I know there has been plenty of halachic literature written about automation on Shabbos as far as timeswitches, eBay bids etc go, but are there any authoritative sources for halacha about artificial intelligence based automation on Shabbos?
Artificial intelligence, speech recognition and related technological systems are steadily improving at the moment. It isn't too hard to imagine that the virtual personal assistants such as Amazon Alexa, Google Assistant, Microsoft Cortana, Apple Siri etc will soon be capable of monitoring what we want without us explicitly asking them to do something (assuming that would actually be socially acceptable). If such a device monitors what we want on Shabbos and takes action, would this be a problem if we don't explicitly ask it to?
I'm envisaging systems which use a variety of sources of data - sensors built into our homes, built into the devices that are around us that we give permission to an app to access. The app then uses all of this data to establish our personal habits and predicts what we are likely to need in advance. For example, it saves us electricity by turning the lights off after we go to bed, but turns them on when it knows we are likely to go into that room if it isn't very light in there. It is predictive, and we can't be sure that any particular action will necessarily trigger a specific outcome (actually we probably could, but those are easy to avoid). I'm imagining a world where systems like this may be pervasive, and we're not conscious of relying on them, and they are likely to exist everywhere - in the streets we walk down as well as in our homes, in buildings we go into and walk past etc. Such systems would most likely become taken for granted and would exist very much in the background - things would just happen to make our lives easier without our conscious effort. Eg. We may not even be aware that the lights go off when we have gone to bed and come back on before we come down in the morning.
This is different from the case of timers, which are intentional (programmed) and near-certain to take effect (barring power outages). It is also different from the case of motion-sensor lights, where our action triggers an unwanted outcome. Profiling by AIs adds more layers of indirection (maybe like a grama?) and uncertainty, with the AI passively monitoring our actions without us directing it, building a model over the course of weeks or months, applying that model to take actions on its own, and tuning the model as it goes. In some cases actions may take place purely based on trained data or in others they may be triggered by a combination of a sensor and learned behaviour.
Additionally, there may be cases where it isn't necessarily clear whether an action even is a melacha. For example, if we are planning on eating out on Shabbat with a family who live the other side of town - an hour's walk away and one of our kids is ill and we decide not to go, the system could notify them that we will not be coming simply because it has heard us saying so, and will then tell us that we don't need to worry about them waiting for us and us not turning up. Has a melacha taken place here? I am not sure.