Above, the "Bar of Soap" prototype device by Brandon Taylor,
Stacie Slotnick, and Michael Bove, all working out of the MIT Media Lab. Its function is not to do anything, but to instead anticipate what you would want to do by dint of how it is held. Hold it like a phone? It guesses phone. Hold it like a TV remote? Its internal accelerometers indicate "TV remote." On its own it's useless, but as the ability to detect its user's intended use improves—earlier versions were in the 60-70% accuracy range if I'm reading the data correctly—it may find its way into a multi-purpose gadget in the future.
Is it just me or does the picture looks like it came out of an '80s-era copy of Omni?