TIM: Artificial intelligence is “machines learning by doing.” NEIL: That’s one piece, yes. B, or the second piece is the ability to understand natural language. Think about how we talk,
we’re using slang, idioms, jargon. Lots of different
kinds of things all the time. If I tell you, Tim, that I’m feeling blue ’cause it’s raining cats and dogs, you know what I’m talking about, right? If I tell a machine, what
does a machine think? Right? That I’m physically the color blue? That small animals are
falling from the sky? I mean, that does not compute, right? But AI understands that,
they’re trying to draw context from the conversation. They’re
not taking everything as literal or as key words.
They’re looking for that, you know, the overall
grammatics, the syntax, the intent, to get the context. TIM: And you’re saying machines can do this? NEIL: And machines can do this.
That was the big challenge, actually, in playing Jeopardy.
A lot of people said, “well, oh can’t it just
look up the answers? It’s like a big search engine.” It’s like think about how they talk, think about jeopardy, right? They give you an answer, you have to figure out the question.
That’s even more complicated. All right, while we take for
granted just being able to talk like this, it’s
incredibly hard for a machine. Natural language processing
is basically just taking raw input like we’re talking
now and the machine is able to understand the context of
what’s being communicated. Beyond just key words,
it’s looking to say, “okay, grammatically,
what’s the structure? What are the key drivers
here? What was said before? What wound up being said after?” To draw information beyond
just the words themselves. You think about us, humans,
90% of our communication is actually non-verbal. There’s
things about our word choice, our tone of voice that
actually convey meaning. The machine has to try
and figure that out. Right, it’s just not, it’s not, you know, second nature to it. (metal clang) (metal
whooshing) (metal whooshing) (whooshing) (electronic zoom
noise) (flipping whooshing)