1) Machines are increasingly impacting daily human routines through technologies like smart home devices and driverless cars.
2) Both humans and machines process information through pattern recognition, but humans excel at piecing together incomplete information in new ways while machines rely more on analyzing large datasets.
3) Early attempts by companies to use only data analysis or only human judgment in developing TV shows met with varying levels of success, showing the value of combining the two approaches.
4. 4
Notes:
1. Intro - When we talk about machines today, weâre referring to artificial intelligence and machine learning
programs (like Siri) and robots like Arnold in Terminator. Do I have to call him the Governator now? Not
sure. In any case, we all see these machines around us, now impacting our day-to-day.
2. Hereâs a simple example. Letâs walk through the images at the top.
- You used to wake up to a mechanical clock a wind-up that would rattle your bedside each morning.
- Once you were up, on a cold day you might wander over to your thermostat check the temperature and
manually turn on your heating.
- After some breakfast you might cozy up to a print newspaper magazine...telling you about the âyuuuuuge
crowdsâ at Trumps inauguration.
- You then jump into your car for your daily commute to work, considering traffic as you go along and
adjusting your route.
3. Today, you wake up to an alarm clock set by your smartphone.
- As you wake up you either control your Nest thermostat device and change the temperature or let the
pre-set mourning temperature kick in.
- Maybe you stroll to the kitchen and grab your espresso, while listening to your Amazon Alexa device walk
you through the morning news.
- Soon, when you step out of your house you may enter your Google driverless car, which will take you to
work while recognizing all traffic patterns and schedule changes faster than any human ever could.
6. Ventral Stream
(recognizes objects,
ties words to what
youâre seeing)
Dorsal Stream
(recognizes objects
in physical space,
ties a 3-D image to
what youâre seeing)
Limbic System
(recognizes feelings attached to
what youâre seeing)
6
Eyes
(as light enters recognizes
geometry/shapes, sends
information 30 parts brain)
1
2
3 4
Source: TED Talk âThree ways the brain creates meaning.â
7. 7 Source: TED Talk âThree ways the brain creates meaning.â
Notes:
Source/Credit: TED Talk âThree ways the brain creates meaning.â
Link: https://www.ted.com/talks/tom_wujec_on_3_ways_the_brain_creates_meaning
Intro
- Letâs talk a little about how our brains processes information.
- Cognitive psychologists tell us that the human brain doesnât see the world as it is.
- Instead it actually creates a collection of âah-haâ moments as it discovers and processes
information.
Item #1 on slide - Eyes.
- Processing begins with the eyes
- Light enters and hits the back of the retina, circulates, and streams to the very back of
the brain at the primary visual cortex.
- Here the brain sees simple geometry, just shapes. But hereâs the really important
part...this area of the brain also sends information to 30 other parts of your brain.
- These other parts then piece together what youâre seeing in a âah-haâ experience.
8. 8 Source: TED Talk âThree ways the brain creates meaning.â
Notes:
Letâs talk about a few of the key parts of the brain that receive this information:
- Item #2 Ventral Stream - this is the part of the brain that recognizes a thing as a thing
(...thatâs a phone). Itâs the part of the brain thatâs activated when you call something by
a name - a word.
- Item #3 Dorsal Stream - this area locates objects in physical space. So right now
youâre looking at your screen maybe in front of a wall, and youâre seeing a 3D mental
map of this space. If you closed your eyes right now you could probably touch the
screen and the wall behind it.
- Item #4 Limbic System - deep inside the brain, this is a super old part of your brain is
what feels. So when you see a picture of your dog and feel love...or you see a picture
of your ex at McDonalds in your stories and feel...hangry?
So what can we learn from this?
- The human eye creates a map of what youâre seeing and the space around you by
identifying dozens or hundreds of objects.
- The brain then processes this information it sees to create one unified mental node of the
world around you.
- Quite simply, humans are amazing pattern-recognition machines.
10. 10
Notes:
1. Letâs run through a simple example - the act of reading.
2. You first recognize the patterns of individual letters
3. Then the patterns of individual words
4. Then groups of words together, then paragraphs, then entire chapters, and books overall.
- And take it from me, as the âold guyâ in the room, the books and articles you read now whether in the
classroom or not, all add up.
- Your brain looks at information and patterns across this massive library of things youâve read in your life.
11. IMDB Rating: 7.5 IMDB Rating 9.0
IMDB Rating: 7.4
11
vs
IMDB Rating: 9.2
Source: IMDB website. TED Talk âHow to use data to make a hit TV show.â
12. 12 Source: IMDB website. TED Talk âHow to use data to make a hit TV show.â
Notes:
Source/Credit: TED Talk âHow to use data to make a hit TV show.â
Link: https://www.ted.com/talks/sebastian_wernicke_how_to_use_data_to_make_a_hit_tv_show
1. Letâs go back in time to 2013. Before original shows online were a normal thing. Back then,
Amazon and Netflix set out to launch original TV Shows.
- At the time Amazon had a senior executive named Roy Price, who was in charge of picking the shows
content the company was going to create.
- Roy decided to to take a bunch of TV show ideas, through an evaluation picked 8, and for each of these 8
candidates created a pilot or first episode.
- Amazon then took these shows and put them online for millions of users to watch...for free.
- So millions showed up and watched these episodes and Amazon used machine learning and algorithms to
analyze in detail when someone pressed pause, when someone skipped a scene, what they skipped,
and what they replayed.
- After crunching all the data an answer emerges and Amazon looked to greenlight a sitcom about 4
Republican US senators. It was called âAlpha House.â How many of you have heard of it? (audience raises
hands)
- Well it was fantastically average. In TV land getting an IMDB rating of around 7.4 means youâre part of the
large pile of shows that are average. Alpha House was a 7.6
13. 13 Source: IMDB website. TED Talk âHow to use data to make a hit TV show.â
Notes:
2. Now in the same year, Netflix also sets out to launch original TV Shows.
- Ted Sarandos, Netflixâs Chief Content Officer, doesnât hold a competition.
- Ted worked with his team and they used the data to discover what kinds of content people liked, the ratings they gave,
what producers they liked, what actors and so on. They realized a show about a single Senator could be really successful.
They then found and revamped a British show called âHouse of Cards.â
- House of cards was a fantastic success. It has an IMDB rating of 9.0, well above the average, and is in the small sliver of
very successful shows.
- Think The Godfather versus The Accountant, when comparing the two shows.
3. The question of course is, what happened?
- Amazon conducted a crowdsourced process to decide on their show versus Netflix which had a human team look at machine
generated data about their subscribers to determine the âHouse of Cardsâ series could be successful with this same audience.
Today, Amazon follows a similar process looking at their Prime subscribers and put out shows that win awards.
- This is just one example, but there are many scenarios where the analysis of millions of data points with no human analysis
does not yield the best result.
4. Why is this the case?
- All data analysis involves taking a problem, ripping it apart and understanding its little pieces. Then putting these pieces
together to come to some conclusion.
- Machines conducting data analysis are not great at putting pieces back together and to drawing a conclusion.
- However, as we now know, the human brain is really good at that. Itâs all about pattern recognition.
- Even with incomplete information, especially if the human brain is that of an expertâs, humans can look at different pieces of
information and rebuild them into a single unified answer.
16. 16 Source: TED Talk âThe jobs weâll lose to machines and the ones we wonât.â
Notes:
Source/Credit: TED Talk âThe jobs weâll lose to machines - and the ones we wonât.â
Link:
https://www.ted.com/talks/anthony_goldbloom_the_jobs_we_ll_lose_to_machines_and_the_ones_w
e_won_t?language=en
1. Letâs start by talking about Artificial Intelligence and Machine learning.
- Machine learning is responsible for most of the disruption weâve seen to date. It's the most powerful branch
of artificial intelligence. It allows machines to learn from data and mimic some of the things that humans can
do.
- Machine learning started in the early '90s with relatively simple tasks like assessing credit risk from loan
applications or sorting mail by reading handwritten zip codes.
- Over the past few years, we have made dramatic breakthroughs. Machine learning is now capable of far,
far more complex tasks... like scanning your eyes and diagnosing diseases such as diabetic retinopathy or
even something like reading human essays and grading them just like a teacher would
2. Now, there have been some things that machines havenât been great at. Machines havenât been
able to handle things they haven't seen many times before.
- The fundamental limitations of machine learning is that it needs to learn from large volumes of past data.
Now, humans don't.
- We have the ability to connect broken pieces of information and solve problems weâve never seen before.
18. 18 Source: Google DeepMind materials and book âWhiplash - How to Survive Our Faster Future.â
Notes:
Source/Credit: Google Deepmind materials and book âWhiplash - How to Survive Our Faster Futureâ
by Jeff Howe and Joi Ito
Link: https://www.amazon.com/Whiplash-How-Survive-Faster-Future/dp/1455544590
1. Today, the limitation of machines not being able to handle the unfamiliar is also changing. Letâs
walk through an example.
2. What youâre seeing in the middle of this slide is the ancient Chinese board game called âGo.â
- Itâs probably one of the most complex games humans have ever devised.
- There are 10 to the power of 170 possible board positions. This is more than there are atoms in the
universe.
- Since there are so many possible positions, there is no way to calculate all possible moves.
- One of the reasons a robot hasnât been able to beat a human player...until it finally happened last year in
March of 2016.
3. The Google DeepMind algorithm beat a human 4 out of 5 times. Where did DeepMind come from
and how did it win?
- What is DeepMind? DeepMind was actually a British artificial intelligence startup founded by Dennis
Hassabis in 2010. Based here in London. Google acquired the company in 2014 for $500M and continued
the teamâs work with Dennis at the helm. Heâs a young guy and a genius, so for anyone thatâs interested in
AI, I would recommend googling him.
19. 19 Source: Google DeepMind materials and book âWhiplash - How to Survive Our Faster Future.â
Notes:
- When the team build the DeepMind algorithm, called AlphaGo, they didnât use brute force to calculate all the moves
it could make. As was done previously. Instead they used reinforcement learning and neural networks to mimic
the process of a human brain.
4. What is reinforcement learning?
- Unlike AI for Siri or IBM Watson, Deepmind uses deep reinforcement learning. The team started training the
algorithm by showing it 100,000 games. At first it just mimicked human players. Then it allowed the machine to play
itself 30 MILLION TIMES, using reinforcement learning the system learned to improve itself incrementally by avoiding
its errors and also by âreinforcingâ its wins.
- The machine knows a certain move resulted in a win more times in the past, than another move, and thus chooses
that move.
- DeepMind combined this memory system with an approach to AI called âneural networkingâ which mimics the
human brain and acts as a bridge between information we give the machine and itâs own memory system.
- In short the machine developed a subconscious that helped determine the moves it played. The same way a
human thatâs a master of the game has a subconscious containing all of itâs prior gaming history.
5. Whatâs next to DeepMind?
- Itâs been tested to play arcade games and is now moving on to immersive 3D games like Doom because itâs a
closer proxy to real life.
- Over time DeepMind will move into healthcare, robotics, computer vision, finance, and even news publishing and
writing.
22. 22
Notes:
1. Weâve already seen what machines can do when theyâre given a âmemoryâ so itâs not hard to
imagine a world maybe just a handful of years away where robots are as lifelike as they are in the
HBO show Westworld.
2. Letâs talk more about WestWorld.
- In the show the machines or âhostsâ as theyâre called are given an extra script of code called "reveries" or
revenant memories.
- These memories can be subtle actions the robots have taken in the past, or violent actions they've taken.
- These memories are like a subconscious and this single feature leads the robots to learn that they are in
fact...robots and that their world is designed by humans.
- Well things go dark from there. CLIP OF WESTWORLD:
https://www.youtube.com/watch?v=0CBRByBCcRU
3. So is that it? Are we all doomed?
- In the thousands of years of meaningful civilization...AI is the first thing that humans have created that
tends to function in ways humans canât predict. It can be scary.
- However, if you put aside the WestWorld scenario...of robot in a cowboy hat killing you in a saloon.
- You can imagine AI helping us to fix some of the biggest issues faced by the human race. It can help us
cure diseases and build things that would have taken us centuries to make otherwise. So the future of
humans vs. machines isnât all carnage :) THE END.