Where's the proof it was?

Microsoft may have been lying. A lot of game companies do that.

I am skeptical to say the least.

The next step will be replacing Milo with Miley.

You posted a story from a website almost nobody has ever heard about yet you're asking about something that was discussed on many major websites?

Look, I'm guessing you are about 12-14 years old by the inability to analyse what you saw of that (absolutely shitty) Milo demo at last E3.  I am old enough to have lived through the "Eliza" storm many, many years ago (I even had the software running on my Pineapple ][). It was fun to play with the program and to watch how the tabloids declared "The End of Psychoanalysis". At that time, this was somewhat understandable as most people had no idea what a computer really was and what it actually did under the hood (which was, at that time, mostly the hood of an Apple ][). Google "Weizenbaum" and "Eliza" to get the story about that program, and how shocked and irritated Weizenbaum was about the reaction to his simple stunt.

Comes Microsoft with Milo at E3, and I see the same thing happening again. The same stunt, (possibly) the same software algorithms (with a lot more memory space to burn, Eliza ran in an 48kB Apple). But with a cute 3D interface, and the crowd goes wild, as it did in those old times with Eliza. I somewhat understand the reaction of the tabloids to the demo, since that is the function of tabloids/FOXNews. What I fail to understand is the reaction of the fanboys which I'd have assigned a better knowledge of analysing what they see a computer doing. Apparently I am wrong here and one can still fool a seemingly unlimited number of people with simple software, as long as the software "looks cute".

You can argue endlessly about whether the last demo was legitimate or not, but until I see it in a complete product I will be very doubtful as to whether it was representative of what the product will be. Even if the demo is legitimate, creating the content for a 5 minute play-session is quite a bit different than creating it for an 8 hour (or longer) play session. I could be wrong, but I suspect that any illusion of realism in interactions like this would be destroyed as you get repeat interactions.

http://www.youtube.com/watch?v=B8aUhsrM0sA

I agree, something like that would be amazing; but go to YouTube and try out their auto captioning. It's only in beta stage at the moment, but, gosh, is it sad. Google from my experience has been one of the best (if not the best) companies in the industry and they can't get speech recognition to work yet. Or you could try to narrate something to MS Word. I'm pretty sure your results will be bizzare. I'm only talking about this because to me it seems that speech recognition would be the easiest part of the whole AI suite to get to work (I'm sure it's a lot easier than the decision making and reactions). Honestly, I think this kind of technology is far far away.