There have been games with NPCs smarter than that. There have been text adventures with sentence understanding at that level. It's a cute demo, but calling it a "living and conversing agent" is way too much.

"Mario, exit the game." "Do you really want to exit the game?" So not impressed.

On the speech front, there is now an automated telemarketing agent which is about as smart as most script-driven human telemarketers. That's a somewhat scary development.

The telemarketing robots that are scripted well enough to continually deny that they're robots are somehow the scariest/most impressive for me:

https://www.youtube.com/watch?v=unRDMy7O62o

sounds like Lenny which is an anti telemarketer SIP bot you can forward calls to:

https://www.youtube.com/watch?v=p5IKbFATKLo

I usually just say "please hang on, while I get my credit card", and then just leave the phone off the hook...

This is amazing. I wonder if anyone has done something similar for email spam.

if you're interested in more of these calls (some are quite funny) there's an entire reddit dedicated to it:

http://www.reddit.com/r/itslenny/

I think that's a person playing back pre recorded sounds. AI isn't advanced enough to impersonate a telemarketer.

It is. Offshore telemarketing - no more fake localized names and accents, just hire a person who understands spoken english and install the MidwestFemale 3.68 module.

I had no idea they were using this type of tech already, the second call is so smooth.

I think most of these comments are missing the point. As I understood the video they are trying to create an AI that plays the game. The AI will respond to its experiences as it learns about the game. The initial Mario state is that it knows nothing then as it learns its environment it will adapt to figure out how to play.

This seems like a pretty bold project. I'm curious if they can get the AI to win the game.

> "Mario, exit the game." "Do you really want to exit the game?" So not impressed.

This depends on how the response is generated. If it's a script, then of course there is nothing impressive about it, but if parts of the system can recognize importance in action and generate a sentence that asks for confirmation for important actions then it is a completely different ballgame.

> On the speech front, there is now an automated telemarketing agent which is about as smart as most script-driven human telemarketers.

Samantha west is human operated. Some dude is just playing back prerecordings that hopefully fit the conversation. It has absolutely nothing to do with AI.

http://newsfeed.time.com/2013/12/17/robot-telemarketer-saman...

Could give a few examples ?

Strictly speaking, this is hardly AI, but a rather traditional approach of handwritten grammars.

"AI is whatever hasn't been done yet." http://en.wikipedia.org/wiki/AI_effect

In this case it has been done quite a few times (and better).

So, SHRDLU and what else?

You do know that SHRDLU is still unique enough to warrant multiple resurrection attempts?

http://www.semaphorecorp.com/misc/shrdlu.html

Would love to see some examples!

The speech recognition and synthesis are primitive compared to any smartphone assistant these days. The natural language processing is about equivalent to SHRDLU from the 1960s [1]. It turns out that this approach based on manually constructing syntax trees and applying simple logic can make some fun demos but is ultimately a dead end in terms of making systems that are actually useful, as was discovered in the "AI winter".

The part that controls Mario looks similar (if not identical) to this: http://aigamedev.com/open/interviews/mario-ai/

If you want to see the state of the art in using AI to play video games, look no further than "Playing Atari with Deep Reinforcement Learning" [2], where a single general AI system learns to play many different games. The generality is what makes it impressive. Its only inputs are pixels and score, and its only outputs are joystick and button state, just like a human player. This makes it unlike these Mario systems which are hand programmed very specifically for Mario, and use special instrumentation of the game state that skips pixels entirely.

[1] http://en.wikipedia.org/wiki/SHRDLU

[2] http://arxiv.org/abs/1312.5602

Note that the Atari playing system is what got Google interested in buying Deep Mind (the company behind it). It was a pretty significant advance on the state of the art at the time.

Deep Mind sold to Google for around $500M

This is absolutely an impressive demo, but the way to look at it is as a 'mash-up' of a variety of components that have all been demoed before in isolation and usually with a lot more attention to the technology behind it but a much less slick presentation.

So kudos to the people that put this together but it is mostly the mashup that is new, not the individual pieces and those pieces have demos all their own that make for very interesting reading and viewing (where applicable).

I should do a blog post on this, but that will be a ton of work to properly represent the state of the art (which this demo does not).

    judge: The cat is green, the dog is blue. 
    ALICE: Is green the dog is blue also cat? 
    judge: What color is the dog? 
    ALICE: orange.

I am not impressed.

Nor should you be. But I think Alice will do better than Mario. All Mario does is talk about his feelings. Alice can do that too.

It involves a lot of techniques which are typically called AI. E.g. planning and learning simple models of the world. The best word is "weak AI" or "narrow AI".

I've always been interested in the use of natural language recognition in games. Does anyone have some good resources about the topic?