Extending the Mind with Cognitive Prosthetics?

Extending the
Mind with
Cognitive
Prosthetics?

Andy Clark
School of Philosophy, Psychology
and Language Sciences (PPLS)

University of Edinburgh,
Scotland, UK

andy.clark@ed.ac.uk

With special thanks to: Rob Rupert, Kenneth Aizawa,
Fred Adams, Mark Rowlands, Dave Chalmers, Julian
Kiverstein, Mark Sprevak, Richard Menary, and Mike
Wheeler.

Where in physical
space lies the
machinery of mind
and cognition?

What it isn’t: Target is that the
machinery of mind might, perhaps in some alien beings, be
smeared across more than the neural economy.

Nor is the claim merely that non-brain activity
impacts the mind.

No-one denies that causal commerce
between mind and world matters, and it
changes what we think.

The contentious claim is that the
mechanisms of mind are not all in the head
(= the 'extended mind hypothesis’- Clark
and Chalmers (1998)

It’s as if someone said that your calculator
or currency converter’s MECHANISMS
were not all inside your laptop.

This is when, e.g., we use a web-based
currency converter.

It is when we use the built-in calculator
on the mac.

TXM: The Main Idea:

The mechanisms of (your)
mind are as free to bleed
into the (rest of the) world
as the mechanisms of
calculation are to bleed into
the web

1.The Extended Mind Claim (super-mini-version)

2. Some Objections and Replies

3.Cognitive Extension versus Cognitive
Shrinkage

The Extended Mind Claim

For the brain, it doesn’t matter whether information is
stored in the head or in the wider world, just so long as
it knows what kind of information is there and how
to get at it as soon as we (the agent) need to put it to
some practical use.

Brains like ours are already adept
at trading easy access against
expensive internal biological
representation and storage.

Roboticists and psychologists have known this for a
while..

Brooks: “The world is
its own best model” O’Regan: “The world
as external memory”

Extending the Mind with Cognitive Prosthetics?

That feeling of seeing all the colour and detail in the
scenes is probably due to a kind of implicit meta-
knowing.

Our brains know that they can usually retrieve more
detailed info when needed, so we feel as if we already
see all the detail.

This is not really a mistake.

For we are poised to access that information just-in-
time for use.

TXM = a cognitive application of the same idea.

Compare: your feeling that you already know what
month this is

This is not due to your constantly rehearsing the
answer in your conscious mind

(continually sub-vocalizing “March” ‘March”
“March”).

Rather, it is due to your implicit meta-knowing that

This is the kind of thing you know and that (in
normal circumstances) you are poised to access that
information pretty much at will and as and when
needed.

So maybe being ‘ready-stored in the head’ is an
optional extra for dispositional believing (standing
beliefs) too?

Perhaps what matters here too (Clark and Chalmers
(1998) is being poised for easy access….

Yields the case of ‘note-book Otto’….

But TXM is not only about dispositional beliefs…many
of our best mind-extending loops into the world (just
like many of the best loops inside the brain) are much,
much fancier than simple access/retrieval loops…
think of loops like these:
gesturing while you talk (actively looping into the
body) – see Clark “Curing Cognitive Hiccups” Journal of Philosophy 2008)

scribbling while you think (looping into ‘passive’
external media, but not a case of simple offloading)

working with a highly-practiced software package
(looping into an active semi-intelligent sub-system)…

This complexity is highlighted in a famous exchange
between Richard Feynman (the Nobel laureate physicist)
and the historian Charles Weiner

“ Weiner once remarked casually that [a batch of notes and
sketches] represented “a record of [Feynman’s] day-to-day
work,” and Feynman reacted sharply.

“I actually did the work on the paper,” he said.

“Well,” Weiner said, “the work was done in your head,
but the record of it is still here.”

“No, it’s not a record, not really. It’s working. You have
to work on paper and this is the paper. Okay?” “

Quoted in Genius (Gleick’s biography of Feynman)

It is not that all the thinking happens inside, and the loop out
into symbols on a page is just a kind of convenience or a way
to avoid forgetting.

Rather, the loops to external media form part and parcel of a
complex, integrated, bio-technologically hybrid system
for thinking.

For lots of examples and discussion, see Clark, Supersizing
The Mind (Oxford Univ Press, 2008))

The extended mind story is most convincing, I think,
when we can by-pass the stage of consciously
consulting an external or internal information store
at all….

Trials (at MIT Media Lab) of so-called ‘memory
glasses’: aids to recall for people with impaired
memory or visual recognition skills.

The Memory Glasses: Wearable Computing for Just-in-Time Memory Support Richard W. DeVaul
(MIT thesis, 2004). See also paper in 7th IEEE International Symposium on Wearable Computers
(2003)

= a Terminator style eye-glass display

The glasses work by matching the current scene
(a face, for example) to stored information and
cueing the subject (using the glasses-mounted
display) with relevant information (a name, a
relationship).

The Memory Glasses: Wearable Computing for Just-in-Time Memory Support Richard W. DeVaul
(MIT thesis, 2004). See also paper in 7th IEEE International Symposium on Wearable Computers
(2003)

The cue may be overt (consciously perceived by the
subject) or covert (rapidly flashed and hence
subliminally presented).

In the covert case, functionality is still improved
without any process of conscious awareness of the
cueing on the part of the subject.

Subjects like this a lot better!

It is easy to imagine cases that then enhance
knowledge rather than merely ‘restore’ it.

Recognizr is a controversial app. purchased by
Apple (for over 15 million dollars).

- it makes a 3D facemap on-the-spot from a
photo input.

That means other applications can then match
that face to pictures taken from other angles etc
on the web, rapidly identifying the person and
retrieving all kinds of associated information.

Upshot: a body-mounted camera could constantly
generate these 3D face-maps, then get and act
upon a bunch of additional information from a
rapid web-trawl.

Imagine a version
where, if the person
meets some desired
condition (e.g. being a
fan of Paris SG) you
get a barely-
perceptible buzz from
a vibrotactile element
somewhere on your
body.

What happens when THAT becomes part of the
suite of robustly available equipment, through
which you encounter the wider world?

Soon you cease to consciously notice the gentle
buzz and simply register what it is telling you.

“just knowing” who is (probably) an SG fan
will then simply become part of how you
experience a new situation.

Your in-the-head cognitive routines will become
geared to the easy availability of the information,
creating a new, co-adapted, cognitive whole.

= ‘cognitive dovetailing’

The operation of a wide variety of such continuously
running programs may be compared to that of your
own (complex, active!) unconscious neural sub-
structures.

You will count as ‘using’ these software entities only
in the same attenuated sense as you ‘use’ your
hippocampus or frontal lobes.

Far better to say that the agent that IS you just is the
larger distributed system.

Speculation:

Such innovations – made increasingly
possible by the combination of web-based
infrastructure and portable technologies
that can learn about the agent as the
agent uses them - will increasingly blur
the boundaries between our own minds
and the technological infrastructures in
which we live, work, and play.

“Google Glasses”, expected to hit the market
within a year, may nudge us in this direction
sooner than we think …

TXM Summary

Portable (or always available/ubiquitous)

Robust.

and

Dovetailed (co-adapted)

Augmentations

“PRaDA accessories become you”
34

Some Worries and Replies

Adams and Aizawa (2008) find TXM
‘outrageous’ and ‘preposterous’ (p.vii).

Whatever plausibility it has, they suggest, it
gets by cheating.

Adams, F and Aizawa, K (2008) The Bounds of Cognition
(Blackwell)

First, it relies on a fuzzy, untriangulated notion of ‘cognition’.
We gave no ‘mark of the cognitive’, so how can we tell
where the machinery of cognition lies?

Second, the best candidate for such a mark involves non-
derived contents and they are all said to be found only ‘in the
head’.

Third, there are characteristic properties that the in-the-head
stuff displays that the rest doesn’t, so we can’t (even
bracketing non-derived content) run a functional-sameness
argument here.

So how come anyone is even tempted?

Only thanks (A and A suggest) to:

1. The error of mistaking (mere) causal
coupling for something more profound, more
‘constitutive’.

= rather like mistaking the inputs to a
calculator for part of the machinery that
calculates

and/or 2.

The error of confusing the cognitive process
with the cognitive system

the latter may include (inner and outer) parts
and processes that aid and abet cognition,
without themselves participating in true
cognitive processing.

(= like mistaking the calculator’s casing or
batteries for part of the calculating engine).

Concerning the mark of the cognitive

A and A suggest, as a plausible ‘mark of the cognitive’ the
presence of “non-derived representations governed by
idiosyncratic kinds of processes” (p.10).

The kinds of inscription found in e.g. some online storage fail
to make the grade on both counts.

They involve derived (that is, in some sense humanly
assigned) meanings.

And they do not behave in the same ways as their in-the-
head counterparts (for example, they fail to display various
well-known psychological effects, such as the recency effect
which systematically favors late entries in a list (p.63)).

But notice: non-derived representations (see Clark
(2005) for discussion) are indeed present in any
putative overall cognizing system

Even on the extended view, every extended mind will
involve some operations defined over
representations whose meanings are non-derived.

So the real question here concerns
the acceptability of derived
representations or contents as
genuine elements in a distributed or
hybrid cognitive process that quite
clearly involves many non-derived
ones too.

I don’t think we have clear intuitions
about this

(consider manipulating
Venn diagrams in the head)

What about the rest of the clause? “non-derived
representations governed by idiosyncratic
kinds of processes” (p.10).

A and A note that human biological memory
systems look to be characterized by certain
psychological laws (eg primacy, recency and
chunking effects).

But to identify cognitive candidacy by
comparison to typical human inner neural
processes threatens (see Wheeler (2008)) to be
question-begging in the context of this debate

In any case, we should reject the idea that the
surface psychological laws that happen to
characterize the inner (bio-cognitive) realm in
human agents should in any way define the
cognitive realm itself
Martian bio-memory, even
if it didn't display e.g. the
recency and chunking
effects found in human
neural memory systems,
could surely count as an
aspect of Martian cognition.

..helps reveal the real role of the Parity
Principle (from Clark and Chalmers
(1998)).

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.

What Parity Isn’t:
PP does NOT require the bio-external elements
to be operating in exactly the ‘same way’ as the
bio-internal elements.

Rather, the Parity Principle is best seen as a
demand that we assess the bio-external
contributions with the same kind of unbiased
vision that we ought to bring to bear on an
alien neural or inner organization.
It is a call not for sameness, but for
sameness of opportunity

Parity Probe =
akin to a ‘veil of metabolic ignorance’

asks what our attitude would be if
currently external means of
information storage and
transformation were found in biology.

= about avoiding a rush to judgment
based on spatial location alone.

PP is a tool that’s meant to help us deploy our pre-theoretic
grip on the cognitive without the distractions of skin and skull.

We surely do have such a grip.

It is only courtesy of such a grip that we can tell that eg the
colour or texture of the brain is not (as far as we know) a
cognitive-processing relevant feature.

PP = thus what Mark Sprevak dubs a ‘Fair Play Principle’: it
helps us avoid a rush to judgment based on the spatial location
and/or the processing idiosyncrasies of human wetware.

Indeed, avoiding human wetware chauvinism is
necessary quite close to home, if we are to allow for
e.g. the minds of cats

Suppose cat-brains turn out not to display
some of the signature features of human
memory systems?

Should we conclude that cat-memory is not
real memory?

Adams and Aizawa are alert (p.71-73) to the
worry, but their discussion is revealing…

“These observations suggest a complication in the
evaluation of the hypothesis of extended cognition.
They suggest that we cannot refute the hypothesis of
extended cognition simply on the grounds that the
combination of brain, body, and environment does not
form a conglomerate that is like a normal human
cognitive processor. The combination could have
some general, non-human, kind of cognition…that is
related to human cognition in only a “family
resemblance” kind of way.” (p.72).

But in this passage ‘like a normal human
cognitive processor’ already seems to mean
‘like a normal human in-the-head
mechanism’.

This makes the response look question-
begging.

For the challenge that the theorist of extended
cognition often means to raise to this very
identification.

What about the putative "coupling
/constitution fallacy” in arguments for the
extended mind?

= the fallacy of moving from the causal
coupling of some object or process to some
cognitive agent, to the conclusion that the
object or process is part of (helps
constitute) the agent's cognitive
processing.

"Question: Why did the pencil
think that 2+2=4?

Clark's Answer: Because it was
coupled to the
mathematician…. That about
sums up what is wrong with [ the]
extended mind hypothesis.”

From Adams and Aizawa (‘Defending the
Bounds of Cognition’ )

Question: Why did the V4 neuron ‘think’ that
there was a spiral pattern in the stimulus?
Answer: Because it was coupled to the (rest of
the) monkey.

Let’s try that again:

…..the coupling is what the V4 neuron,
whose response characteristics are such-and-
such, to in virtue
of which , in the
larger Monkey-system, is exhibited.

Unlike, say, the created in
that neuron in isolation, which wouldn’t be part
of any cognitive process at all

The Appeal to Coupling (Revisited)

Coupling is just the that
allows extended or distributed cognitive
processes to emerge, and be maintained, while
processing proceeds.

Examples:
Inter-hemisphere coupling, as in part enabled by
the corpus callosum.

Neural-bodily coupling, as between neural systems
and movements of hand and arm. See e.g. the case
of gesture, discussed at length in Clark (2007) (2008)

Neural-bodily-wordly coupling, as between neural
systems, bodily effectors, and bio-external resources
such as sketchpads,notebooks, and the web. See
e.g. discussions in Clark (2008) Supersizing the Mind

But still, I agree that not all coupling creates extended
cognitive systems…

Many things (like the weather, or a bang on the head) may
impact cognition but are not thereby parts of the cognizing
machine.

59

Thought Experiment 1

Suppose the rhythmic pulse of rain on my Edinburgh window
somehow helps the pace and sequencing of a flow of
thoughts.

Is the rain now part of my cognitive engine? Probably not.
.

60


A robot that deliberately seeks those
conditions, because it is designed to use
raindrop sounds to time, sequence, and pace
some internal operations essential to proper
cognizing.

??

61


Imagine a robot that evolved to spit
stored water at a plate on its own
body so as to use the auditory signal to
time and sequence key neural
information-processing operations.

62

Those self-maintained, self-stimulating signals are best
seen (I claim) as part of the cognitive mechanism itself. A
neural clock or oscillator would surely count after all…

Much of advanced cognition involves the deployment of
cognitive processes that create (or sometimes just elicit)
the inputs that continuously drive those and/or other
cognitive processes along (speech, sketching, writing, and
gesture, seem like prime examples of such self-created
systemic inputs).

63

In these special loop-y contexts, the simple input vs
part-of-processing distinction, with its associated ban
on counting inputs as parts of processing
mechanisms seems wrong.

= Self-stimulation as one clean route from mere inputs
to parts of mechanisms..

Compare: the car makes exhaust fumes (outputs) that are
also inputs that drive the turbo that adds power (often
around 30% more power!) to the engine.

The exhaust fumes are outputs that are also self-
created inputs that surely form a proper part of the
overall power-generating mechanism

= automotive self-stimulation!

65

Another Kind of Worry

Rob Rupert (2009) looks able to allow the spitting
robot to possess a bodily extended cognizing circuit,
but would reject the use of paper or other off-body
storage

This is because Rupert argues for a special status
for the most portable bundle of processing powers
that characterize the biological organism.

He sees this bundle as the constant target
(implicitly or explicitly) of most work in psychology
and neuroscience.

Various arguments: I’ll look just at two: asymmetry
and integration

Asymmetries
Eg (Rupert) If you destroy a
notebook, a cognizing agent may
well replace it. But destroy the
brain and that’s (literally) all she
wrote!

Or (Harry Collins) When my props and aids go wrong it is I
who have to repair them. They will never repair me.
There seems to be a deep asymmetry, or lopsidedness,
between the role of the notebook and that of the brain.

Take a small part of the neural crew, and very
often ‘I’ can survive perfectly well without it (a
neuron or two, visual cortex, MT)
Similarly, when aspects of my own bio-memory
start to become unreliable, I may deliberately shift
towards alternative means of storage and retrieval.

The apparent lopsidedness (I have to take steps
to offset the loss of my own bio-memory
functioning) does not threaten the claim that, prior
to the loss, those internal resources were
realizing my cognitive activities.
Ditto, then, for the notebooks and sketchpads…

(Sprevak) Don’t hold the external stuff
to higher standards than we’d hold
aspects of the brain’s own
functioning.

Integration

Rupert claims there are severe scientific costs to
adopting the extended perspective, as we may begin to
lose our experimental grip on the integrated bundles of
processing resources (agents) that psychology and
neuroscience seeks to study.

Sally-the organism (call that ‘O-Sally’)

O-Sally + iPhone

O-Sally +notebook

O-Sally + Tommy

Re these putative costs

I just don’t see them.

No need to lose our grip on the core biological bundle.
Any more than attention to whole brains makes us lose
track of the special contribution of the hippocampal
bundle, or of the right hemisphere bundle…

The invitation is to let a thousand flowers bloom.

If our goal is to understand what a (a socially and
technologically situated entity) can do, we’d better study the
class of systems that includes loops through the body,
artifacts, the web, other agents etc.

If the goal is to understand what the persisting biological
organism alone can do (say, by way of mathematical
reasoning) we might want to restrict the use of all non-
biological props and aids. Fingers yes, notepads no

If it is to discover the stand-alone capacities of the neural
apparatus, we might want to impede subjects from using their
fingers as counting buffers during an experiment. No fingers,
no gestures

If it is to track the contribution of a specific neural sub-
structure, we might want to use TMS to get a better grip on
that.

All these targets are both
theoretically and experimentally
viable!

TXM invites us to tackle them all,
and to do so as part of a single
interdisciplinary project of
understanding the distinctively
human mind.

A last question to ponder:

so…is all this potential change and
cognitive ‘upgrading’ a GOOD thing,
or is it a dangerous early step on the
road to some dark and ‘post-human’
future?

A common worry:

To allow all these well-fitted, transparent tools to
count as genuine aspects of OURSELVES is to
lose sight of our essential humanity.

It is to risk a kind of bodily, sensory, and cognitive
dissolution, as we slowly but surely lose track of
where WE stop and the world of tools and
technologies around us begins.

= a kind of personal dissolution into the bio-
technological matrix..

A kind of bodily, sensory, and cognitive BLOAT

Keith Butler tries to stop the bloat by appeal to a notion of the
biological brain as ultimate controller

“Even if external elements sometimes participate in processes
of control and choice ( your software agent might choose
some stocks and shares, and so on) still it is always the
biological brain that has the final say”

So the brain is the controller and chooser of actions in a way
all that external stuff is not.

So the external stuff should not count as part of the real
cognitive system. See eg Butler (1998), see also Adams and
Aizawa (2002, 2008)
81

But I am not convinced.

Re-applying the “locus of control” criterion inside the head
helps reveal what’s going wrong.

Do we now count as not part of my mind or myself any
neural subsystems that are not the ultimate arbiters of
action and choice?

Suppose only my frontal lobes have the final say- does that
shrink the “real mind” to just the frontal lobes!?

What if no subsystem has the ‘final say (Dennett)?

Has the mind and self just disappeared?
82

It is a mistake to think that all those
“cognitive tools” need some kind of
wafer-thin user…

This is where the ghost of Descartes
seeps out from under the contemporary
materialist rug

83

I think, though, that we should
be MUCH more worried by the
alternative, which is a kind of
unprincipled shrinkage of the
mind and self!

Brainbound’s Last Stand?

Brie Gertler (2007) has argued for what she calls ‘the
narrow mind’ (TNM)

According to TNM, the realm of the mental consists only
of the contents of occurrent conscious, processing.

This allows her to reject the arguments for TXM by e.g.
rejecting standing beliefs (classing them as not ‘mental’)
hence sidestepping the parity considerations.

If only what is active and conscious here and now is
mental, then the physical base of mind (thus reduced)
plausibly does shrink back to well within the bounds of
skin and skull….

But restricting the mental/cognitive to the
occurrent and conscious is a drastic step

It renders huge swathes of crucial in-head
processing non-mental.

Do we really want to avoid cognitive ‘bloat’ at the
cost of shrinking the mind so dramatically?

This seems scientifically unwarranted and
ethically dubious…

A Closing Story: Deacon
Patrick Jones

Jones suffers severe
memory impairments as
a result of repeated
traumatic brain injury.

Yet he lives a surprisingly
normal life as a working
catholic deacon in
Colorado Springs.

This is not due to any
super hi-tech interventions.

Jones relies upon a combination of the popular
software Evernote, a Mac program for visualization
called Curio, and an iPhone.

Courtesy of these off-the-shelf packages and
devices Jones is able to create massive webs of
interlinked notes and pointers that allow the
saving, searching, retrieving, and diagramming of his
own contacts, thoughts, meetings, decisions,
and interactions.

See “What if HM had a Blackberry?” Gary Marcus,
Psychology Today, December 2008

Amazingly, it is only in virtue of this whole up-and-
running web of structure that he able to recall who
he has spoken with, what was decided, and so
on.

Yet he carries through complex long-term projects of
pastoral care with incredible skill, optimism, and
good humour.

Patrick’s mental life is now built (it seems to me)
upon a foundation of both biological and non-
biological processing and storage.
If you were to hack into and destroy his EVERNOTE
records, that would be a crime against the person,
not merely a crime against his cyber-property.

It would be tantamount, as Dan Dennett once
commented, to inflicting brain damage on someone
while they sleep.

Issues of ownership and legal protection must
soon loom here.

Do Patrick’s software providers have the right to
delete his records if he fails to keep up
payments?

Do they have the right to cease to support old
software, even if it has become deeply dovetailed
with an ageing human’s biological brain?

What if Patrick and his spouse create a shared
resource then split up?

Issues like these will surely arise as our
cognitive technologies grow better and better,
and the ongoing dovetailing of brains and
technologies becomes more and more
pronounced.

Our laws, educational practice, and social
policy need to plan for a near-future in which
individual minds are web-extended,
technology-permeated artifacts, apt for all
kinds of transformation, repair, extension,
and enhancement

Maybe the best way to do so
is start by recognizing that it’s
cognitive technologies all the
way down….

Extending the Mind with Cognitive Prosthetics?

More Related Content

What's hot (20)

Viewers also liked (6)

Similar to Extending the Mind with Cognitive Prosthetics? (20)

More from PhiloWeb (20)

Recently uploaded (20)

Extending the Mind with Cognitive Prosthetics?