HUMANOBS

1
HUMANOBS
Toward a programming paradigm for control systems
with high levels of existential autonomy
Eric Nivel, Kristinn R. Thorisson Reykjavik
University / Icelandic Institute for Intelligent
Machines
AGI 2013 - Beijing August 2013
2
Overview
gt A G I domain independence ? scalability (in
terms of complexity) gt Natural intelligence vs
manageable complexity gt Constructivist approach
delegate to the system (some of) its own
construction gt Used for S1, an AERA based
system, that learns socio- Communicative skills
by observing people (HUMANOBS EU-funded FP7
project)
3
Overview
gt Main characteristics of Replicode - Based on a
non-axiomatic logic real-valued and time
constrained. - Data-driven, computation based on
pattern-matching. - Stateless executable code
(models, composite states, programs). - No
explicit ifs, loops, OR, AND. - All executable
code run concurrently massively parallel - Code
can be an input for some other code. - Code can
be active or inactive.
4
Overview

• gt Main characteristics of Replicode
• - A kind of functional language, LISP-like
  syntax, not typed.
• - Data can be salient or not.
• - Data and code (objects) have a limited life
  time (resilience).
• Once produced, objects cannot be modified.
• Objects live in workspaces (groups), possibly in
  several of them.
• Groups control the lifecycle and state of
  objects at various frequencies.
• - Extensible set of operators hooks for custom
  C code (icpp_pgm).

5
Dynamic Model Hierarchy
gt A model captures a causal relationship
(deduction, i.e. prediction). gt A model can thus
be used to perform abductions (ex
sub-goaling). gt A model is bidirectional and
performs deductions and abductions concurrently.
A(X,Z,W) B(X,Y,Z)
M0
Forward a ? predicted b and ? iM0(x,z,w,y) Y
can be a function of X, Z and W. Backward goal b
? sub-goal a W can be a function of X, Y and
Z Predicted b and sub-goal a are monitored,
success of M0 is assessed in due time
6
Dynamic Model Hierarchy
gt Indirect coupling / pattern affordances.
M0 A ? B
M1 B ? C
M2 C ? D
M0
M1
M2
M3 E ? C
M4 C ? F
M5 G ? C
M3
M4
M5
Forward a ? predicted b ? predicted c ?
predicted d Backward goal d ? sub-goal c ?
sub-goal b ? sub-goal a ?
sub-goal e ? sub-goal g
7
Dynamic Model Hierarchy
gt Pre-conditions.
iM0 instance of M0, NOT M0 itself
IF M0 fires (at some time), it will succeed
M0 A ? B
M1 C ? iM0
M2
Pre-conditions (weak) - OR
M1
M2 D ? iM0
M0
M3 E ? iM0
Pre-conditions (strong) - AND
M3
M4 F ? iM0
M4
IF M0 fires (at some time), it will fail
Execution of a model, success/failure thereof ARE
REGULAR (INTERNAL) INPUTS.
8
Dynamic Model Hierarchy
gt Post-conditions.
M0 A ? B
WHEN M0 has fired (at some time)
M6
M5 iM0 ? C
M0
M6 iM0 ? D
WHEN M0 has not fired (at some time)
M5
M15
M7 iM12 ? iM0
M12
M10
M8 iM13 ? iM0
M7
Combinations
M9 iM14 ? iM0
M0
M8
M10 iM15 ? iM0
M9
M13
M14
9
Dynamic Model Hierarchy

• gt Control hierarchy
• gt Dynamic models are built/deleted dynamically.

predictions
I/O
goals
10
Dynamic Model Hierarchy
gt Dynamic models are activated accordingly to
their success rate under a threshold, no
execution. gt Inputs hold a confidence value
(saliency) under a threshold, no input. gt
pred.cfdinput.cfdmodl.sr (likelyhood for the
pred to be true). gt goal.cfdsuper-goal.cfdmodel.
sr (likelyhood for a goal to succeed).
11
Dynamic Model Hierarchy
gt Given limited resources, only the paths
consisting of the best models will be followed.
I/O
12
Dynamic Model Hierarchy
gt Drives and top-level models are hand-coded. gt
Drives are non-observable states. gt Drives
are re-generated dynamically (defined by the
programmer). gt Top-level models are given ways to
satisfy drives.
I/O
drives
top-level models
learned models
13
Dynamic Model Hierarchy
gt States atomic or composite constitute the
patterns found in models. gt States can be
composed of other states ? state hierarchy. gt
Reflectivity internal operations are reflected
as states. Composite states can encode concepts
in an operational fashion. gt In addition to a
control hierarchy (procedural) we also have a
concept (structural) hierarchy.
A(X,Z,W) B(X,Y,Z) C(X,T)
S0
A, B and C are synchronous states synchronous
hold within a common time interval
14
Dynamic Model Hierarchy
gt Indirect coupling models are not hardwired to
each other. They are coupled via events in the
workspace. gt Pattern affordance. gt Control
hierarchy affordances, pre- and
post-conditions. gt Dynamicity models are
added/deleted/activated continuouslsy. gt Concept
hierarchy states encode concepts at different
levels of specification.
15
Logic
gt Real valued confidence (not a probability) in
0,1. gt Time constrained time interval after,
before bounds in µs. gt Inference rules
govern - Deduction (forward chaining) -
Abduction (backward chaining) - Induction -
Commitment (resolution of simulation)
16
(No Transcript)