The Meaning Code | Mark Solms and Michael Levin on the Attempt to Build and Test an Artificial Consciousness @TheMeaningCode | Uploaded July 2022 | Updated October 2024, 4 hours ago.
For a complete intro of these two guests, see their previous conversation.
youtu.be/2vYfcotRRjk
Timestamps
00:00 Musical Intro
Affect is fundamental
Unless we can make a conscious machine, we have not solved the hard problem
Cognitive systems as predictive of the outside world
02:00 Mark Solms describes his project to create an artificial consciousness
04:15. The quest to find a neural correlate of consciousness (Francis Crick)
05:00 The wrong place to start
06:25 Children born without a cerebral cortex are still responsive and reactive
10:00 The mechanism of homeostasis, the periaqueductal gray, error signals
11:00 The architecture of the artificial consciousness
11:50 Autonomic regulation can be unconscious
12:30 Uncertainty triggers consciousness (ex: blood gases rising to conscious reaction),
Uncertainty tells you how well you’re doing
15:00 Minimal conditions -
self-organizing system - registers its own Markov blanket state (point of view)
3 needs
17:30 It needs energy, it suffers damage w/in tolerable limits, it needs rest
& repair of structural damage - only purpose is to keep going
19:30 It wanders about is environment exploring and experiencing
20:43 An active inference agent, inferring its own state and state of its environment
Error signals, updates, improving its model of itself in the world based on its
3 hemeostats
Statistical confidence
23:50 Qualitatively distinctive categories
24:30 Calculating its expected free energy - the ability to look ahead
25:00 Equivalent of working memory, able to make decisions about how many steps needed to think ahead.
25:50 Create uncertainty by changing the environment. Its confidence in its policies is tethered to its affect. Environment changes & returns. Seasons.
Explore and Exploit
28:20 How do we know when such a system has feelings?
30:20 Non-conscious is how we look at things that are in a different space than we are, but
are they really non-conscious? Levels of “selfness”
32:40 Competencies in non-3D, problem spaces, transcriptional spaces, etc. In your virtual world is that level of consciousness as valid as physical organisms exhibit in 3-D space?
36:50 Measuring its own state in a qualitative sense
Virtual v. Real agents
Substrate v. Mechanism
What new problems arise with embodiment?
41:00 Anomaly. Navigating morphospace
Stress
Changing set point. Rewritten the set point they gauge where to go.
Second order monitoring system (metacognition)
43:30 We interact with our own representation of that state and our own internal bodily state
The nerve impulses represent the state of a reality
45:30 The thermostat makes no resistance to changing the setpoint, but biological organisms need to resist, to monitor the set point.
learning v being trained
Has your set point been changed? Has someone else changed it?
47:40 The survival need of resisting change to your set point
50:00 Introduce other agents into the same environment to infer
other agents. Isolation is a biologically implausible situation.
51:30. The problem of other minds
55:50 Zebrafish with hedonic condition place preference research
See whether what motivates the entity is only a feeling.
57:30 Anxiolytics
58:50 Where does the desire to exist come from?
"Affect is the fundamental (elemental) aspect of consciousness." Mark Solms
There is an attempt to recast physics "to use the dynamics we use to look @ cognitive systems - a specific kind o predictive interaction with the outside world." Michael Levin
"The ultimate test of a mechanistic consciousness is to engineer the mechanism... as Richard Feynman said, 'If I can't create it, I don't understand it.'" Mark Solms
Raw feelings. A feeling state with survival consequences
The mechanism of feeling is primarily homeostatic
Periaqueductal Grey - error signals converge there
Trying to develop a functional architecture of the PAG for an artificial consciousness.
Choice making. Choices must be rooted in a value system.
Minimal requirements for an artificial consciousness:
Self-organizing
Bounded
Registers its own state
Mark posits that these three requirements are necessary for a point of view. They establish a Markov boundary
If it's doing badly, it will cease to exist
It needs to improve its model of itself in the world, to increase confidence in its greater or less precision to its present policies.
The system makes decisions has needs
Free energy is average error
To minimize free energy, we must maximize the efficiency of its predictive model
In your virtual simulated world, is it as valid as what physical creatures have? Can you have a consciousness that is as valid?
A toxic environment creates an imbalance in blood gases -- air hunger that rises to the conscious level. Anomaly
I require a metacognitive monitor to determine the origin of whatever might be messing with me.
Are you resisting changes to your set point?
For a complete intro of these two guests, see their previous conversation.
youtu.be/2vYfcotRRjk
Timestamps
00:00 Musical Intro
Affect is fundamental
Unless we can make a conscious machine, we have not solved the hard problem
Cognitive systems as predictive of the outside world
02:00 Mark Solms describes his project to create an artificial consciousness
04:15. The quest to find a neural correlate of consciousness (Francis Crick)
05:00 The wrong place to start
06:25 Children born without a cerebral cortex are still responsive and reactive
10:00 The mechanism of homeostasis, the periaqueductal gray, error signals
11:00 The architecture of the artificial consciousness
11:50 Autonomic regulation can be unconscious
12:30 Uncertainty triggers consciousness (ex: blood gases rising to conscious reaction),
Uncertainty tells you how well you’re doing
15:00 Minimal conditions -
self-organizing system - registers its own Markov blanket state (point of view)
3 needs
17:30 It needs energy, it suffers damage w/in tolerable limits, it needs rest
& repair of structural damage - only purpose is to keep going
19:30 It wanders about is environment exploring and experiencing
20:43 An active inference agent, inferring its own state and state of its environment
Error signals, updates, improving its model of itself in the world based on its
3 hemeostats
Statistical confidence
23:50 Qualitatively distinctive categories
24:30 Calculating its expected free energy - the ability to look ahead
25:00 Equivalent of working memory, able to make decisions about how many steps needed to think ahead.
25:50 Create uncertainty by changing the environment. Its confidence in its policies is tethered to its affect. Environment changes & returns. Seasons.
Explore and Exploit
28:20 How do we know when such a system has feelings?
30:20 Non-conscious is how we look at things that are in a different space than we are, but
are they really non-conscious? Levels of “selfness”
32:40 Competencies in non-3D, problem spaces, transcriptional spaces, etc. In your virtual world is that level of consciousness as valid as physical organisms exhibit in 3-D space?
36:50 Measuring its own state in a qualitative sense
Virtual v. Real agents
Substrate v. Mechanism
What new problems arise with embodiment?
41:00 Anomaly. Navigating morphospace
Stress
Changing set point. Rewritten the set point they gauge where to go.
Second order monitoring system (metacognition)
43:30 We interact with our own representation of that state and our own internal bodily state
The nerve impulses represent the state of a reality
45:30 The thermostat makes no resistance to changing the setpoint, but biological organisms need to resist, to monitor the set point.
learning v being trained
Has your set point been changed? Has someone else changed it?
47:40 The survival need of resisting change to your set point
50:00 Introduce other agents into the same environment to infer
other agents. Isolation is a biologically implausible situation.
51:30. The problem of other minds
55:50 Zebrafish with hedonic condition place preference research
See whether what motivates the entity is only a feeling.
57:30 Anxiolytics
58:50 Where does the desire to exist come from?
"Affect is the fundamental (elemental) aspect of consciousness." Mark Solms
There is an attempt to recast physics "to use the dynamics we use to look @ cognitive systems - a specific kind o predictive interaction with the outside world." Michael Levin
"The ultimate test of a mechanistic consciousness is to engineer the mechanism... as Richard Feynman said, 'If I can't create it, I don't understand it.'" Mark Solms
Raw feelings. A feeling state with survival consequences
The mechanism of feeling is primarily homeostatic
Periaqueductal Grey - error signals converge there
Trying to develop a functional architecture of the PAG for an artificial consciousness.
Choice making. Choices must be rooted in a value system.
Minimal requirements for an artificial consciousness:
Self-organizing
Bounded
Registers its own state
Mark posits that these three requirements are necessary for a point of view. They establish a Markov boundary
If it's doing badly, it will cease to exist
It needs to improve its model of itself in the world, to increase confidence in its greater or less precision to its present policies.
The system makes decisions has needs
Free energy is average error
To minimize free energy, we must maximize the efficiency of its predictive model
In your virtual simulated world, is it as valid as what physical creatures have? Can you have a consciousness that is as valid?
A toxic environment creates an imbalance in blood gases -- air hunger that rises to the conscious level. Anomaly
I require a metacognitive monitor to determine the origin of whatever might be messing with me.
Are you resisting changes to your set point?
