Machine Learning for Morality by Using Integrated Information Theory

An experience is a conceptual structure specified by a complex of elements in a state. Specifically, the form of the conceptual structure in cause-effect space completely specifies the quality of the experience. A conceptual structure C can be plotted as a constellation of points in cause-effect space. If only we could find a way to translate points from cause-effect space into pixels or conversely, implement machine learning across 2 x 2^N axes, this would then allow the search for moral truth to commence. I do not say this is technically possible, this consideration is merely meant to illustrate that some level of moral realism may be achieved in principle.

Say there’s experience 4 and your brain recognizes this as experience 4. 4 is the experience of the guilt of failure with a particular valence tone and a particular snow on the ground and a particular sound of a truck. However, your brain can just recognize it as experience 4, existing on a continuum where falling in love ranks higher, death of a dear sibling ranks lower, and other similar feelings of guilt are also 4. {Relativists, please appreciate how automatically brains do this.} I mean, there are versions of guilt of failure where there is a white room and red flowers, and one where there is a grey sky and smaller stature, and yet the brain still recognizes these otherwise different experiences as guilt of failure – as a 4. So the specific constellation of points in cause-effect space can be very different from one experience/conceptual structure to the next. The particular set of neurons that are firing in one guilty moment’s brain are quite different from those firing in another guilty moment’s brain, but something in that mutual, evolved cortex of ours resolves these as representing the same idea – while at the same time recognizing other experiences as their own distinct value number.

As we know, with machine learning, you can convert an array of pixels with greyscale values for a sloppy handwritten number into a single, actual and concrete number. Similarly, we may be able to input varied constellations of points in cause-effect space and output an assessment of the concrete ethical value corresponding to that constellation and those like it. All we would need is a neural network that can do its “learning” beyond 2-d inputs.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s