Ask HN: Open problems in ML/data prediction

I have been competing in data prediction contests (like those on kaggle.com), and I am interested in trying a prediction collaboration.

I'm hoping to recruit 10-20 people who would work together with the common purpose of putting together a single prediction system. What open questions would:

1) Apply machine learning/prediction techniques 2) Use publicly available data 3) Hold enough general appeal to facilitate recruiting other ml folks.

I'm hoping that, by removing the competitive aspect, participants will have more incentive to share ideas... and as a result it could be a more educational experience.