We will build a platform for software developers who want to use
realistic cognition in their AI models. It is based on insight of
the way the brain works. Currently, we believe that AI has reached a
technological impasse since it relies up to this day on the idea of
adjusting connection weights as a main form of parameter learning. And
this is true for many variants of neural networks out there, and these
models are accordingly incredibly costly and incredibly large. They have
no other way of learning, no pre-structured or structured learning.
The brain operates differently. It uses neuronal cells with all their
structural differentiation, many types of cells, and all the different
forms of plasticity that neurons undergo. I myself have studied this for
many years. And we are now at the point that we are defining abstract
units, we call them neural processing units, which encompass the most
important computational principles of neurons. A main novelty is that
they will use both external membrane based and internal molecular
storage mechanisms and exchange mechanisms between these areas of the
neuronal cell.
So it is clear that this is a radical departure from conventional AI,
but it is a form of computation which is based on the way the brain
works. So we can use these neural processing units to assemble them
for elementary tasks, and we can build whole networks from these groups,
a hierarchical system, for more complex tasks. This problem of
modularity and hierarchical structuring has not been able to be solved
within a paradigm where you only use synaptic weight adaptation (or
connection weight adaptation). And this will be solved once we
have this more complex neural processing units as basic
mechanisms. Overall there may be an equivalent number of parameters but
only very few need to be learned to solve new problems because there is
a lot of re-use possible from the modularity of the system.
We want to re-boot AI by offering new tools and technologies. We believe
this is absolutely necessary, and many different approaches are needed
at this point in time to overcome the restrictions of the existing
algorithms and technology. And on this basis, our team was assembled. We
have a NASA software engineer with much experience in complex software
construction. We have a founder and professor of neuro-robotics, and me,
and together we will try to create new tools for useful cognition.
Our work is funded by grants, by innovators and visionaries who support the ideas of cellular computation for combating disease, and by contributions from the general public, from people like you. Our general fund is the most important fund because it allows us to venture into new areas, and to try out projects which don't receive funds from other sources. Please contribute to our general fund!