Here are my notes on this talk about the IoT and UX:

The Internet of Things for consumers has a very different business model than regular consumer electronics. In normal consumer electronics, somebody buys an object and the relationship is over until the consumer chooses to buy a new device (I would, however, dispute this as being simply a bad paradigm for consumer goods companies). But certainly for IoT, the sale is just the beginning of the relationship. The device alone has almost no value to the customer or the manufacturer, it is just a way to get people to use a service. Hardware will start to become more specialized.

Connecting stuff to the internet is easy but much of it is pointless. The internet can help you maximize efficiency of a fixed process, but that's not a problem that most people have. What is interesting is to make sense of the world on behalf of people: to reduce cognitive load and allow people to interact with services at a higher level of complexity. People are good at pattern finding but when there's lots of data. This is where predictive analytics, machine learning, pattern matching, etc. comes into play.

There are three issues with the UX of predictive analytics. Expectation, Uncertainty, and Control.

Expectation - How do devices that adapt communicate that they're adaptable? Does a chair vibrate when it changes posture? Should a coffee machine let you know it's schedule for you. What machines are adaptive and what are simply reactive? Have you ever gotten used to an automatic sink that does not require turning a knob and then stood in front of a non-automatic sink for 30 seconds wondering if the thing is broken?
Uncertainty - Predictive systems are unpredictable at first which is not what we expect from them. We judge systems on their mistakes. So a system that works 95% of the time still fails us quite often. And a device is typically just one in a whole bunch so something may always be failing.
Control - How do you control a device when things are wrong? Is there overtraining?

There are also four patterns for addressing predicting UX issues:

Set behavior expectations so that a person can build a mental model for understanding what the computer will do.
Explain when a state has changed or is going to change.
Create clear ways for people to adjust the predictive behavior.
Support people, don't work to replace them.