How to make an artificial general intelligence... today!

1. Call IBM and order 100,000 truenorth chips. This will give you 100 billion neurons and 26.8 trillion synapses.; the same as a human brain.
2. Create an AI kindergarten by connecting sensors to the chips(e.g. cameras, microphones, speakers, etc) and allow people to interact with it.
3. Connect the chips+sensors to additional chips in series or parallel, or a combination of both, until they have enough neurons and connections to match human accuracy and brain-inspired regions. Think of a visual cortex as 2 cameras with specialized chips(same truenorth chips but with different neuron and synapse configuration) that send output to specialized chips and so on until you get to just a few central chips that are abstracted from the data so far that they just make executive decisions like "yes" or "no" based on all the sensors and processing in parallel.
4. Call it Skynet and sell it to the Air Force. Or use it to trade stocks or some other self-serving purpose.
5. Wait for the Air Force to connect it to the defense network, or just plug it into your LAN.
6. Let it interact with billions of people.

So, it's not very easy or practical, but certainly within reason and current technology. You might even be able to get Samsung to whip you up a special board for all the chips. It would probably consume 7,000W of power which is on par with a home standby generator. You could give the AGI a little boost by translating all of the deep learning algorithms over to it, things like: image recognition, speech recognition, lexical parsing, etc.

I was a little shocked, myself, to realize that we have reached this capability in 2017 which could put us on track to achieve better than human intelligence in the near future. I wouldn't be too surprised to learn that multinational corporations like google or governments like the US have already made one.