And here's me thinking the "©2009 Artem S. Tashkinov" at the end of the article was a bit of a giveaway.

Scott Adams published his first adventure in 1978. Infocom published Zork for microcomputers in 1980. While they may have been relatively contemporary, Adams did most of his games alone or with at most one collaborator, compared to the group working on Infocom's technologies. They're both important pioneers, even if Infocom's efforts have dated better.

Bear in mind also that the original version of Zork and Infocom's interpreter was improved over the years, too. I can distinctly remember it being easier to play later in life - not just because I was older, but because it was a little more forgiving with its vocabulary compared to the original version (I booted up the TRS-80 to check).

Cpt. Asseline was a stunt pilot?

So wait, a plane switched to "landing" mode was being used for something other than landing, and then suddenly behaved in an unpredictable fashion by trying to land? I'm shocked!

It's not as simple as blaming "software fault" or "human error": it's in the interaction between the systems and the human, and the assumptions that each maintains about the other's actions in a marginally-stable, high risk environment.

No, not at all. The lesson that if you hide the risks from people, they tend to make more risky decisions.

Rather than simply cushioning people from the risks in their environment, they need to be made aware of them and their consequences.

Anyway, geeks? Football? Going outside to play? That's crazy talk! :-)

Back in 1988, an Airbus A320 crashed at an air show during low-level flight manoeuvres. The brand new fly-by-wire systems made the plane easier to control in situations that a non-wire flight system would have problems. By making it that easy, the system also made it easier for pilots to push closer to the unstable edges of that envelope without the same level of feedback that things could go wrong. Things went wrong. People died.

Twenty years later, we're still learning the same lessons, it seems.

From the article at least, it doesn't look like they're doing anything particularly special, here. Segmenting people from the background and running something like eigenface classification or template matching on the foreground... anyone who's halfway competent with some of the major computer vision libraries out there could probably write something like this without really straining. Especially if it's in a partially-controlled environment with good lighting.