(Angus Taylor) Hi, I'm Angus Taylor and welcome to today's Tech Talk.
I'm really delighted to be here today with Peter Norvig
who is Director of Research at Google. Welcome Peter.
(Peter Norvig) Great to be here Angus. Thank you.
(Angus Taylor) Peter, as the Head of Research at Google, can you tell us
something about how you use data to innovate
and deliver better products and services at Google?
(Peter Norvig) Everything we do is based on data
and based on continual improvement.
So, we're always trying to keep track of how our services are doing,
look at how our users are responding and then make continual improvements.
Look at the results, can we come up with something better
and improve what we have
and then try to invent new services and products as well.
(Angus Taylor) And how quickly do you try and do that?
Do you try and cycle that through at a rapid pace?
(Peter Norvig) We certainly do.
So, you know every day we're running dozens or hundreds of experiments
for ideas to do something new and not all of them make it
but we're making changes every day.
(Angus Taylor) Right, so, the key is experiment
and keep the stuff that works?
(Peter Norvig) Right and don't worry about failure
but keep going ahead fast.
(Angus Taylor) Now one of the issues there is you are using a lot of data
and it's customer data in many cases.
How do you maintain trust with your users, with your customers?
(Peter Norvig) Trust is really crucial and we have to guarantee
that we're going to be great shepherds for their data and we…
first is understanding that it is theirs. It's not ours.
They're entrusting us with it but it belongs to them
and that means we're not going to give it away to anybody.
We're not going to sell it to anybody
and we're going to do world class defence
against anybody hacking into it
and we're going to help our customers with that as well.
So, we've gone beyond just simple passwords
to having multi factor identification to make it safer for them
and then we also give them confidence by saying
"You know what, we're happy to serve you
but if you're not happy we want to make it easy for you
to take your data with you".
So, we have services to say
"If you want to take the data some place else,
you want to leave us completely, or you want to be in two places at once,
we'll make it really easy for you to get your data out.
(Angus Taylor) Right. Now one older way of creating trust
was through long, incomprehensible User Agreements
but transparency is another way.
Can you tell us about what you do to make your approach
and the way you use data more transparent to users?
(Peter Norvig) So, we try to explain to people what we're doing
and we try to do that in terms they can understand
rather than in terms lawyers can understand
and when we make changes we try to highlight that.
(Angus Taylor) And transparency and how that data's used, what do you do?
How can I see how Google is using my data?
(Peter Norvig) That's in the terms and we like to give explanations for,
for where, why it's there, what we can do with it
and why you might want to have us have that data or not
and we certainly give you options.
So, for example if you use a search and you log in and you say,
"Keep my search history" then we'll do that
and we'll explain to you that there's some benefits you get.
So, one is you could search over your search history
and say "What was that query I was doing a week ago?
I forget exactly what the terms were".
You could go back and find it.
(Angus Taylor) And that's a true… I've used that many times.
(Peter Norvig) You've used that many times and also if you're logged in
and you give us permission to track then we can start personalising
and give you better results that are personalised to you.
But if you decide you don't want that. You don't want to log in.
You want the searches to not be associated with your account
then you can do that too. It's up to you.
(Angus Taylor) Now you're a global expert in artificial intelligence
and predictive algorithms are an important part of your work
as I understand it. Can you tell us something about
how you're using, or could use predictive algorithms
to improve economic and social outcomes?
(Peter Norvig) So, we like to provide new products
and you know we're providing goods for our users
and we're getting to the point where we're doing things
that they could never do before.
Right and so I know for me, as an amateur photographer,
boy I've wasted countless hours sorting through pictures
and putting them into folders and labelling them with key words
and now I can just throw them all into Google photos
and it does all that labelling for me automatically.
Right, so that's a small thing.
We're also saying we built all these capabilities to do object recognition
in pictures, to do machine translation, to do speech recognition
and so on and we incorporate them into our own products
but we also want to let other people use them.
So, we've built these services where you as a small company
may not have the expertise to do that yourself.
You can call in to our services and say,
"Tell me what's in this picture, recognise this speech".
(Angus Taylor) And what about Government, Peter?
How could Government use these sorts of predictive algorithms?
(Peter Norvig) Right, Government could certainly participate
in the same way if that made sense for them.
They could call in to you with our data
and of course Government has its own data
and they could be looking at the kinds of things we're doing
and see if they can duplicate that.
And I think Government should also be thinking about
the whole ecosystem, right.
So, Governments collect data
and they've shown that they're good stewards for it
but they don't have to do all the analysis.
They could just make it possible for others to do the analysis.
So, you know, popularise the fact "I've got this data here" and say,
"Who wants to do something useful with it?".
(Angus Taylor) And what would be an illustration of an area
where that could really have a positive impact?
(Peter Norvig) I spoke with a group in the U.S.
that's working on making it easier to access Government services
that have long and complicated forms.
Right and when we had things like the Veteran's Administration Form
where Veterans should get their benefits
but a large percent of them were dropping out
because the forms were too complicated and they didn't finish them.
And so, a third party, non-profit stepped in and said
"Instead of these long forms that you fill out paper
and you can only hand them in at this Office
between 7.00 and 9.00 o'clock on Tuesdays
we're going to have a phone app that's going to be much simpler.
You can go through step by step
and it greatly increased the completion process.
(Angus Taylor) Well, as the world's worst form filler
I would really love some of this kind of help like that.
(Peter Norvig) And so there… Government didn't have to do that.
They just had to say,
"We're going to accept these forms and you can fill them in this other way
rather than in the traditional way".
(Angus Taylor) Well, what limitations do you need to be careful of
when you're using these sorts of algorithms?
(Peter Norvig) So, one is you know you hear this maxim of
"Garbage in, garbage out", right.
So, if you built a system based on data and you have the wrong data
then it's going to produce the wrong answers.
So, that's something to worry about.
It could be completely wrong or it could be biased
in a way that you don't want it to be, right.
So, there are biases in the real world
and if your data comes from the real world, unless you're careful
it's going to pick up some of those biases.
And so, you have to protect against that.
(Angus Taylor) Now, as a Government, we have a real focus
on data integration and analysis.
Can you tell us what Google has learnt that we might learn from,
as the Australian Government in that area?
(Peter Norvig) Move quickly. Make it easy for people to try new ideas.
You know you generally have to be careful of what you do and so,
we try not to tell our employees what to do but say
"Go out and surprise us and come up with something interesting"
of course we also put processes in place
where not anybody can look at any data.
You have the, there's need to know and there's anonymization and all that.
But we just try to set a direction of saying
"We're trying to help our users. We want to go in this general direction"
but we're not going to have a step by step plan of how to go there.
We're going to let people experiment and come up with their own ideas.
(Angus Taylor) And just your own employees,
or do you effectively allow third party researchers
who are appropriately qualified to do that sort of work?
(Peter Norvig) So, if it's user data then only our employees can see that.
We can't, we can't let anybody who's not our employee see user data
but there's other types of data, like data we might have collected
off the Web that we can supply that to third party researchers.
(Angus Taylor) Traffic data for instance?
(Peter Norvig) Yeah, something like that,
that's not personally identifiable.
That would be fine
and we do a lot of sponsoring of university research
for people to look at those kinds of things.
(Angus Taylor) Now, one of the things we're particularly interested in using
longitudinal data sets, data sets that over time…
can you tell me something about how Google uses those sort of data sets
to improve products and services?
(Peter Norvig) Because we are so worried about
the personally identifiable data,
we tend not to do a lot with longitudinal studies of say
your search history because we just… we don't want to know,
right because it would be too dangerous for that to leak out.
There are other places where we do have longitudinal data.
We're working with a company called Planet
which is trying to take pictures of the earth, every spot, every day
and you can use that to track a lot of different things.
So, for agriculture, you know what do things look like?
For figuring out economic indicators,
what parts of what cities are growing faster?
(Angus Taylor) So, what would be an agricultural example for instance,
of how that data might be…?
(Peter Norvig) Varied. Is it green or brown?
Has there been more rain?
(Angus Taylor) Is there a crop in?
(Peter Norvig) Yeah. Or are people cutting down trees?
And so, we can track that on a day by day basis.
(Angus Taylor) Fantastic. Well look Peter, thank you so much.
We really appreciate you spending this time with us today.
Fantastic to hear about... to hear about what you're doing
at Google.
(Peter Norvig) Thank you.
For more infomation >> Jose could dump five inches of rain on Boston - Duration: 2:35. 
For more infomation >> Preparations underway ahead of Hurricane Jose - Duration: 0:59. 
For more infomation >> Martin County First Responders Working in the Keys - Duration: 1:29. 
For more infomation >> Donald Trump Meets With World Leaders At U.N. Assembly | NBC Nightly News - Duration: 2:11. 
For more infomation >> Recycled Percussion helps big NASCAR fan - Duration: 0:43. 
For more infomation >> Category 4 Hurricane Maria Takes Aim At Puerto Rico, Virgin Islands | NBC Nightly News - Duration: 2:10. 
For more infomation >> Coast prepares for effects of Hurricane Jose - Duration: 1:35. 
For more infomation >> More Than 140 People Arrested In St. Louis Protests | NBC Nightly News - Duration: 1:29. 
For more infomation >> Messi puso a bailar a Cristian Ronaldo | Suelta La Sopa | Entretenimiento - Duration: 0:37. 
For more infomation >> Game of Thrones Conquest & Rebellion: An Animated History of the Seven Kingdoms Clip (SUB ITA) - Duration: 4:31. 
For more infomation >> Chris Van Hollen Calls New Health Care Plan A 'Sneak Attack' | MTP Daily | MSNBC - Duration: 7:11. 
For more infomation >> J Balvin puso a bailar a sus seguidores de La gran manzana | Suelta La Sopa | Entretenimiento - Duration: 0:41. 
For more infomation >> Kate del Castillo y Ninel Conde entre las premiadas de Micrófonos de Oro en L.A. - Duration: 2:51. 
For more infomation >> Nonstop Việt Mix 2017 - Anh Vẫn Giữ Remix ft.Cứ Thế Chờ Mong Remix - Nhạc Trẻ Remix Hay Nhất Tháng 9 - Duration: 32:54. 
Không có nhận xét nào:
Đăng nhận xét