Extended abstract
In an ongoing research project we work with the ordered set E = {“observations”, “explanations”, “knowledge”, “theories”, “sciences”}. Any inductive reasoning relies upon observations and explanations. In our research we use the following template for understanding what inductive reasoning is:
Inductive reasoning is made by experiencing subjects. The “atoms” of inductive reasoning are observations and the observations are about some mental contents. This is to say that the mental content has two fundamental layers. The second layer is the observation whereas the first layer is about non-mental contents. So, when we move with observations of observations, we are on the second layer. The first layer concerns consciousness about our five senses in addition to bodily sensations. Ontogenetically, therefore, the first observation must be about non-mental contents.
This understanding of inductive reasoning may not be that non-traditional. However, we also see observations as explanations. This allows us to equal observations and explanations so that we can build higher level theories and explanations. The ontogenetically very first observation, however, is unconventional since it (also) is an explanation. So, there are now two gaps. The first one is the ordinary gap between neurological activity and the conscious contents. The other gap is between a conscious contents and the awareness of it.
The template poses two problems for the possibility of machines’ inductive learning given that inductive reasoning and learning both are founded upon observations.
We can exemplify the template by considering two adjoined cases. First we have the biological organism. Think of a cow’s newborn calf that can walk. Now think of a student studying in a different culture and that learns the social codes by experience. We can now join the cases into a human infant.
Think of an infant turning for the first time without knowing how to turn. Now think of the same infant later in its development doing the same move but now after an intention to turn. In the first case the infant evidently has learned to turn but without learning it via observations. In the second case the infant has observed its movements and has tried to repeat them (or at least to turn). In the second case, therefore, the infant has used inductive learning.
Conclusions. The first issue concerning machine learning is if machines experience anything. If we look at a robot vacuum cleaner there is no doubt that it can adapt to the environment. If that is learning machines can learn. Here we can see the first issue as a matter of convention. The other issue, however, cannot be solved in this way. If inductive learning takes observations and observations are about mental contents, then “agents” without mental contents cannot observe and therefore neither learn inductively.
Keywords
Observations as explanations
Observations are about mental contents
Machine learning
