holidays. It is also focused on primary school education. Assuming a school of 160 students, at least 160 XOs will be needed. Assume a steady-state situation in which each child entering the school in the second grade gets an XO. Each child keeps his XO throughout the primary years (grades 2 thru 5 as assumed in the study). The school purchases an additional 5 XOs per class to provide for loss, damage, repair, incoming transfers, and the teacher. In addition the OLPC model assumes a dedicated computer (XS) per school which provides internet access, provides backup for the XOs, caches instructional materials, and supports a CMS (Moodle). Each child has a computer which could be used each day for six hours in class and two hours at home (plus ten hours on weekends). This is approximately 1800 hours per year - let's assume 1000 hours.

First, look at the direct computer cost, excluding initial setup, training, connectivity to the internet, electricity, and Tier 1 (software) support.

Case 1: Public Secondary School (grades 9-12) sets up a computer lab with 16 desktops (one is also a server) to teach ICT. The total cost is $8000 with a per seat cost of $125. This is a approximately $0.40 per hour for computer access ($125/312).

Case 2: Public Primary School (2-5) provides each child and four teachers with an XO. The initial purchase for this school is 160 + 4 (teachers) + 4 (reserve) for a rounded-up total of 170 at $200 each ($34000). The annual buy is 45 laptops at $200 per student ($9000). A dedicated server per school costs $500 (headless mainstream desktop). This is a total cost of $34500 ($216 per student). If we assume utilization of 1000 hours per year, the cost is($216/4000) or 5 cents per hour. Note: if leaving students keep their XO, the cost in the first four years is $34000 + $27000 (3 * $9000).

Other costs:

Case 1 the computer lab pays $33886 total for initial setup ($5351), training ($10620), software support ($10920), hardware support ($2432), electricity and internet connectivity ($5351), and damage and theft ($2640). This increases the computer lab cost to $2500 (41886/16) per seat and approximately $2 per computer access hour ($41886/16/4/312).

In case 2, the costs of Tier 2 and 3 (hardware maintenance) and the costs of damage and repair are handled by purchasing extra XOs. In the OLPC case the extra cost is $28314 (excluding $5072). The OLPC total cost is $70000 ($41500 + $28316). This is $437.50 per student or 4.5 cents per hour.

Posted by: Tony Anderson on September 29, 2008