The document discusses AAAI - EDGeS, which will take place March 27-29, 2023 in Burlingame, CA. It focuses on lifelong formal modeling agents and includes sections on why code is important, using code as knowledge representation, implementations, and the vision/goals. Specific topics discussed include Von Neumann's argument, biological inspiration, the complexity of search/forward running code, formal fields, concepts as a step to "language as code", the ARC challenge, using speech-to-text in their framework, bAbI as a "language as code" example, definitions of intelligence/understanding, agency/learning, and present/future goals.
Genuine love spell caster )! ,+27834335081) Ex lover back permanently in At...
Lifelong Formal Modeling Agents at AAAI-EDGeS 2023.pdf
1. AAAI - EDGeS
2023
Lifelong formal-modelling
agents
AAAI - EDGeS
Hyatt Regency San Francisco Airport, Burlingame, CA
March 27-29, 2023
AAAI Spring Symposium Series
Jacques (aka. Santiago) Basaldúa
5. AAAI - EDGeS
2023
1. Why code?
- Von Neumann's argument
- All models are code
- Biological inspiration
2. Code as Knowledge Representation
3. Putting Ideas into Action: Implementations
4. Our Vision: Purpose and Long-Term Goals
6. AAAI - EDGeS
2023
All models are code …
… just using the same sequence of operations on different parameters.
Is it good enough to wait for the convolutional network to learn how to multiply?
7. AAAI - EDGeS
2023
1. Why code?
- Von Neumann's argument
- All models are code
- Biological inspiration
2. Code as Knowledge Representation
3. Putting Ideas into Action: Implementations
4. Our Vision: Purpose and Long-Term Goals
12. AAAI - EDGeS
2023
1. Why code?
2. Code as Knowledge Representation
- Complexity of the search and forward running code
- Formal fields and what we cannot do with blocks
- Concepts as a step towards “language as code”
3. Putting Ideas into Action: Implementations
4. Our Vision: Purpose and Long-Term Goals
13. AAAI - EDGeS
2023
Macroscopic
Computer resources by the € Clock Cycles Evaluations @(1M/sec)
1 euro 4 x 1013
4 x 1010
1 million euros 4 x 1019
4 x 1016
1 billion euros 4 x 1022
4 x 1019
GDP of the whole planet 1014
(USD)
Age of the universe 1.4 x 1010
(years) 4.3 x 1017
(sec)
Weight of all biomass of the planet 5.6 x 1011
(ton) 5.6 x 1014
(Kg)
Human population
Human population 8 x 109
Doing something 1000 times (taking
pictures, buying, eating, …)
8 x 1012
Doing something a million times 8 x 1015
15. AAAI - EDGeS
2023
Combinatorial
• Tic-tac-toe positions 26·103
• Chess game tree complexity 10120
• Fitting GPT3 s^(170·10⁹) (s == size of “parameter”) if there was only 1 solution
• Writing Shakespeare at random
• …
16. AAAI - EDGeS
2023
Formal
A(4) = 4!!!! cannot be computed on Earth
B(9) = A(A(A(A(A(A(A(A(A(9))))))))) cannot be computed using the whole
universe as a computer
C(9) = B(B(B(B(B(B(B(B(B(9))))))))) is still a finite number
17. AAAI - EDGeS
2023
Countable, Continuum and the
Power Set of ℝ
Countable:
0
1
2
Continuum:
Power Set of ℝ:
• The set of all integers
• The set of rational numbers
• The set of tuples of integers
. . .
• The set of all real numbers
• The set of tuples of real numbers
. . .
• The set of all sets of real numbers
• The set of functions from ℝ → ℝ
. . .
19. AAAI - EDGeS
2023
1. Why code?
2. Code as Knowledge Representation
- Complexity of the search and forward running code
- Formal fields and what we cannot do with blocks
- Concepts as a step towards “language as code”
3. Putting Ideas into Action: Implementations
4. Our Vision: Purpose and Long-Term Goals
21. AAAI - EDGeS
2023
An object cannot possibly
represent a concept. At the
minimum, we need a data
structure that can dynamically
grow (e.g., a graph)
Formal fields and what we
cannot do with blocks
22. AAAI - EDGeS
2023
1. Why code?
2. Code as Knowledge Representation
- Complexity of the search and forward running code
- Formal fields and what we cannot do with blocks
- Concepts as a step towards “language as code”
3. Putting Ideas into Action: Implementations
4. Our Vision: Purpose and Long-Term Goals
24. AAAI - EDGeS
2023
1. Why code?
2. Code as Knowledge Representation
3. Putting Ideas into Action: Implementations
- The ARC challenge
- Speech to text using in our framework
- bAbI as our “language as code” bAbY
4. Our Vision: Purpose and Long-Term Goals
34. AAAI - EDGeS
2023
1. Why code?
2. Code as Knowledge Representation
3. Putting Ideas into Action: Implementations
- The ARC challenge
- Speech to text using in our framework
- bAbI as our “language as code” bAbY
4. Our Vision: Purpose and Long-Term Goals
36. AAAI - EDGeS
2023
1. Why code?
2. Code as Knowledge Representation
3. Putting Ideas into Action: Implementations
- The ARC challenge
- Speech to text using in our framework
- bAbI as our “language as code” bAbY
4. Our Vision: Purpose and Long-Term Goals
37. AAAI - EDGeS
2023
The primitives in this example
are: {person, location, to move
to, to be at, to be with, then,
and, they, where, who}
Using blending, abstraction
and interrogation the text runs
as code and generates the
answers.
“I assume Jabari is a person” is
a question asked by the
blender since Jabari is an
unseen word.
bAbI as our “language as code” bAbY
43. AAAI - EDGeS
2023
1. Why code?
2. Code as Knowledge Representation
3. Putting Ideas into Action: Implementations
4. Our Vision: Purpose and Long-Term Goals
- We need specific definitions of intelligence and understanding
- Agency and learning
- Present and future goals
44. AAAI - EDGeS
2023
We need specific definitions of
intelligence and understanding
Once we use code as our knowledge representation: Understanding is
being able to successfully run and evaluate code in order to optimize it.
45. AAAI - EDGeS
2023
1. Why code?
2. Code as Knowledge Representation
3. Putting Ideas into Action: Implementations
4. Our Vision: Purpose and Long-Term Goals
- We need specific definitions of intelligence and understanding
- Agency and learning
- Present and future goals
46. AAAI - EDGeS
2023
Agency and learning
Lifelong systems
exploit what they
have learned
requiring less
learning, not
more.
Most system in
production today,
don’t learn at all.
47. AAAI - EDGeS
2023
1. Why code?
2. Code as Knowledge Representation
3. Putting Ideas into Action: Implementations
4. Our Vision: Purpose and Long-Term Goals
- We need specific definitions of intelligence and understanding
- Agency and learning
- Present and future goals
48. AAAI - EDGeS
2023
Present and future goals
We need interpretable natural language
systems do to human/computer interaction.
This Not this
52. 52
This is the BBVA AI Factory
Our offices
Who we are
We produce data products
We work to respond to the bank’s priorities creating
new products based on Artificial Intelligence
Multidisciplinary teams
We share our day-to-day work among data scientists, engineers, AI
architects, domain experts and other professionals
We are a fully-owned subsidiary of BBVA
BBVA AI Factory is the global development center where BBVA
builds the Artificial Intelligence capabilities