Video/Audio Open Access
[lecture for the Dept. of AI, University of Groningen, Tuesday, March 3rd, 2020]
As impressive as the robots of the Boston Dynamics company are (no AI involved) and as impressive the many results of deep learning are (no AI involved, either), the goal of creating autonomous, intelligent machines is as far away as it ever was.
In this presentation, I will give a brief overview of several deep-learning projects in our group. As a next step I will try to indicate
what may be missing, as regards 'real' AI. We may need a closer look at biological systems, i.e., the brain of animals. There exists a wide gap between the control systems at the low level of reflexive movement and the equilibria that need to be maintained ('Boston') versus the higher levels of processing, up to the levels of cognition and reasoning, which are very much upstairs ('Eden'). The missing middleware layer should not be underestimated: It contains the brain stem, up to the thalamus in animals and humans.
It corresponds to the 300-million year period before the 200 million years period where the neocortex was present.
What is this middleware doing? The conclusion may be that there is no autonomy without self protection, possible due to the presence of a separate and specialized valuation network that determines probability times utility (p*U), similar to what brain stem, midbrain and amygdala are doing in animals.