26/09/2013
1
A Computational Approach to the Discovery of the
Correlates of Consciousness
DAVID G AMEZ
Wiston House Consciousness Retreat,
10th September 2013
Sackler Centre for Consciousness Science / Department of Informatics, University of Sussex, UK
Talk Overview
Scientific study of the correlates of consciousness.
Informational correlates of consciousness, including information integration.
Physical correlates of consciousness.
Computational discovery of the correlates of consciousness.
CORRELATES of Consciousness
Correlates of Consciousness
Not interested in thought experiments or theories about consciousness.
Want to make as few metaphysical commitments as possible.
Gather data about the relationship between consciousness and the physical world.
Develop a mathematical theory of consciousness that can make testable predictions.
Experiments on the Correlates of Consciousness
Correlates of Consciousness
This talk will discuss two types of correlate:
Informational
Physical
I think that computational correlates of consciousness are unworkable – contact me for a paper if you are interested.
26/09/2013
2
INFORMATIONAL Correlates of Consciousness
Information Correlates of Consciousness Tononi (2008) proposes that information
patterns are consciousness.
Will focus on more moderate claim that some information patterns could be correlated with consciousness.
Tononi’s algorithms have factorial dependencies and might not identify the informational correlates.
Will focus on the possibility that some pattern of information is correlated with consciousness.
Information Correlates of Consciousness Key claim: Information patterns are correlated
with consciousness independently of the physical structures that happen to instantiate them at a particular point in time (neurons, electromagnetic waves, etc.).
If appropriate information patterns occurred in a computer, the Internet or traffic lights, they would be associated with conscious as well.
Definition of Information To search for informational correlates of
consciousness we need a clear definition of information.
Floridi’s (2009) work is a sensible place to start. He distinguishes between:
Dedomena. The physical world prior to human measurement.
Data. Differences in the physical world measured at a level of abstraction.
Information. Meaningful data.
Definition of Information
Identifying meaningful data in the brain is difficult – representations?
Data is a better place to start.
Tononi’s information integration theory of consciousness is actually a data integration theory of consciousness (Gamez 2011).
Level of Abstraction
Data is measured at a level of abstraction (Floridi 2008).
In the brain we need to define:
What is measured (neurons, glia, EM waves etc.)
Spatial and temporal scale.
Units of measurement.
These are subjective choices; consciousness is an objective observable.
To minimize subjectivity need to work across multiple (and possibly all) levels.
Tononi (2010) discusses this point.
26/09/2013
3
Cross-level Maximization Potential Issues
A complete cross-level analysis cannot be done:
The number of levels of abstraction of a real world system is probably infinite.
The lower levels of the system (molecules, atoms, quarks) will be inaccessible for the foreseeable future.
The results at each level might contradict each other.
Data Pattern or Physical Correlate?
A cross-level algorithm’s output could be consistently correlated with consciousness at one level of abstraction, for example the neuron level.
Two possibilities:
Correlation between data patterns and consciousness
Correlation between neural patterns and consciousness
How do we decide between these two possibilities?
Data Pattern or Physical Correlate?
Could show that the data pattern was independent of the substrate if:
One level correlated with consciousness at one time.
A different level correlated with consciousness at another time, either in the same or different subjects.
Could then be claimed that there are data correlates of consciousness, and not just physical correlates at one particular level.
Proof that there are Data Correlates of Consciousness
Data Correlates of Consciousness
Could only produce a conclusive result if the data pattern that was correlated with consciousness was found to naturally shift between levels of abstraction.
Otherwise it is going to be impossible to prove that data patterns are correlates of consciousness.
26/09/2013
4
PHYSICAL Correlates of Consciousness
Physical Correlates of Consciousness
Consciousness could be correlated with a specific pattern in a particular physical structure.
Some potential physical correlates of consciousness:
Neural synchronization
Electromagnetic waves
Quantum effects
Connection patterns
Activation in particular brain areas.
Physical Correlates of Consciousness
Want a mathematical description of the physical correlates that can make testable predictions.
Tononi (2008) has attempted to write down the mathematics.
I favour a computational approach to the discovery of the correlates of consciousness.
Example
Hypothesis: conscious colours and sounds could be correlated with the information-processing carried out by the eyes and ears.
I have a recent paper on this – available on request.
Structuring of Information by the Senses
Cortex Trained by Structured Information
26/09/2013
5
Conscious Sensations Correlated with Learnt Patterns
Conscious Sensations Correlated with Learnt Patterns
Activation of Trained Brain Areas in the Absence of Stimulation
Testing the Hypothesis
Hypothesis: conscious colours and sounds could be correlated with the information-processing carried out by the eyes and ears.
General procedure:
Carry out experiments to identify which patterns are correlated with which conscious sensations.
Look for concise mathematical descriptions of the patterns.
Use mathematics to generate testable predictions.
Identify Correlations Identify Correlations
26/09/2013
6
Record Brain Activity Mathematical Description of Brain Activity
Generate Prediction about Consciousness
Compare with First-Person Reports
Testing the Hypothesis
The mapping of brain activity onto a set of mathematical equations plays a key part in this process.
A human could look at brain activity and come up with a mathematical description.
But ...
Enormous quantity of data
Humans have limited working memory, limited mathematical ability, limited imagination, etc.
If a simple mathematical description does not exist, the human will fail.
Computational Scientific Discovery
Could use computers to generate a mathematical description of the patterns.
Examples of computational discovery of scientific knowledge:
Adam and Eve - automated discovery of yeast genomics (Sparkes et al. 2010).
Inference of differential equations from mechanical data (Schmidt and Lipson 2009).
Summaries of earlier work in Dzeroski and Todorovski (2007).
26/09/2013
7
Adam and Eve
Sparkes et al (2010)
Inferring Natural Laws from Experimental Data
Schmidt and Lipson (2009)
Computational Discovery of the Correlates of Consciousness Use machine learning to identify brain activity
patterns that are correlated with different conscious sensations.
These could be expressed as differential equations or another formalism.
Use equations to generate predictions that can be tested by further experiments.
Multiple levels could be analysed simultaneously.
CONCLUSIONS
Conclusions
Likely to be impossible to prove that information patterns (or information integration) are correlated with consciousness.
Claims about computational correlates of consciousness are untestable.
Might be possible to scientifically prove that consciousness is correlated with a spatiotemporal pattern in a particular physical structure.
Conclusions
Physical correlates of consciousness could be very mathematically complex (possibly thousands of pages of differential equations).
Could be embedded in a vast amount of data.
Human scientists have limited working memory, mathematical ability, etc.
26/09/2013
8
Conclusions
More likely to make progress if we use a machine learning approach to identify the physical correlates of consciousness.
Some work has already been done on the computational discovery of scientific knowledge.
New area with many challenges!
Science is likely to go in this direction in the future.
Acknowledgements
Many thanks to Barry Cooper and the John Templeton Foundation for supporting this work (Project ID 15619: 'Mind, Mechanism and Mathematics: Turing Centenary Research Project').
I would also like to thank the Sackler Centre for Consciousness Science at the University of Sussex for hosting me as a Research Fellow during this project.
References Dzeroski, S. and Todorovski, L. (2007). Computational Discovery of Scientific Knowledge :
Introduction, Techniques, and Applications in Environmental and Life Sciences. Berlin, Springer.
Floridi, L. (2008). The Method of Levels of Abstraction. Minds and Machines 18(3): 303-329.
Floridi, L. (2009). Philosophical Conceptions of Information. Lecture Notes in Computer Science 5363: 13-53.
Gamez, D. (2011). Information and Consciousness. Etica & Politica / Ethics & Politics, XIII(2): 215-234.
Humphreys, P. (2004). Extending Ourselves : Computational Science, Empiricism, and Scientific Method. Oxford ; New York, Oxford University Press.
Schmidt, M. and Lipson, H. (2009). Distilling Free-Form Natural Laws from Experimental Data. Science 324(5923): 81-85.
Sparkes, A., Aubrey, W., Byrne, E., Clare, A., Khan, M. N., Liakata, M., Markham, M., Rowland, J., Soldatova, L. N., Whelan, K. E., Young, M. and King, R. D. (2010). Towards Robot Scientists for Autonomous Scientific Discovery. Automated Experimentation 2: 1.
Tononi, G. (2008). Consciousness as Integrated Information: A Provisional Manifesto. Biol Bull 215(3): 216-242.
Tononi, G. (2010). Information Integration: Its Relevance to Brain Function and Consciousness. Arch Ital Biol 148(3): 299-322.
More Information
This talk is based on several papers that I am preparing for publication - can send copies on request.
Slides:
www.davidgamez.eu/talks/
Other papers related to this material:
www.davidgamez.eu/publications/
Feel free to contact me if you have any questions:
Top Related