TMF Jeff
TMF owner and co-founder
- Joined
- Apr 2, 2001
- Messages
- 27,068
- Points
- 83
Google has developed a new AI that can converse so naturally that, as long as it's on task, the person talking to it won't even know that they're talking to software.
https://ai.googleblog.com/2018/05/duplex-ai-system-for-natural-conversation.html
In just a couple of years, they've gone from completely robotic, fake-sounding robocallers, to artificial assistants that can call a restaurant for you and the person taking the reservation doesn't even know they just talked to a bot.
I remember reading an article about an AI that had been trained to play Go (The Chinese game with black and white stones.) It's a game that had been considered impossible for a machine to beat the best humans at, because it has 300 times as many possible moves as chess.
So they trained the AI to play and set it to playing against itself. Thirty days later it had played 30 million games, and went on to beat the world's best human players.
Imagine if a person set out to become a Go master - starting at the age of 18, they manage to play 2 games of Go a day, literally every single day of their life. 50 years later they'll have played about 36,000 games. Or .1% of the number of games the AI was able to play in 30 days.
I remember a guy explaining it like this once - imagine if I asked you to answer a question for me that was within your area of expertise... except I'm going to give you 20,000 years to research and consider your answer.
We're not there yet, but we're not super far either. It seems like every time I turn on my phone, my traffic software and voice recognition software get noticeably smarter, which is just a small example of what I mean about what happens once the machines start doing the learning at their own pace.
https://ai.googleblog.com/2018/05/duplex-ai-system-for-natural-conversation.html
In just a couple of years, they've gone from completely robotic, fake-sounding robocallers, to artificial assistants that can call a restaurant for you and the person taking the reservation doesn't even know they just talked to a bot.
"Hi, I'd like to make a..."
"Hold on one second please."
"OK."
"Sorry, thank you for holding how can I help you?"
"No problem... uhm... I'd like to make a reservation for four people at seven O'clock next Wednesday."
"Ok... let's see... you said four people at seven for Wednesday the eighth?"
"Uhm... yes, that's right, Wednesday the eighth at seven PM for four people."
"Ok, you're all set."
"Great, thanks!"
I remember reading an article about an AI that had been trained to play Go (The Chinese game with black and white stones.) It's a game that had been considered impossible for a machine to beat the best humans at, because it has 300 times as many possible moves as chess.
So they trained the AI to play and set it to playing against itself. Thirty days later it had played 30 million games, and went on to beat the world's best human players.
Imagine if a person set out to become a Go master - starting at the age of 18, they manage to play 2 games of Go a day, literally every single day of their life. 50 years later they'll have played about 36,000 games. Or .1% of the number of games the AI was able to play in 30 days.
I remember a guy explaining it like this once - imagine if I asked you to answer a question for me that was within your area of expertise... except I'm going to give you 20,000 years to research and consider your answer.
We're not there yet, but we're not super far either. It seems like every time I turn on my phone, my traffic software and voice recognition software get noticeably smarter, which is just a small example of what I mean about what happens once the machines start doing the learning at their own pace.