The histories of parallel computing systems and artificial intelligence (AI) have one aspect in common: Both have experienced dramatic changes from a flourish period, followed by a long dormant one and decline, and then coming back again to another period of fantastic recovery/growth, with now a seemingly very bright prospect for the future.
 
In this talk, we review the situation and the lessons learned from the angle of parallel computation models and systems, based on the experience of early dataflow models. We will also comment on  the challenges and opportunities in exploring the design of future computer systems from the recent flourish in the field of AI, machine learning, and brain-like (or brain-inspired) computing.
 
We articulate why we should start system design from the foundation of a parallel program execution model (PXM).  To this end, we present a proposal of a PTM – a Parallel Turing Machine model.  The PXM of the proposed PTM is based on recent works inspired by early dataflow models – especially the event-driven codelet execution model.
 
We conclude the talk with a few open problems and conjectures on the limitations of PTM for future intelligent computing systems (智能计算系统) in the general sense.