Every problem has a software based solution

Think of any difficult problem, you can probably find a way to relate it to software. Create and host the software, let people with money or data use it.

Criteria

• You have no money (to spend on acquiring
  data, expertise, equipment), only a
  computer and internet access.
  
• You have no data.
  
• You only have skills in web development,
  such as LAMP stack or appengine.
  
• You do not have specialized knowledge
  gained from years of study. You are
  reasonably smart but time limited and can
  only study from textbooks/websites during
  the duration of your coding, which is
  limited to a maximum of 1 year
  
• You can only work alone or with one or
  two other people
  
• The software should generate money for
  you. It should not generate it from
  advertising, only from paying users. This
  is to force it to be more useful than just
  another web 2.0 social
  networking/aggregation site. The paying
  users could be in the form of charity
  donations or government funding (it is
  likely to be useful if people are willing
  to donate money to it).
  
• You only want to contribute on huge
  world changing problems.

Some problems looking for a solution

• Tissue Engineering
  
  
  • You want to replace aged or
    damaged organs and tissues such as the
    heart, liver, lungs or pancreas.
    
  • This can be organized as a giant
    optimization problem. What sequence of
    cells/chemicals need to be output on
    which xy coordinate at which time in
    order to create functioning biological
    material.
    
  • Hypothetically, you would set up a
    million identical dispensing robots,
    each loaded with cells/chemicals, and
    intrumentation to check the
    functioning of the biological material
    produced. You do not have the money or
    expertise to create this setup,
    however you can set up the
    optimization problem on amazon
    mapreduce to direct the robots.
    
  
  
• Speech Recognition
  
  
  • This is a problem of having enough
    labelled data.
    
  • Mechanical Turk workers contribute
    and label the data.
    
  • Hosted speech recognisers are
    automatically generated from the
    labelled data. Users pay to use the
    speech recognisers, revenue going
    in, for example, a 90/10 split
    between the funders of the data
    creation/labelling, funders of the
    compute resources, used and you.
    
  • Users should be able to team up
    through the site and get a share of
    the produced speech recogniser's
    revenue (by donating either their
    time, money, computer power, or data).
    
  
  
• Self Driving Car
  
  
  • The hard, software only, part of
    this is basically image recognition
    (which direction/speed to go while
    viewing video), which is about
    having enough labelled data, and so
    the solution is very the similar to
    the speech recognition item above
    (the produced classifier is not
    hosted but downloaded to a car computer).
    
  
  
• Artificial Doctor
  
  
  • The problem is create an expert
    system(s) for medical diagnosis. It
    has to be sufficiently comprehensive
    and detailed to not give too many
    errors.
    
  • Create a site where people can
    upload expert system rules and test
    them in combination with possibly
    millions of rules other users have
    uploaded.
    
  • Workers are compensated by
    assessments of their contributions
    by other workers. For example, other
    workers rank 10 randomly selected
    contributions, and results are
    collated and money dispensed
    accordingly.
    
  
  
• Loneliness
  
  
  
  • Create a virtual companion.
    
  • The virtual companion needs to be
    able to understand the user's speech
    and facial expressions through video
    in real time. It needs to know what
    to say needs to have a pleasing
    voice and facial expressions itself.
    
  • Use the software solution created
    for speech recognition above. Run
    that system for recognition of
    speech, emotion in speech, body
    position, body language, facial
    expressions, recognition of
    household objects (these are called
    classifiers).
    
  • Train machine learning algorithm
    to generate speech and facial
    expressions, using the above created
    classifiers (the compute resources
    used will be funded by users as was
    with the data collection/labeling).
    
  • Use the collaborative expert
    system website from the artificial
    doctor problem above, for creating
    rules for what to say/emote.
    
  • Charge lonely people to converse
    with the generated virtual
    companions, splitting revenue 90/10
    between funders and you.