The AI is nothing but a pattern recognition and response software on a computer. Ideally, to start with, the AI would be based on a huge room of server computers with a user inferface consisting of only a monitor, a microphone and earphones.
In popular culture we can see how Hollywood has tried to change the image of Artificial Intelligence to a humanoid robot, (Terminator, Metropolis, I, Robot). In reality, this method is simply not practical. The human-AI interaction is not one-sided. This means that the robot does not need to go shopping, go to the doctor, or even get any sunlight at all. Imagine a room in every city with various outlets. Consider it like a library. Each person has a desk, a monitor, earphones and a microphone that are all wired to the server in another room. They can then ask the bot for advice or information.
The bot would learn and get smarter everytime somebody interacts with it. The AI would get smarter from birth to growth similar to how a human being does as a baby and small child. We are told that "1" is one, and from then on we refer to this experience in our memory so that everytime we see the digit, we know what it is.
The teaching process for the AI could take between 1 and 20 years, depending on how much time is given. Or, we could develop a software that will feed information through to the bot, teaching itself. Then, the bot would be left alone for 6 months to process the information. Then, we will dedicate another year after that, so that it can recognize speech patterns.