The quest to create artificial general intelligence has largely followed a “brain in a vat” approach, aiming to build a disembodied mind that can carry out the kinds of logical reasoning and inference that humans are capable of, usually demonstrated through language. This approach may some day pay off, but it’s not how nature did it. Intelligence did not evolve to solve abstract problems – it evolved to adaptively control behaviour in the real world. Living organisms are agents that can act, for their own reasons, in pursuit of their own goals – most fundamentally, to persist as a self through time. By charting the evolution of agency, we can see the origins of action and the concomitant emergence of behavioural control systems; the transition from pragmatic perception-action couplings to more and more internalised semantic representations; and, on our lineage, a trajectory of increasing cognitive depth and ever more sophisticated mapping and modelling of the world and the self. The resultant accumulation of causal knowledge grants the ability to simulate more complex scenarios, to predict and plan over longer timeframes, to optimise over more competing goals at once, and ultimately to exercise conscious rational control over behaviour. In this way, intelligent entities – agents – evolved, with greater and greater autonomy, flexibility, and causal power in the world. To realise intelligence in artificial systems, it may similarly be necessary to develop embodied, situated agents, with meaning and understanding grounded in relation to real-world goals, actions, and consequences.
4. Intelligent is as intelligent does
Intelligence is often equated with the capacity for
abstract, logical reasoning
But cognition evolved for controlling behavior and
solving problems in the real world
Understanding causal relations, anticipating,
simulating, predicting outcomes of action
Intimately linked with the evolution of agency
5. The Evolution of Agency
Metaphysics
How can physical
systems that obey
physical laws exercise
causation in the world?
Implications for free will,
moral responsibility
Biology
How do organisms
actually exercise choice
and control?
Where does that causal
power come from?
How is it supported by
cognition?
6.
7. What is Life?
An improbable pattern of matter –
resistant to the 2nd Law of
Thermodynamics:
• Self-sustaining
• Self-replicating
• Self-organising
9. Life in chemistry
x, y, z…
+ energy
self-sustaining system
(dynamic stability)
x, y, z…
+ energy
A
B
10. From geochemistry to biochemistry
Alkaline hydrothermal vents
- proton gradients
- stable microenvironments
- Fe, S chemistry
(Bill Martin, 2011)
11. A
B
C
D
E
ATP
ADP
H+
ATP
ADP
H+
Fe, Ni, S
pH ~9
pH ~6
CO2 H2
ATP
synthase
A
B
C
D
E
ATP
ADP
H+
F
H+
lipid
membrane
proton
pump
microcompartment
alkaline hydrothermal vent
From geochemistry to biochemistry
15. DNA: a stable template
ATP
ADP
H+
A
B
C
D
E
F
H+
RNA
RNA
enzyme
proteins
catalysis
translation
metabolism
replication
ATP
ADP
H+
A
B
C
D
E
F
H+
RNA
RNA
enzyme
proteins
catalysis
translation
metabolism
transcription
DNA
stable
reference
template
allows robust maintenance of dynamical pattern,
replication and recapitulation (i.e., reproduction),
and stable substrate for evolution (saves progress)
self-sustaining system
(but precarious, not
stably evolvable)
invention
of DNA
16. Life goes on…
Selection for persistence:
• Things that have the property of tending to
persist, persist
• Things that don’t, don’t
• Things that out-persist other things come to
predominate, especially if that persistence is
achieved through replication
This is the origin of purpose
18. Doing things for reasons
A
B
C
D
E F
G
FOOD
(x, y, z)
DANGER
(high temp,
low pH)
Information
- internal signal is
about x, y, z
Value
- relative to goal of
persisting
Meaning
- approach or avoid
Avoidance Approach
19. The origins of meaning
The signals now mean something for the organism
Physically correlated with something in environment
Have value relative to its “aim” of persisting
Given through feedback from natural selection
Meaning is distributed and temporally extended
Not semantic – system doesn’t understand anything
Pragmatic consequence: approach or avoidance
Terrence Deacon
Alicia Juarrero
21. Reductive vs holistic views
X
X
Y
Z
osmolarity
integration
of multiple
signals
feedback à
temporal integration,
gain modulation
temperature
metabolic
state
cell density
An overly reductive, mechanistic, linear, passive, stimulus-response model
An integrative, contextual, holistic, endogenously active, information-accommodating model
22. The exercise of agency
It’s not passively being pushed around by things in the
environment
It’s not being pushed around by its own parts either
It’s actively seeking out and taking in information
Holistically integrating it in context of its own state, its
history, and other signals from the environment, in
order to regulate its own behaviour
(Potter and Mitchell, “Naturalising Agent Causation”, Entropy, 2022)
23.
24. From pragmatic couplings to representations
Pragmatic couplings are good for simple reflexes in
response to single signals
But limited for integration of multiple signals
Need to convey these to another layer where they
can be integrated without directly causing an action
- That internalised signal is a representation
- And that process of integration is cognition
33. C. elegans: pre-wired circuitry – configured for simple
responses, integrated with internal state
Capable of simple associative learning
Now not just configured to enact natural selection’s reasons
An individual animal can reconfigure based on experience to
enact its own reasons
This happens on vastly faster timescale than evolutionary
adaptation
Associative learning
34. Historicity
System carries a memory of previous activity
- embodied in physical changes in synaptic connectivity
These set new criteria for action
They recalibrate the value attached to various mental states
This happens across evolution → innate meaning and value
(purpose is to stay alive and breed)
And it happens across life → learned meaning and value
(purpose is various short-term and long-term goals)
Causation in living systems is temporally extended
35. Simple cognition – but not thinking about much
Very simple sensorium – responds to chemicals
or direct touch – local, immediate stimuli
Inhabit the here and now
Cognitive depth
36. Olfaction and touch – things have to be in contact
Vision and hearing – don’t respond to the things
directly, respond to disturbances in the medium
- vibrations of air or water
- refractions of electromagnetic waves
Lets organisms infer the existence of objects out in
the world
Perception at a distance
40. Hypothalamus
(Cisek 2019)
Monitors physiological state
Regulates physiological homeostasis
Provides motivating signals (drives) for behavioral
homeostasis
Forebrain structures (neocortex and basal ganglia)
evolved to integrate drives with accumulated
knowledge and systems of action simulation,
evaluation, and selection
45. Internalised representations
Patterns that stand in exploitable relation to things
in the world (or in the organism)
Semantic content – (about something)
Meaning grounded by experience, web of context
Decoupled from obligate action, but used to
inform action (for something)
Disentangled, compositional – can be recombined,
drawn on, simulated, operated over
52. Intelligence and Agency
Cognition emerges to better control behaviour
More sophisticated cognition leads to better anticipation,
more accurate predictions, ability to optimise over more
goals and over longer timeframes
Meaning of internal states anchored by experience, in
relation to goals or “master functions”
Understanding is egocentric, action-oriented, causal,
-> enables simulation, counterfactual reasoning
Eventually in humans -> truly abstract reasoning
53. Artificial intelligence may require agency
May need to create an embodied self
With:
- continuity through time
- ability to behave and learn
- reasons to care
Being intelligent means
acting intelligently
54. Princeton University Press, 2018 Princeton University Press, 2023
AGENTS
How Life Evolved the
Power to Choose
KEVIN J. MITCHELL