Michael Wheeler's presentation in Sorbonne, "Philosophy of the Web" seminar, March 3 2012.

Cognitive Extension and the Web-Enabled

Mind

Mike Wheeler

School of Arts and Humanities: Philosophy

University of Stirling

Poor Memory or Adaptive Memory?

There is evidence that, in an era of laptops, tablets and smartphones, with powerful Internet search engines, our organic brains tend to internally store not the information about a topic, but rather how to find that information using the available technology

See data from Sparrow, Liu and Wegner (2011), ‘Google Effects on Memory: Cognitive Consequences of Having Information at Our Fingertips’, Science 333 (6043).

The Guardian reported this research under the heading ‘Poor Memory? Blame Google’.

By contrast, the experimenters talk of “an adaptive use of memory” in which “the computer and online search engines [should be counted] as an external memory system that can be accessed at will”

Technologies Are (Part of) Us

Clark’s description of human beings as natural born cyborgs reminds us that it is of our very nature as evolved and embodied cognitive creatures to create tools which support and enhance our raw organic intelligence by dovetailing with our brains and bodies to form shifting human-artefact coalitions operating over various time-scales.

This is no less true of our engagement with the abacus, the book or the slide-rule than it is of our engagement with the laptop, the tablet or the smartphone.

Defining the Positions

Embedded Cognition (the default view): the distinctive adaptive richness and flexibility of intelligent behaviour is regularly, and perhaps sometimes necessarily, causally dependent on (a) non-neural bodily structures and/or movements, and/or on (b) the bodily exploitation of environmental props or scaffolds.

Extended Cognition: there are actual (in this world) cases of intelligent action in which thinking and thoughts (more precisely, the material vehicles that realize thinking and thoughts) are spatially distributed over brain, body and world, in such a way that the external (beyond-the-skin) factors concerned are rightly accorded cognitive status.

Cognitive Self-Stimulation

As Clark (Supersizing the Mind) explains, cognitive self-stimulation occurs when

a) neural systems are causally responsible for producing certain bodily movements and (sometimes) beyond-the-skin structures and events which are then recycled as inputs to those and/or other neural systems, and

b) this feedback process sustains sophisticated brain-body or brain-body-environment loops of exploitation, co-ordination and mutual entrainment, with various problem-solving benefits.

Self-Stimulation and Cognitive Extension

Clark introduces his treatment self-stimulation in the case of gesture as a “worked out example of extended cognizing in action”

But his cornerstone claim is that the “key distinction between “merely impacting” some inner cognitive process and forming a proper part of an extended cognitive process looks much less clear . . . in cases involving the systematic effects of self-generated external structure on thought and reason” (Supersizing the Mind)

So his argument is actually that where there is cognitive self-stimulation (e.g. in gesturing), there is no clear distinction between cognitive embeddedness and cognitive extension.

A Different Interpretation

In cases of cognitive self-stimulation:

(i) the distinction between cognitive embeddedness and cognitive extension is eroded in such a way that whatever evidence there is that tells in favour of the embedded view, tells equally in favour of the extended view, and

(ii) under such circumstances, we are theoretically permitted to adopt the extended view.

But (i) flouts the causal-constitutive distinction

And (ii) flouts the thought that the embedded view is the default position in the debate

A Line of Response The empirical evidence of self-stimulation that supposedly

undermines the embeddedness-extension distinction also undermines the causal-constitutive distinction.

“Sometimes, all coupling does is provide a channel allowing externally originating inputs to drive cognitive processing along. But in a wide range of the most interesting cases, there is a crucially important complication. These are the cases where we confront a recognizably cognitive process, running in some agent, that creates outputs (speech, gesture, expressive movements, written words) that, re-cycled as inputs, drive the cognitive process along. In such cases, any intuitive ban on counting inputs as parts of [cognitive] mechanisms seems wrong.” (Clark)

Arguing for Cognitive Extension I:

the Parity Principle (PP)

“If, as we confront some task, a part of the world functions as a process which, were it to go on in the head, we would have no hesitation in accepting as part of the cognitive process, then that part of the world is (for that time) part of the cognitive process.”

Clark, Supersizing the Mind

(drawing on Clark and Chalmers, ‘The Extended Mind’)

Notice that parity considerations cut both ways

Arguing for Cognitive Extension II: Extended Functionalism

PP depends on the multiple realizability of the mental

Functionalism in the philosophy of mind provides a well-established platform for securing multiple realizability.

Thinking Through Parity

But PP, even if an explicitly functionalist register, doesn’t solve the problem of determining which functional differences matter when deciding what counts as cognitive

A way forward: we understand the relevant concept of parity not as ‘parity with the inner simpliciter’, but rather as ‘parity with the inner, with respect to some sort of locationally uncommitted account of what counts as cognitive’.

PP itself is ‘merely’ an heuristic device designed to free our intuitions from neural chauvinism

Folk Intuitions

If an environmental protester had stolen the plans of Heathrow Terminal 5, would the folk have been concerned about the whereabouts of part of Richard Rogers’ mind?

A plausible explanation: our folk grip on the cognitive involves an internalist presumption

Arguing for Cognitive Extension III: a Mark of the Cognitive

A proposal: what the hypothesis of cognitive extension needs is some kind of scientifically informed theory that tells us which functional differences are relevant to judgments of parity and which aren’t.

More specifically: first we give a scientifically informed account of what it is to be part of a cognitive system, one that is independent of where any candidate element happens to be spatially located. Then we look to see where cognition falls.

This is what Adams and Aizawa (The Bounds of Cognition) have dubbed a mark of the cognitive

Cognitive Self-Stimulation as a Mark of the Cognitive

So maybe we could adopt the view that being the kind of self-generated input that supports a process of cognitive self-stimulation is a mark of the cognitive

But is this an independently plausible claim?

Problem: a self-generated input in a cognitive self-stimulating loop may very well make its turbo-charging contribution to thought while remaining non-cognitive in character.

Hybrid Mechanisms Now recall Clark’s claim that, in cases of cognitive self-

stimulation, “any intuitive ban on counting inputs as parts of [cognitive] mechanisms seems wrong.”

It is unclear that the right to add the term ‘cognitive’ has been earned here.

Embedded and extended theorists agree that self-generated inputs that support cognitive self-stimulating loops operate within well-defined mechanisms that turbo-charge thinking.

For the embedded theorist, however, the properly cognitive mechanisms in play are sub-systems of larger, performance-enhancing loops, where the latter are not cognitive mechanisms in their own right, even though they contain cognitive mechanisms.

Embedded Rowers

Consider Baca and Kornfeind’s self-stimulating rowing training loop for the acquisition and honing of bodily skills.

Although the self-generated inputs are key aspects of the mechanism by which the rower is tuned for improved performance, there is no temptation to categorize those inputs as realizers of the observed bodily adaptation, as opposed to elements that have a critical causal impact on that adaptation.

Why should things carve up any differently when the focus of attention is a self-stimulating loop that enhances thought?

From Symbolic Coupling…

So can we provide a scientifically informed mark of the cognitive?

Bechtel argues that cognitive achievements such as mathematical reasoning, natural language processing and natural deduction, are the result of sensorimotor-mediated interactions between internal neural (connectionist) networks and suites of external symbols.

Now consider the phenomenon of systematicity

The “property of systematicity, and the compositional syntax and semantics that underlie that property, might best be attributed to natural languages themselves but not to the mental mechanisms involved in language use” (Bechtel, Natural Deduction in Connectionist Systems)

Is this a case of cognitive extension?

…to Extended Physical Symbol Systems

PP would suggest so, but, as know, PP provides only an incomplete argument. We need a mark of the cognitive.

Here is a possible mark of the cognitive: a physical symbol system (PSS), when sufficiently complex and suitably organized, and when placed in the operating context of a complete cognitive architecture, has the necessary and sufficient means for certain aspects of cognition.

I claim that the Bechtel-style network-plus-symbol-system architecture is an instantiation of an extended PSS and thus, if we adopt the above mark of the cognitive, it’s an instantiation of an extended cognitive system (or subsystem)

Notice that questions of revisionism no longer (obviously) favour the embedded view

Riding the Waves

So-called second wave ExC downplays PP in favour of considerations such as complementarity

Sutton (‘Exograms and Interdisciplinarity’): “in extended cognitive systems, external states and processes need not mimic or replicate the formats, dynamics, or functions of inner states and processes”, so “different components of the overall… system can play quite different roles and have different properties while coupling in collective and complementary contributions to flexible thinking and acting”.

But by placing the stress on difference in this way – even necessary difference – the second-wavers risk being trapped within an embedded internalist prison

Three Issues

1. Integration into an individual cognitive architecture

2. Portability versus Reliability

3. Agency and Cognitive Ownership

Education and Technology: a Slippery Slope or a Cognitive Incline?

Pen and Paper

Slide Rules

Limited capability generic calculators

Restricted Internet Access

Largely unrestricted Internet access

The user’s own smartphone

Mainlined Google

One way of focussing the issue here is to ask under what conditions our children’s intelligence should be formally examined

Crossing the Line

Swimsuits that improved stability and buoyancy, while reducing drag to a minimum, were outlawed by swimming’s governing body FINA after the 2009 World Championships.

FINA stated that it “[wished] to recall the main and core principle that swimming is a sport essentially based on the physical performance of the athlete”

Perhaps education is a process essentially based on the unaided cognitive performance of the learner

This Time it’s Personal

One might argue that generic technology is permissible in an exam setting, but individualized technology isn’t.

If the extended cognition view is correct, however, this may be unsustainable

The extent to which some external element is configured so as to interlock seamlessly with the desires, preferences and other personality traits that are realized within the rest of the putative cognitive system will be one factor in determining whether the cognitive system includes that external element.

Put crudely, individual tailoring will, if other conditions are met, indicate that the technology in question counts as part of the learner’s mind (and surely we want to allow that into the examination hall).

Dwellers on the Threshold

Increasingly, architects will be designing buildings that, via embedded, Internet-enabled computers, autonomously modify our spatial and cognitive environments in the light of what those buildings ‘believe’ about the needs, goals and desires of their users.

“An intelligent building is… a building that has the

ability to respond (output) on time according to processed information that is measured and received from exterior and interior environments by multi-input information detectors and sources to achieve users’ needs and with the ability to learn.” (Sherbini and Krawczyk, Overview of Intelligent Architecture, 2004)

The Cybertecture Egg

Planned building from James Law Cybertecture International

“Integrates technology, multimedia, intelligent systems and user interactivity to create customizable living and working spaces that focus on experience.”

http://gazette-world.blogspot.com/2008/05/hi-tech-building-from-james-law.html

Interactive features that monitor occupant’s vital health statistics (e.g. blood pressure, weight).

Users can customize their views with real time virtual scenery.

Ambient Assisted Living

Consider research from the Ambient Assisted Living Research Department at the Fraunhofer Institute for Experimental Software Engineering, Kaiserslautern

An intelligent home automation system that uses a network of hidden sensors to monitor the daily routine of the occupants, detecting and assessing risks.

Sensors automatically report data to a control centre somewhere in the house. So, e.g., the system can tell if someone has fallen and is able to send that information to a designated contact person.

The bathroom has a toilet that recognizes the user and adjusts to the proper height, a light that turns on and off automatically, and a tap that turns itself off to save water.

It also contains a mirror with illuminated pictograms to help those who are easily confused remember what to do next, such as brush their teeth, wash, shave or take medication.

From Portability to Reliability

The portability objection to cognitive extension: a material element may count as part of the vehicle of a cognitive state or process only if you carry (or at least are able to carry) that material element around with you

This cuts the cognitive cake as follows: your brain meets this necessary condition; intelligent architecture doesn’t.

But consider mobile access to the Internet The message: a material state or process may count as

part of the vehicle of a cognitive trait only if that state or process meets a dynamic reliability condition.

Intelligent architecture may meet this necessary condition

Interactive Architecture: Usman Haque

Haque argues that we need to shift from reactive architecture to a genuinely interactive architecture

Single-loop interaction: particular outputs for particular inputs determined in advance

Multiple-loop interaction: depends upon the openness and the continuation of cycles of response, and on the ability of each system, while interacting, to have access to and to modify each other’s goals

Talk to the Walls

“I concede that reactive or single-loop devices that satisfy our creature comforts are useful for functional goals (I am thinking here of Bill Gate's technologically-saturated mansion; or building management systems that seek to optimise sunlight distribution; or thermostats that regulate internal temperature). Such systems satisfy very particular efficiency criteria that are determined during, and limited by, the design process. However, if one wants occupants of a building to have the sensation of agency and of contributing to the organisation of a building, then the most stimulating and potentially productive situation would be a system in which people build up their spaces through "conversations" with the environment, where the history of interactions builds new possibilities for sharing goals and sharing outcomes. In such architectural systems, inhabitants themselves would be able to determine efficiency criteria.” (Haque, Architecture, Interaction, Systems, 2006)

Paskian Systems

“In such systems, there may be an environmental sensor/actuator device which monitors a space and is able to alter it. However, rather than simply doing exactly what we tell it (which relies on us knowing exactly what we want within the terms of the machine, i.e. within the terms of the original designer) or alternatively it telling us exactly what it thinks we need (which relies on the machine interpreting our desires, leading to the usual human-machine inequality, or, as some would say, mistreatment), a Paskian system would provide us with a method for comparing our conception of spatial conditions with the designed machine’s conception of the space” (Haque, Architecture, Interaction, Systems, 2006)

Paskian Systems and Cognitive Extension

Although non-Paskian intelligent architecture may qualify as proper parts of the dweller’s cognitive economy, Haque’s Paskian architectural systems will not.

This is because of the very conditions that make possible the capacity of Paskian systems to enter into richly interactive dialogues, the fact that they may operate with categorizations and goal-states that diverge from those of their human users (e.g. Haque’s own Evolving Sonic Environment).

Paskian systems exhibit a kind of agency that raises questions of cognitive ownership and thus prevents them from being incorporated into the cognitive systems that are centred on their human users.

"I go up", said the elevator, "or down."

"Good," said Zaphod, "We're going up."

"Or down," the elevator reminded him.

"Yeah, OK, up please." There was a moment of silence. "Down's very nice," suggested the

elevator hopefully. "Oh yeah?" "Super." "Good," said Zaphod, "Now will

you take us up?" "May I ask you," inquired the

elevator in its sweetest, most reasonable voice, "if you've considered all the possibilities that down might offer you?"

Conversation with an elevator designed by the Sirius Cybernetics Corporation The Restaurant at the End of the Universe, Douglas Adams

A Possible Response The conclusion drawn here is too strong Consider collaborative activities in which no one individual

could complete the cognitive task. Hutchins’ ship navigation example (Cognition in the Wild,

1995) provides a paradigmatic case This may look like an instance of cognitive extension If it is, then we can resuscitate the idea of extended

Paskian systems However:

(i) The fact that no one individual could complete the task is not relevant to securing cognitive extension

(ii) Hutchins’ example is a case of an unowned or group-owned distributed cognitive process, not of cognitive extension as standardly understood

(Tentative) conclusion: where there’s more than one will, there’s no way to cognitive extension