The AI Takeover in Hollywood by Yves Bergquist
- 2. ENTERTAINMENT TECHNOLOGY CENTER
Data & Analytics Project
Audience
Intelligence
Creative Marketing
Data/ML/AI applied to
Production
Project
Snowflake
Operations
Content
Security
AI Database
- 4. Leadership
Yves Bergquist
CEO
• Data Scientist.
• Director of the “Data & Analytics”
Project, Entertainment Technology
Center at USC.
• Leading researcher on applying
NLP to measure narrative
intelligence and its impact on
performance in marketing and film.
• Built AI applications which
understand human stories.
• Leading AI researcher with
hundreds of publications.
• Creator of the OpenCog AGI
project.
• Chairman of the Artificial Brain Lab,
Xiamen University.
• Chairman of the AI Society.
• Former Senior DARPA + IARPA
Researcher.
• PhD in Mathematics, Temple
University.
Dr. Ben Goertzel
CTO
Dr. Gian Gonzaga
Chief Data Scientist
• Former Head of Content
Algorithms, Netflix.
• Former Chief Scientist,
eHarmony.
• PhD in Psychology, UC
Berkeley.
Decades of applied AI experience
- 6. Examples of our work
We have built some of the most advanced AI platforms to date.
We built the AI
behind Hanson
Robotics’ “Sophia”
(the world’s most
advanced
humanoid robot)
Sophia was
interviewed by
Charlie Rose
on “60
Minutes”
RoboticsMedia
National
Security
An AI-driven application
that identifies deep
semantic narratives
around brands in social
media conversations and
assesses their
performance.
Finance
Entirely AI-powered
hedge fund (no humans
involved in trading).
AI application modeling
the human brain’s
probabilistic decision-
making process in the
parietal cortex, to seed
human-level cognition in
geospatial analysis
systems.
Profiled in “Wired”
Magazine
- 8. Narrow AI vs AGI
Narrow AI AGI
Machine Learning
Doesn’t generalize
Machine Understanding
Generalizes Across Domain
- 9. Our Unique Approach
Deep Neural Nets
+
Integrative, Neural-Symbolic
Cognitive Architectures
Deep neural nets currently touch only a
few of the numerous cognitive
processes underlying human
intelligence;
Integrative
neural-symbolic
cognitive architectures
attempt to embrace all of them
From Machine Learning to Machine Understanding
- 10. OpenCog stores and manipulates
knowledge in the form of weighed labeled
hypergraphs called “The AtomSpace”
OpenCog Features multiple cognitive
algorithms, each acting on different sorts of
knowledge within the common “AtomSpace”
The world’s most extensive open source AI architecture
OpenCog
- 13. Cognitive Approach
AI & NEUROSCIENCE
APPLIED TO
AUDIENCE
INTELLIGENCE
Understanding
Deep Human Drivers
Detecting / Classifying Them
in Natural Language
A New Age of Behavioral Communication
- 15. Measuring “interestingness”
(Schmidhuber)
low
compression
rate
high
compression
rate
there is an ideal compression rate (ratio of known to new) that
draws the brain to want to collect new training data
familiar people,
problems, characters,
stories, emotions and
situations
unfamiliar people,
problems, characters,
stories, emotions and
situations
known new
too known too newPERFECT RATIO
Neuroscience & AI Research Application
we built an application that mines social media
data to extract and score how interesting a
brand is based on the ratio of traditional /
emerging themes around that brand
Neuroscience R&D
- 16. Storygraphics Research
Our CEO, Yves Bergquist, has been conducting research at USC’s Entertainment Technology
Center on the performance of various narrative structures in feature length film, film trailers
and video ads
Sample Findings
Feature film audiences like when a movie ends with a “collective
community win” which end in a“collective win” vs a “personal win”
• Stories focused on “internal states” perform poorly.
• Main character mechanics that are visual in nature
(revenge, obtaining, etc.) perform very highly.
• Temporal elements (flashbacks, flash-forwards) perform
highly.
• 3-act structure does not work very well.
• Acts 2 and 3 are where most film stories fail at the box
office.
• Audiences care about solving plot problems at the
community level vs the individual level.
• Audiences don’t care much about heroes obtaining what
they set out to obtain.
• Highest-performing MBTI main character types: INTP,
ISFP, ENTJ.
the algorithmic structure of story.
- 17. Overall Story Signpost 1
The
Crucible,
Donnie
Darko,
Apt Pupil
Othello, The
Babadook,
Searching
for Bobby
Fischer, The
House of
Yes
Beginning of the Movie
My Big Fat Greek Wedding, The
King's Speech, There's
Something About Mary
The Dark Knight, The
Producers, How to Train
Your Dragon Juno,
Toy Story,
Frozen
In
Bruges,
Hamlet,
The
Importanc
e of Being
Earnest
Charlotte's Web,
Inside Man, Pride
and Prejudice
The Age of
Innocence,
The Silence of the
Lambs
- 20. the first AI-based enterprise database.
1. Knowledge Representation:
Rich semantic framework build on
top of a custom hypergraph DB
2. AI tools:
AGI Plugins
(neural-symbolic
architecture)
3. Interface:
ChatBot, Query language, Graph Viz UI
CortoDB:
• Hypergraph DB: next step beyond graph DBs - best
knowledge representation system for complex, dynamic
data.
• Powerful pattern matcher, to enable a very wide variety
of user queries to be answered flexibly and efficiently.
CortoAI:
• AI, AGI, ML plugins integrated with CortoDB for data
analysis, decision support and hypothesis formation.
Uniquely powerful AI tools:
• Evolutionary learning.
• Probabilistic logic.
• Neural nets.
• Integrated with TensorFlow/Theano, etc., for deep
learning-based data processing.
• Natural language processing plugins.
CortoChat:
• Enterprise chatbots with customizable dialogue, and the
ability to interact with the knowledge in CortoDB.
SOLUTION | CORTO
- 21. Video Data
social media data
Production Data
Financial Data
Now
• data in dozens of
silos.
• data in
thousands of
formats and
taxonomies.
Studio has few specialized experts and no
unified framework for representing and
analyzing its data. Data sits largely unused.
Then
• data is
represented in a
way that reflects
its semantics.
• relevant patterns
are found by AI
plugins.
• interaction via
query language,
graphical UI or
chatbot .
All studio data is represented in a
consolidated semantic hypergraph,
analyzed using AI plugins and flexibly
queryable through CortoChat
Market Segmentation Data
Post-Production Data
CASE STUDY | FILM STUDIO