it's the Cube.

Covering South by Southwest 2017.

Brought to you by Intel.

Now here's John Furrier.

Okay we're back live here at the South by Southwest

Intel AI Lounge, this is The Cube's special coverage

of South by Southwest with Intel, #IntelAI

where amazing starts with Intel.

Our next guest is Dr. Dawn Nafus who's with Intel

and you are a senior research scientist.

Welcome to The Cube.

>> Thank you.

>> So you've got a panel coming up and you also

have a book AI For Everything.

And looking at a democratization of AI

we had a quote yesterday that,

"AI is the bulldozer for data."

What bulldozers were in the real world,

AI will be that bulldozer for data,

surfacing new experiences. >> Right.

>> This is the subject of your book, kind of.

What's your take on this and what's your premise?

>> Right well the book actually takes

a step way back, it's actually called

Self Tracking, the panel is AI For Everyone.

But the book is on self tracking.

And

it's really about actually

getting some meaning out of data

before we start talking about bulldozers.

So right now we've got this situation where

there's a lot of talk about AI's going

to sort of solve all of our problems in health

and there's a lot that can get accomplished, whoops.

But the fact of the matter is

is that people are still struggling with gees,

like, "What does my Fitbit actually mean, right?"

So there's this, there's a real big gap.

And I think probably part of what the industry

has to do is not just sort of build new great technologies

which we've got to do but also start to fill that gap

in sort of data education, data literacy,

all that sort of stuff.

>> So we're kind of in this first generation of AI data

you mentioned wearable, Fitbits.

>> Dawn: Yup.

>> So people are now getting used to this,

so that it sounds this integration into lifestyle

becomes kind of a dynamic.

>> Yeah. >> Why are people grappling

>> John: with this, what's your research say about that?

>> Well right now with wearables frankly

we're in the classic trough of disillusionment. (laughs)

You know for those of you listening

I don't know if you have sort of wearables

in drawers right now, right?

But a lot of people do.

And it turns out that folks tend to use it,

you know maybe about three or four weeks

and either they've learned something

really interesting and helpful or they haven't.

And so there's actually a lot of people

who do really interesting stuff to kind of combine it

with symptoms tracking, location, right

other sorts of things to actually

really reveal the sorts of triggers for

medical issues that you can't find in a clinical setting.

It's all about being out in the real world

and figuring out what's going on with you.

Right, so then when we start to think about

adding more complexity into that,

which is the thing that AI's good at,

we've got this problem of there's only so many data sets

that AI's any actually any good at handling.

And so I think there's going to have to be a moment where

sort of people themselves actually start to say,

"Okay you know what?

"This is how I define my problem.

"This is what I'm going to choose to keep track of."

And some of that's going to be on a sensor

and some of it isn't.

Right and sort of being really intervening

a little bit more strongly in what

this stuff's actually doing.

>> You mentioned the Fitbit and you were seeing

a lot of disruption in the areas, innovation

and disruption, same thing good and bad potentially.

But I'll see autonomous vehicles is pretty clear,

and knows what Tesla's tracking with their hot trend.

But you mentioned Fitbit, that's a healthcare

kind of thing.

AIs might seem to be a perfect fit into healthcare

because there's always alarms going off

and all this data flying around.

Is that a low hanging fruit for AI?

Healthcare?

>> Well I don't know if there's any such thing

as low hanging fruit (John laughs)

in this space. (laughs)

But certainly if you're talking about

like actual human benefit, right?

That absolutely comes the top of the list.

And we can see that in both formal healthcare

in clinical settings and sort of imaging for diagnosis.

Again I think there's areas to be cautious about, right?

You know making sure that there's also

an appropriate human check and there's also

mechanisms for transparency, right?

So that doctors, when there is a discrepancy

between what the doctor believes and what the machine says

you can actually go back and figure out

what's actually going on.

The other thing I'm particularly excited about is,

and this is why I'm so interested in democratization

is that health is not just about,

you know, what goes on in clinical care.

There are right now environmental health groups

who are looking at slew of air quality data

that they don't know what to do with, right?

And a certain amount of machine assistance

to sort of figure out you know signatures

of sort of point source polluters, for example,

is a really great use of AI.

It's not going to make anybody any money anytime soon,

but that's the kind of society that we want to live in right?

>> You are the social good angle for sure,

but I'd like to get your thoughts 'cause you mentioned

democratization and it's kind of a nuance

depending upon what you're looking at.

Democratization with news and media

is what you saw with social media now you got healthcare.

So how do you define democratization in your context

and you're excited about.?

Is that more of

freedom of information and data is it

getting around gatekeepers and siloed stacks?

I mean how do you look at democratization?

>> All of the above. (laughs) (John laughs)

I'd say there are two real elements to that.

The first is making sure that you know,

people are going to use this for more than just business,

have the ability to actually do it

and have access to the right sorts of infrastructures to,

whether it's the environmental health case

or there are actually artists now who use

natural language processing to create art work.

And people ask them, "Why are you using deblurting?"

I said, "Well there's a real access issue frankly."

It's also on the side of if you're not the person

who's going to be directly using data

a kind of a sense of, you know...

Democratization to me means being able

to ask questions of how the stuff's actually behaving.

So that means building in

mechanisms for transparency, building in mechanisms

to allow journalists to do the work that they do.

>> Sharing potentially?

>> I'm sorry? >> And sharing as well

more data? >> Very, very good.

Right absolutely, I mean frankly we still have

a problem right now in the wearable base of

people even getting access to their own data.

There's a guy I work with named Hugo Campos

who has an arterial defibrillator and

he's still fighting to get access

to the very data that's coming out of his heart.

Right? (laughs)

>> Is it on SSD, in the cloud?

I mean where is it? >> It is in the cloud.

It's going back to the manufacturer.

And there are very robust conversations about

where it should be.

>> That's super sad.

So this brings up the whole thing that

we've been talking about yesterday

when we had a mini segment on The Cube is that

there are all these new societal use cases

that are just springing up that we've never seen before.

Self-driving cars with transportation,

healthcare access to data, all these things.

What are some of the things that you see emerging

on that tools or approaches that could help

either scientists or practitioners or citizens

deal with these new critical problem solving

that needs to apply technology to.

I was talking just last week at Stanford with folks

that are looking at gender bias and algorithms.

>> Right, uh-huh it's real. >> Something I would never

have thought of that's an outlier.

Like hey, what? >> Oh no, it's happened.

>> But it's one of those things were okay,

let's put that on the table.

There's all this new stuff coming on the table.

>> Yeah, yeah absolutely. >> What do you see?

>> So they're-- >> How do we solve that

>> John: what approaches?

>> Yeah there are a couple of mechanisms

and I would encourage listeners and folks in the audience

to have a look at a really great report

that just came out from the Obama Administration

and NYU School of Law.

It's called AI Now and they actually propose

a couple of pathways to sort of making sure

we get this right.

So you know a couple of things.

You know one is frankly making sure that women

and people of color are in the room

when the stuff's getting built, right?

That helps.

You know as I said earlier you know making sure that

you know things will go awry.

Like it just will we can't predict how these things

are going to work and catching it after the fact

and building in mechanisms to be able

to do that really matter.

So there was a great effort by ProPublica to look at

a system that was predicting criminal recidivism.

And what they did was they said, "Look you know

"it is true that

"the thing has the same failure rate

"for both blacks and whites."

But some hefty data journalism and data scraping

and all the rest of it actually revealed that

it was producing false positives for blacks

and false negatives for whites.

Meaning that black people were predicted

to create more crime than white people right?

So you know, we can catch that, right?

And when we build in more system of people who

had the skills to do it,

then we can build stuff that we can live with.

>> This is exactly to your point of democratization

I think that fascinates me that I get so excited about.

It's almost intoxicating when you think about it

technically and also societal that there's

all these new things that are emerging

and the community has to work together.

Because it's one of those things where

there's no, there may be a board of governors out there.

I mean who is the board of governors for this stuff?

It really has to be community driven.

>> Yeah, yeah.

>> And NYU's got one, any other examples of communities that

are out there that people can participate in or?

>> Yup, absolutely.

So I think that you know, they're

certainly collaborating on projects that

you actually care about and sort of asking