I'm really interested in exploring some AI ideas related to emobidment. To that end, I needed a "body" to put an "AI" in. So I started out by gutting out an old boombox, and putting a Raspberry Pi, a USB powerbank, and a USB charger in there. I also built a power supply multiplexer board that can allow switching from the battery to the charger, and back, without causing the Pi to reboot. And that stage is pretty much where I am. Getting that multiplexer thing built proved more challenging that I expected, but that's a story for another day.

Next steps are to start adding sensors to allow the "AI" to experience the physical world. Plans include:

1. GPS module, so it can experience motion and sense its velocity and location in space, etc. The same board also happens to be a cellular data board that will allow the device to upload data to the cloud.

2. One or more accelerometers for more localized perception of motion / orientation in space.

3. One or more cameras for vision. Ideally two so it can do stereo vision and have depth perception, etc.

4. One or more microphones to receive audio.

5. Anything beyond that is getting into pretty speculative area, but I can imagine adding sensors for temperature, barometric pressure, and other environmental factors, and even get into "senses" that humans don't really have, like sensing IR, ultraviolet light, electromagnetic radiation, magnetic flux density, etc.

Now obviously a Raspberry Pi doesn't have enough computing power to do a whole lot from an AI perspective, so the basic concept is to have the data being received by the box pushed to a remote server, where complex training and reasoning stuff will happen. Modules can then be pushed back down to the box so that some operations can be executed locally. That's where both Wifi and cellular data come into play. One of my "next steps" right now is to get the cellular data board configured, and figure out how to get NetworkManager to seamlessly switch between that and the various Wifi AP's that will be available depending on where I take this thing.

Anyway... there's a lot of speculation to all of this, and a lot of it is still very fuzzy in my mind. But the basic concept is simply to focus on the idea of how an AI can learn from physically experiencing the world in a way that's loosely analogous to the way a human does. If you've read any of Ben Goertzel's stuff and are familiar with his whole "baby AI" notion, then you may recognize some spiritual kinship in terms of thinking. I've been heavily influenced by Ben's work, and by OpenCog[1]. In fact, OpenCog (or some parts of it) may become part of this project before all is said and done.

[1]: https://opencog.org/