To me, there is a huge pressing need in the world to help people communicate more effectively and it is not being addressed adequately. The foundation for this requires systems that enable introspection. When people argue or misunderstand each other, often it is like they are just shuffling around these large immutable “lego blocks” of concepts. They can’t break these things down into their much smaller atomic components and thus they cannot construct what another person, who has a different set of blocks in a different atomic arrangement, has constructed.

Making models of how people think isn’t new, but there are newer developments in machine learning that give us much more traction on handling one component of this problem: modeling subsymbolic representations.

There are two fundamental types of reasoning: symbolic and subsymbolic. The words you read here are symbols, immutable bits of information. What they mean to you, however, is subsymbolic. A car is a car, what thoughts are invoked in your mind by the perception of this symbol is much different than mine. The reason why people are impacted so differently by the same piece of text, art, etc, is due to the vastly different subsymbolic representations we have for the symbols we use. Advances in neural networks and natural language processing have given us much better tools of modeling the subsymbolic component to the perception of language – which has been the missing piece of the puzzle.

About five years ago I started KindVoice, a platform for peer mentorship and support. People post here when they are looking for guidance or when they are looking to help people. Since its inception, we’ve connected thousands of people and I’ve gotten some of the most heartfelt feedback from people that were helped by the system and by people that were at a point in their lives where it seemed like the world was so dark and cold, but they came across our beacon of good will towards mankind in that darkness and that gave them hope to keep moving forward and not give up on life.

I learned here that the effectiveness of peer support is based on two things:

  1. A human connection was the primary source of healing. Just having someone there to listen and show some semblance of care by doing so, gave people a better feeling about the world.
  2. The ability of a peer to guide someone through an introspective processes and help repair the points in one’s philosophy that was amplifying bad thoughts.

It lies in this second point where I believe machine learning can be brought in.

Once we can learn the effective introspective processes, considering joint introspection (i.e. conflict resolution) becomes a more straightforward problem – although of course not trivial.