Predicting the three-dimensional structure of a protein from its amino acid sequence is one of the most important current problems of modern biology. The GASP (Critical Assessment of Structure Prediction) blind prediction experiments aim to assess the prediction capabilities in the field. A limitation of GASP is that predictions are prepared and filed by humans using programs, and thus, what is being evaluated is the performance of the predicting groups rather than the performance of the programs themselves. To address this limitation the Critical Assessment of Fully Automated Structure Prediction (CAFASP) experiment was initiated in 1998. In CAFASP, the participants are programs or Internet servers, and what is evaluated are their automatic results without allowing any human intervention. In this paper, we review in brief the current state of protein structure prediction and describe what has been learned from the CAFASP1 experiment, the evolution toward CAFASP2, and how we foresee the future of automated structure prediction. We observe that the histories of "in silico" structure prediction experiments and computer chess tournaments show some striking similarities as well as some differences. We question whether the major advances in automated protein structure prediction stem from novel insights of the protein folding problem, of protein evolution and function, or merely from the technical advances in the ways the evolutionary information available in the biological databases is exploited We conclude with a speculation about the future, where interesting chess might only be observed in computer games and where the interpretation of the information encoded in the human genome may be achieved mainly through in silico biology.