Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

AI changes everything - FST

No description
by

Peter Siljerud

on 5 December 2018

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of AI changes everything - FST

Peter Siljerud
All separated elements:
Create a unique layout
Artificial intelligence
changes everything
Gordon Moore har sett slutet för Moore's lag...
1995 trodde han 2005

2005 trodde han 2015

2015 trodde han 2025
Background
How does it work?
AI examples
"This [AI] is the most dramatic that has happened in computer science during the last 40 years. I've never seen anything similar."

Stefan Carlsson, computer science professor, KTH
Change
The Future
"Nobody mentioned AI in the party leader debate. Hey, hello, wake up! This will be the most important issue in a 10 year time frame."

Max Tegmark
När Patrik Sundström från SKL under ett föredrag om e-hälsa i Halland ställde frågan om deltagarna kände att de fick stöd av lagstiftningen i sitt arbete skrattade de 300 personerna i publiken.
PARADIGMSKIFTE
"Jag är en ständig fiende till mobila tjänster."

Min blogg, januari 2009
Riv sönder
aktiebrev med
mA-logo!

Man vs. machine
The 7 stages of robot replacement
A robot/computer cannot possibly do what I do.

OK, it can do a lot, but it can’t do everything I do.

OK, it can do everything I do, except it needs me when it breaks down.

OK, it operates without failure, but I need to train it for new tasks.

Whew, that was a job that no human was meant to do, but what about me?

My new job is more fun and pays more now that robots/computers are doing my old job.

I am so glad a robot cannot possibly do what I do.
November 2013: Listen to this crazy conversation between Time's Michael Scherer and a telemarketing robot who refuses to admit her true artificial nature.
Journalist?
Author?
Speaker?
Photo model?
"About 100 years ago, electricity transformed every major industry. AI has advanced to the point where it has the power to transform every major sector in coming years."

Andrew Ng
Då ligger det mesta "under vatten", alla våra jobb, etc.
Teknikskeptikerna är numera i minoritet.
Vi ska inte utbilda oss i det som ligger vid stranden.
Ingen automatisk katastrof om teknik tar alla mänskliga jobb, en evig semester med stort välstånd.
Svenska cupen 2016/2017
UEFA Europa Leauge
Statistik om löpvägar, passningsvägar, avslut, postitionsspel och praktiskt taget allt annat som händer på planen matas in tillsammans med uppgifter om sömn och kost.

Med hjälp av av algoritmer letar sedan deras plattform upp mönster i spelet som sedan presenteras för tränaren som en siffra.

“Det kan vara allt ifrån hur troligt det är att en spelare blir skadad till hur troligt det är att en spelare gör mål från en viss position.”
2015 Superettan,
avancemang till
Allsvenskan
Investments
Thank you!
Above 30 y.o.: peter.siljerud@futurewise.se

Under 30 y.o.: @siljerud
futurewise.se/fst
"The shrimp that falls asleep in the stream,
wakes up in the soup bowl."

"Well over half of our business is scientific or science-related"

Martin Sorrell, vd för WPP med 162 000 anställda
AI-tränare: person som är skuggskribent till chatbot










































Om AI är hjärnan, så är nätverken artärerna.
Astrofysik data
Crowd cheering
Action recognition, such as high fives or fist pumps
Commentator excitement (tone of voice)
Commentary (exciting words or expressions obtained from the Watson Speech to Text API)
Shot-boundary detection
TV graphics (such as lower third banners)
Optical character recognition (the ability to extract text from images to determine which player is in which shot)
https://www.ibm.com/blogs/research/2017/04/ai-technology-auto-curates-2017-masters/
Metadata:
Nyckelord
Koncept
Bild
Ljud
Känslor
För att bättre matcha reklam med tittarnas intressen
Från officiellt till personlig trailer
(den som tycker om romantiska komedier får dessa scener framlyfta, etc.)

PROCESSING POWER
BETTER ALGORITHMS
MORE DATA
"Software is eating the world."

Marc Andreessen, Internet pioneer
my $err_fornamn = "";
my $err_efternamn = "";
my $err_gatuadress = "";
my $err_postnummer = "";
my $err_postadress = "";
my $err_telefon = "";
my $err_epost = "";

my $sel_postforskott = "";
my $sel_faktura = "";
my $sel_forskott = "";

read(STDIN,$temp,$ENV{'CONTENT_LENGTH'});
@pairs=split(/&/,$temp);
foreach $item(@pairs)
{
($key,$content)=split(/=/,$item,2);
$content=~tr/+/ /;
$content=~s/%(..)/pack("c",hex($1))/ge;
$fields{$key}=$content;
}

if ($fields{no_black_M} == 0 && $fields{no_white_M} == 0 && $fields{no_black_L} == 0 && $fields{no_white_L} == 0) {
reload_page();
}

The world's first AI solution
1959
Echo-cancelling on phone lines
= 20 atomers bredd
Intel's latest fab in Chandler, Arizona = 5 billion USD
7 nm
Unity: Book of death
Google 2013: x2 antal datacenter
Google's processor (TPU) for
AI calculations in their datacenters
AI chips for greater speed
Do we need faster computers?
DeepMind's text to speech

"Software is eating the world,
but AI is going to eat software."

Jen-Hsun Huang, CEO Nvidia
AI software platforms
Lagrad data i petabytes (miljoner GB)
Källa: Göran Lindsjö, rådgivare åt Regeringskansliet
Better
processors
More demanding
algortihms
Need for
more data
ENABLES
DRIVES
REQUIRES
The law about Moore's law:

The number of people predicting the end of Moore's law doubles every 24 months.
10 nanometer
2 beardseconds
1997
2007
2011
2015
2017
AlphaGo: Studerade tidigare partier av människor
AlphaGo Zero: Ingen info om människor, 3 dagars egen träning (slog gamla AlphaGo med 100-0)
"AlphaGo Zero is so powerful because it is no longer constrained by the limits of human knowledge."

Demis Hassabis, CEO DeepMind
30 million US households
Source: Morning Consult, 2017









5 000 anställda som jobbar med utvecklingen av Alexa och företagets Echo-produkter = Astrazeneca i Sverige
AI bättre än radiologer
170 miljoner CCTV-kameror i Kina
Drönardesign som påminner
om bäckenbenet hos en flygekorre.
Yes, they wanted the Elbo to reference the Danish mid-century modern style. And they wanted the seat 18 inches off the floor, and capable of supporting 300 pounds. But they let algorithms decide everything else.
En treåring behöver ju inte tränas att titta på en miljon katter för att fatta att det är en katt...
What has AI got to do with the human brain?
Machine learning: a type of algorithm with the ability to teach itself by analyzing (big) data sets, in order to find patterns in other data sets
Teorin har funnits länge
- men inte data kapaciteten att analysera
1 miljon bilder på katter.
Deep learning: Machine learning with advanced
"deep" (multi-layered) neural networks
What does it look like inside an AI brain?
Andrew Ng
Google 2012: Ng’s breakthrough was to take these neural networks, and essentially make them huge, increase the layers and the neurons, and then run massive amounts of data through the system to train it. In Ng’s case it was images from 10 million YouTube videos. Ng put the “deep” in deep learning, which describes all the layers in these neural networks.
Neural Network is less complex and requires more information about features for performing feature selection and feature engineering method. On the other hand, Deep Learning Neural Network does not require any information about features rather they perform optimum model tuning and model selection on their own.
Source: University of Cambridge
Number of Facebook likes to know you as well as...
Colleague: 10 likes
Friend: 70 likes
Family member: 150 likes
Spouse: 300 likes
Neural networks: A number of connected artificial "neurons" that create a system for machine learning
Klassiska AI: Intelligensen programmerades av människor
Då ligger det mesta "under vatten", alla våra jobb, etc.
Teknikskeptikerna är numera i minoritet.
Vi ska inte utbilda oss i det som ligger vid stranden.
Ingen automatisk katastrof om teknik tar alla mänskliga jobb, en evig semester med stort välstånd.
There is always free cheese in the rat trap...
DeepFace can look at two photos, and irrespective of lighting or angle, can say with 97.25% accuracy whether the photos contain the same face. Humans can perform the same task with 97.53% accuracy.
If you don't pay for a service
- then you are the product
Machine learning is a type of computer program or algorithm with the ability to teach itself by analyzing data (inputs) and coming up with a solution (output).














Error rate: 0,23%
https://arxiv.org/abs/1202.2745
List of unwanted expressions:

Viagra
You've won the lottery
Nigerian prince
Meet hot Russian women
...
1) Rule-based system
2) Machine learning
Har möjligheten att generalisera
200 hundraser
Geometry part of SAT-test: 49% correct (=average student)
AI gets better and better,
humans do not get better.
AI challenges
Results are hard to interpret
AI with prejudice
1 of 44 winners had dark skin,
a few were Asians.
AI algorithms analyzed:

youth (in relation to age)
skin quality
face symmetry
likeness with models and actors
https://www.engadget.com/2018/01/23/photo-stitch-ai-fail-the-big-picture/
Felaktig stitching hos bilder,
AI förstår inte att människor
inte kan vara 20 meter höga

What if AI finds a connection between
freckles and ice hockey skills?

Target vet om en kvinna är gravid
(och i vilken månad)
The freaky line
Personal integrity
Erroneous results
Up to 300 USD discount for
"predicted" safe drivers
https://techcrunch.com/2016/11/02/uk-car-insurance-firm-wants-to-scan-social-media-posts-to-price-premiums/
http://playground.tensorflow.org
Can you make a computer write
Leo Tolstoy's
War and peace
?

100 iterations:
2 000 iterations:
700 iterations:
300 iterations:
500 iterations:
1 200 iterations:
Mambojambo, men en idé om mellanslag,
även om det är två ibland
Börjar få en idé om citattecken och
punkter.
Börjar forma korta ord som: “we”, “we”, “his”, “which”, “and”
Nu börjar det se ut som engelska
Börjar lära sig längre ord, samt utropstecken
Nu börjar det nästan likna litteratur!
We’ll give the AI a huge chunk of text and ask it to model the probability distribution of the next character in the sequence given a sequence of previous characters. This will then allow us to generate new text one character at a time.
https://google.github.io/tacotron/publications/tacotron2/index.html
3 neuroner 20 neuroner
Overfitting!
Jmf en student som lär sig alla svar
utantill - klarar alla rätt så länge alla
frågor är desamma. Jämfört med en
som förstår innehållet och kan tänka
själv.
Möjligheter att lura AI
The neurons get good at identifying different types of features, e. g. dog eyes, noses or complete dogs
Andra neuroner har koll på helt andra saker,
exempelvis sköldpaddsskal, bollar och hjul
Is AI better than human beings?
Text to speech
Driving
Chess
Jeopardy!
Image recognition
Translation
RISING "SEA LEVEL" WITH AI
Design
Calculation
Försvinnande öar i Bengaliska bukten - och mänskliga kunskaper där AI överträffar oss
DeepFace: 97,3% accuracy (human 97,5%)
Liten övervikt för människa vs. AI pga svåra ord, t. ex. "Merlot"
För trafikskyltar är AI dubbelt så bra som människan
Humour
https://techxplore.com/news/2016-08-deep-neural-network-approach-sarcasm.html
Well, this day was a total
waste of makeup.
AI identifies sarcasm with 87% accuracy
Image recognition contest ILSVRC 2017: 97,8% accuracy
Tävlingen nedlagd
(2010 var det 25% fel!)
Professional transcriber: 5,1% error rate
Chinese robot pass medical exam
http://www.chinadaily.com.cn/bizchina/tech/2017-11/10/content_34362656.htm
Memorizing
Investments
Science
Programming
Writing
Social interaction
Leadership
Art
Education
Speech to text
Which voice is synthetic?

“George Washington was the first
President of the United States.”
Alternative 1
Alternative 2

NameTag
Syntetisk verklighet skapar så avancerade fake news att det inte längre är möjligt att bedöma vilka nyheter som är sanna
Neuralink (Elon Musk)
EEG headset
"Okomprimerad" kommunikation - man går inte via komprimerat tal utan via hjärnimplantat kan man kommunicera direkt
Experiment 2014: hjärnvågskommunikation mellan Frankrike och Indien (binär)
1 apr 2010
“It was extremely limited. It could tell you the dog was tired or angry. But you can kind of see that anyway.”
I don't know if you want your dog to have a headset like this, looking like Lady Gaga on stage...
Telepathy between India and France (2014)
Now engineers at Imperial College in London have developed an EEG device that can be worn inside the ear
Samtal med avatarer / döda släktingar
Memorial bots
Intel Vaunt
"A young girl played football."
"The lawyer drank coffee."
"The witness yelled during the trial."

(some of 240 short sentences)
Carnegie Mellon University i USA
87 procent rätt
http://illvet.se/manniskan/hjarnan/artificiell-intelligens-laser-dina-tankar#cxrecs_s
"We anticipate that computers in the not-so-distant future will interact directly with the human brain in a fluent manner, supporting both computer-to-brain and brain-to-brain communication routinely. "

Alvaro Pascual-Leone, neurology professor, Harvard University
Who are we, as human beings, when more decision making is AI-based?
Source: CB Insights, Nov 2017
Number of AI patents
AGI (strong AI): Artificial general intellicence on a human level, that can solve a wide range of tasks (even without specifically being trained for them)




Pell's law of AI lock-in
Superintelligence: Artificial intelligence that is far smarter than the best human in all areas (including creatitivity, wisdom, social skills, etc.)
AI editing images like a photo pro
Weak AI: Artificial intelligence that can solve a specific task
Everybody beats a chess computer playing noughts and crosses
Anticipated year when we reach AGI
(average of leading AI researchers):

Source: Future of Life Institute
2047
(although some say never)
Companies
experimenting
with AI
Commercial
success
More AI
investments
Competition
needs to follow
More AI-based solutions on
the market
Self-enhancing AI:

Facebook tests 300 000
AI-algorithms every month
(one new every10 seconds!)
Superintelligence is humanity's
last innovation
Energy cost savings: 40%
"Precis om elektriciteten förändrade stora
delar av samhället för hundra år sedan,
har jag idag svårt att se någon bransch
som AI inte kommer att förändra under
de kommande åren."

Andrew Ng
Stort tack för mig!
Över 30 år: peter.siljerud@futurewise.se

Under 30 år: @siljerud
futurewise.se/iamcp
"Räkan som somnar i strömmen,
vaknar upp i soppskålen."








False reality (synthetic reality)
(images to the right are generated by AI)
SK Telecom använder ett AI-baserat system (Tango) för att upptäcka fel i näten, liksom att åtgärda felen och optimera prestandan.
Falsk verklighet (syntetisk verklighet)
(bilderna till höger är genererade av AI)
300+ AI acquisitions since 2012

Source: CB Insight, Feb 2018
Predicted brain acitivity (top)
Actual brain acitivity (bottom)
The ability to achieve complex goals
Intelligence
Biological intelligence
Artificial intelligence
What is intelligence?





Google speech-to-text: 120 languages
"Cloud Speech-to-Text also now includes automatic punctuation in speech transcriptions thanks to a new [type of] neural network."
"We first unveiled the Cloud Speech API in 2016, [...]
with usage more than doubling every six months."

Dan Aharon, Product Manager, Cloud AI, April 2018
Microsoft: 6 languages
IBM: 7 languages
What[Well],
Prince
,
so Genoa and Luc
c
a are now just family
is states[estates]
of the
body parts[Buonapartes],

b
ut I warn you
,
if you don't tell me that
T
his
M
eans
W
ar
,
if you still try to defend the
infamous[infamies]
and
H
orrors perpetrated by Antichrist
,
I really believe he is Antichrist
.
I will have nothing more to do with you and you are no longer my friend
,
no longer my
'
faithful slave,
'
as you call yourself
!
First part of chapter 1, War and Peace
Cost per hour:
1,44 USD
Source: https://cloud.google.com/speech-to-text/pricing
Source: IBM
https://arxiv.org/pdf/1703.02136.pdf
Full transcript