While the vision of widespread co-habitation with robots
is beyond the 2020 horizon, recent advances in machinelearning
techniques are being experimented with to
model and support human behaviour in other ways.
Knowing what a person is thinking or wanting will enable
robots to be programmed to respond and adapt to their
needs accordingly. In the past, most machine-learning
applications operated ‘off–line’, where a set of training
data would be collected and used to fit a statistical model.
Nowadays, new techniques are being used to solve realtime
inference problems in which multiple streams of data
are processed from diverse sources. Statistical analyses
are then used to make inferences about the state of the
world. For example, when new information is received,
probabilities can be updated using Bayes’ theorem. This
allows machines to learn by reducing the uncertainty of
particular variables based on new information being fed
into it.