Human‐machine Coexistence in GroupsGroups

George Kampis Paul Lukowicz Stuart Anderson DFKI (German Research Institute for Artificial Intelligence)DFKI (German Research Institute for Artificial Intelligence), 

Kaiserslautern, Germany

Edinburgh University, Edinburgh, United Kingdom

26/09/2013 www.smart‐society‐project.eu

“ l l” l“classical” telemonitoring

hScoring scheme

lResults Patients like the system – Scoring is de‐personalisedPatients like the system primarily because it gave them privileged access to doctors...

Doctors orried abo t

Scoring is de personalised Contextual factors are excluded

Doctors worried about: High false positive rate Over diagnosis

Difficult to prioritize patients Brittle response even at small scale

Over treatment

“In participants with a history of admission for exacerbations

small scale

of admission for exacerbations of COPD, Telemonitoring was not effective in postponing hospital admissions orhospital admissions or reducing healthcare costs.”

Diagnosis Capturing Context Shifting Context Capturing Context Pervasive and mobile data capture

Shifting Context Individual Global systemicdata capture

Formal monitoring activity

Global, systemic Individualised

Open‐ended Social Negotiation of key features 

26/09/2013 www.smart‐society‐project.eu 5

Diagnosis Cultures Learning Cultures Domestic: concerned, inexpert

Learning More or less open loop Easy to createinexpert

Call Centre: risk averse, inexpert

Easy to create “revolving door” cases.

General Practitioner: risk averse, expert, resource constrained

Acute Care: uninvolved

26/09/2013 www.smart‐society‐project.eu 6

“ l k ”“Social Sensemaking” Individual family and carers “curate” the monitoring time Individual, family and carers curate  the monitoring time series.

Capture rich context – environment patient condition Capture rich context – environment, patient condition, … Learning algorithm Interactive supports negotiation Interactive, supports negotiation Linking cultures supporting the transfer of context Individualised Individualised

Learning communityH i t l li k Horizontal links

Transferring experience

26/09/2013 www.smart‐society‐project.eu 7

Settings Cultures Context Cultures Long‐lived Value rich (empirical

Context Rich Open ended Value rich (empirical 

work) Range of scale

Open ended Unexpected Capture is inherentlyg

Shifting and overlapping Capture is inherently hybrid

Sensors always miss ythings

Negotiated

26/09/2013 www.smart‐society‐project.eu 8

hMechanisms Semantics Mediation Semantics Tightly linked to culture/context

Mediation Value systems differ Boundary practicesculture/context

“Good enough” to understand now

Boundary practices enable inter‐working without agreement

Open‐ended Structured by cultures

Key element in crossing cultures

26/09/2013 www.smart‐society‐project.eu 9

hMechanisms Incentives Hybridity Incentives Easy to get wrong E g quality of care in

Hybridity All computation is social In particular the E.g. quality of care in 

the telemonitoring case “Keeping people out of 

In particular the evolution process

Develop more p g p phospital” is probably better

pparticipatory approaches to 

Heterogeneous Privacy harming?

evolution.

26/09/2013 www.smart‐society‐project.eu 10

ffEffects Learning Alignment Learning Opportunity to reflect Incorporate reflection

Alignment Temporary, requires repeated repair Incorporate reflection

Drives design and evolution

repeated repair Incentives encourage alignment

Linkage between values and incentives 

26/09/2013 www.smart‐society‐project.eu 11

lConclusions Focus on Hybridity and Diversity Focus on Hybridity and Diversity Hybridity drives the programming model…Di it d i th d t d l Diversity drives the data model…

26/09/2013 www.smart‐society‐project.eu 12