We know that we are in an AI take-off, what is new is that we are in a math take-off. A math take-off is using math as a formal language, beyond the human-facing math-as-math use case, for AI to interface with the computational infrastructure. The message of generative AI and LLMs (large language models like GPT) is not that they speak natural language to humans, but that they speak formal languages (programmatic code, mathematics, physics) to the computational infrastructure, implying the ability to create a much larger problem-solving apparatus for humanity-benefitting applications in biology, energy, and space science, however not without risk.
1. Denver CO, 8 Jul 2023
Slides: http://slideshare.net/LaBlogga
Melanie Swan, PhD, MBA
DIYgenomics.org (Research Lead)
University College London (Research Associate)
“Nothing is more abstract than reality”
– Giorgio Morandi
The Math Take-off
Space Humanism, AI-Quantum Computing Convergence,
and the Future of Intelligence
2. 8 Jul 2023
AI Math Agents 1
Goal: solve biosystem pathology (aging, Alzheimer’s disease)
with physics mathematics (renormalized multiscalar entropic
near-far correlations) or other AI-aided mathematical analysis
AdS/Biology
Research Program
2015 2019 2020
Blockchain Blockchain
Economics
Quantum
Computing
Quantum Computing
for the Brain
2022
AdS/Biology: application of AdS/CFT (anti-de Sitter space/conformal field theory) bulk-boundary modeling to biosystems
AI Math Agents:
https://arxiv.org/abs/2307.02502
https://huggingface.co/papers/2307.02502
AI Genomics and Alzheimer’s Disease:
https://www.diygenomics.org/files/AI_Math_Agents_poster_AAIC2023.pdf
3. 8 Jul 2023
AI Math Agents
AI Science Project Landscape
2
Sources: https://openai.com/blog/chatgpt-plugins#code-interpreter; Boiko et al (2023). Autonomous scientific research capabilities of
LLMs. arXiv: 2304.05332; https://opencatalystproject.org/; Tu et al (2023). Towards Generalist Biomedical AI. arXiv:2307.14334v1;
Mialon et al (2023). SSL with Lie Symmetries for Partial Differential Equations. arXiv:2307.05432v1. WizardLM: 2304.12244.
DeepMind Med-PaLM
biomedical AI
Meta AI/CMU Open
Catalyst: 1000x faster
molecular dynamics
Code Interpreter (OpenAI): using ChatGPT to upload files, analyze
data, create charts, solve math problems, edit files, produce code
WizardLM: LLM creating instructions for other LLMs
(math, code, reasoning, complex data formats)
Quick move into biochemistry, biophysics
with LLM Math Agent functionality
Lie symmetry PDE
solving network
https://ibm.co/3XviRVV
Smart-
biology.
com
4. 8 Jul 2023
AI Math Agents
AI Genomics
Multiscalar approach
Gene regulatory elements
influence expression in
cell types and tissues
Alzheimer’s disease
2,676 differentially
expressed genes
Up/downregulate
proteins in cell types
Upregulation of APOD, INSR
and COL4A1 in brain tissue
Downregulation of SLC6A1 in
GABAergic neurons and
astrocytes, PDGFRB in
pericytes and ABCB1, and
ATP10A in endothelial cells
3
SNP: single nucleotide polymorphism Sources: Kellis Lab: Sun et al. (2023.) Single-nucleus multi-region transcriptomic analysis of brain
vasculature in Alzheimer’s disease. Nat Neurosci. 26, 970–982. https://doi.org/10.1038/s41593-023-01334-3. Cirillo et al. (2017). A
Review of Pathway-Based Analysis Tools That Visualize Genetic Variants. Front. Genet. 8:174. doi: 10.3389/fgene.2017.00174.
Pathway
Protein
Blood plasma, CSF
RNA
Expression
miRNA, mRNA
DNA
Gene, Variants (SNPs), Gene
Regulation, Epigenomics
AI Genomics Research Copilot
5. 8 Jul 2023
AI Math Agents 4
The message of generative AI and LLMs (large language models like GPT) is not that they speak natural
language to humans, but that they speak formal languages (programmatic code, mathematics, physics) to
the computational infrastructure, implying the ability to create a much larger problem-solving apparatus for
humanity-benefitting applications in biology, energy, and space science, however not without risk
Thesis
Formal
Language:
Math, Physics,
Software Code
Natural
Language
Human
Computational Infrastructure
Interface Reality
AI
A math take-off is using math as a formal language, beyond the
human-facing math-as-math use case, for AI to interface with the
computational infrastructure
We know that we are in an AI take-off,
what is new is that we are in a math take-off
6. 8 Jul 2023
AI Math Agents
Language Space Program Space Mathematics
Space
infinite
infinite
A. Software 1.0 (human-discovered)
B. Software 2.0 (machine-derived)
A.
B. automated
theorem proving
human-discovered
theorems
computer
algebra systems
Existing Spaces
New Spaces
AI Space
Computational
Complexity
Space
Planck Space
AI Science
Space
Now Treating the Entire Possibility Space
Source: Swan, M., Kido, T., Roland, E. & dos Santos, R.P. (2023). AI Math Agents: Computational Infrastructure, Mathematical
Embedding, and Genomics.
Digitizing a possibility space (e.g. natural language) makes it formal
7. 8 Jul 2023
AI Math Agents
6
Reality Interface
Abstraction: Mathematics is the Interface
Multiscalar Renormalization
One System Two Modes
Mathematics as a High-order Lever for Interacting with Reality
Data de-emphasized in the Math-Data Relation
Big Data -> Big Math Era
AI “speaking” formal languages implies math as a higher-
order lever for interacting with reality (beyond data)
Source: Swan, M., Kido, T., Roland, E. & dos Santos, R.P. (2023). AI Math Agents: Computational Infrastructure, Mathematical
Embedding, and Genomics.
8. 8 Jul 2023
AI Math Agents
New Conceptualization of Math
Traditional conceptualization
Practical math-as-math: build bridges, space shuttles
Foundational reality has a mathematical structure
Mathematical universe hypothesis
Quark properties are quantitative (mass, charge, spin)
Expanded conceptualization to also include
Math as a language
A formal language for human-AI entities to formulate problems
A language for AI to speak to the computational infrastructure
Math as a means not an end
Mobilized as a digital tool, as software is a digital tool
Math as a framework
Math as “truthier” content: high-validation, subject to proof
7
2014
F(x)
math-certified
F(x)
Penrose
tile
9. 8 Jul 2023
AI Math Agents
Math Agent: AI agent operating in digital mathematical
domain to identify, analyze, integrate, write, discover,
solve, prove, and steward mathematical ecologies
AI Math Agents
Source: Swan, M., Kido, T., Roland, E. & dos Santos, R.P. (2023). AI Math Agents: Computational Infrastructure, Mathematical
Embedding, and Genomics.
Mathematical Embedding:
476-equation ecology (LaTeX)
(SymPy)
Mathematical embedding: math entity
(symbol, equation) represented as a
character string in vector space for
high-dimensional AI analysis
Mathematical ecology (mathscape):
set of related mathematical equations
Equation Cluster: similar equations
grouped in mathematical ecology
embedding visualization
(LaTeX)
10. 8 Jul 2023
AI Math Agents 9
Citizen 2 heterozygous (1 ALT allele) SNPs in Illumina VCF file (Legend: Cit2-1)
Citizen 2 homozygous (2 ALT alleles) SNPs in Illumina VCF file (Legend: Cit2-2)
Genes: APP,
ASXL3, ABCA7,
SLC24A4, ANK3
PLCG2
Embedding Visualization examples with Academic Papers as the Data Corpus
AdS/CFT Equation Clusters in Embedding Visualization (LaTeX and SymPy)
Source: AdS/CFT: Kaplan, J. (2016). Lectures on AdS/CFT from the bottom up. Johns Hopkins Lecture Course.
https://www.diygenomics.org/files/AI_Math_Agents_poster_AAIC2023.pdf
The Mathematical Embedding
11. 8 Jul 2023
AI Math Agents 10
Citizen 2 heterozygous (1 ALT allele) SNPs in Illumina VCF file (Legend: Cit2-1)
Citizen 2 homozygous (2 ALT alleles) SNPs in Illumina VCF file (Legend: Cit2-2)
Genes: APP,
ASXL3, ABCA7,
SLC24A4, ANK3
PLCG2
Source: AdS/CFT: Kaplan, J. (2016). Lectures on AdS/CFT from the bottom up. Johns Hopkins Lecture Course.
https://www.diygenomics.org/files/AI_Math_Agents_poster_AAIC2023.pdf
The Mathematical Embedding
Annotated equation
clusters illustrate
(a) how similar groups
of equations are
grouped in the
embedding
method and
(b) the mouse-over
view of equation
images by
equation number
(OpenAI inlay from
previous figure)
12. 8 Jul 2023
AI Math Agents 11
Citizen 2 heterozygous (1 ALT allele) SNPs in Illumina VCF file (Legend: Cit2-1)
Citizen 2 homozygous (2 ALT alleles) SNPs in Illumina VCF file (Legend: Cit2-2)
Genes: APP,
ASXL3, ABCA7,
SLC24A4, ANK3
PLCG2
Source: https://www.diygenomics.org/files/AI_Math_Agents_poster_AAIC2023.pdf
AD, PD, ALS: Alzheimer’s disease, Parkinson’s disease, Amyotrophic lateral sclerosis
Mathematical Ecology analysis: math + data
Mathematical Ecologies (a) Alzheimer’s + SIR Model (control math); (b) Chern-Simons + AD SNPs
(a) AdS/CFT Mathematical Ecologies + AD SNPs; (b) SIR Mathematics; (c) Multi-disease Genomic view: AD, PD, ALS
13. 8 Jul 2023
AI Math Agents 12
Citizen 2 heterozygous (1 ALT allele) SNPs in Illumina VCF file (Legend: Cit2-1)
Citizen 2 homozygous (2 ALT alleles) SNPs in Illumina VCF file (Legend: Cit2-2)
Genes: APP,
ASXL3, ABCA7,
SLC24A4, ANK3
PLCG2
Source: https://www.diygenomics.org/files/AI_Math_Agents_poster_AAIC2023.pdf
AD, PD, ALS: Alzheimer’s disease, Parkinson’s disease, Amyotrophic lateral sclerosis
Alzheimer’s Genomics Precision Health
Embeddings Visualization of Data: Alzheimer’s SNPs applied to Citizen 1, Citizen 2 Precision Health initiative
Each individual is
homozygous (two
alternative alleles)
for different subsets
of genes suggesting
a starting-point for
personalized
intervention
Citizen 2 is homozygous for cancer-upregulated
membrane proteins (TREM) and cytokine-dependent
hematopoietic cell linkers (CLNK)
Both are homozygous for the solute carrier protein (SLC24A4) and the intracellular trafficking protein nexin (SNX1).
Citizen 1 is homozygous for more immune system related
genes (CD33, HLA-DRB1), and Alzheimer’s-related
clathrin binder (PICALM)
Alzheimer’s disease genomic risk is analyzed for two precision health participants with whole-human genome sequencing
An embedding visualization is performed for all GWAS-linked Alzheimer’s disease SNPs and presented for Citizen 1 and Citizen 2’s
heterozygous (one alternative allele) and homozygous (two alternative alleles) SNP
14. 8 Jul 2023
AI Math Agents
Source: Karpathy, A. (2017). Software 2.0. Medium. 11 November 2017. https://karpathy.medium.com/software-2-0-a64152b37c35.
Software 2.0: machine designed & programmed code
Machine coding: AlphaCode, Codex API (Github Copilot)
Search interface for internet-available code
Ability to seed code for new applications
New software development paradigm
Human specifies
Data, objective, framework, problem space
Machine learning optimizes
Node weights, network architecture
Algorithm for compilation and transfer
Algorithms more effective at code-writing than humans
Theorem-proving, code security audit, bug fixing
Software Coding Copilot
Software 2.0
Algorithms can explore a
larger possible program space
15. 8 Jul 2023
AI Math Agents
Reality Interface
14
Representation
Kantian
Goggles: the
manifold of
Space and
Time
Perception
Human
Kantian goggles of the perceptual manifold
Any object appears in some space and some time
We cannot know the “thing in itself” only our representations of it
Interface Reality
Human
16. 8 Jul 2023
AI Math Agents
Reality Interface
15
Representation
Perception
Human
Projects extending Kantian goggles with telescopes and
microscopes, now into relativistic and quantum domains
Interface Physical Reality
Human
Classical
Relativistic
Quantum
Kantian
Goggles: the
manifold of
Space and
Time
Classical
Relativistic
Quantum
17. 8 Jul 2023
AI Math Agents
New ideas of where we fit
The Large and Small Scale Universe
16
Scale Measure Comment
1 5.1 x 1096 Planck density Kg/Meter3 Density of the universe immediately after the Big Bang
2 1 x 1080 Particles Total particles in the observable universe (est.)
3 1 x 1014 Cells Cells in the human body (9 out of 10 are bacteria)
4 8 x 1010 Stars Number of stars in the Milky Way galaxy (est.)
5 1 x 102 Meter Earth Earth’s atmosphere: 10,000 ft life support, 62 mi to space
6 1 x 101 Meter Human Human-scale: Classical Mechanics
7 1 x 10-9 Nanometer Atoms Quantum mechanics (nanotechnology)
8 1 x 10-12 Picometer Ions, photons Optics, photonics
9 1 x 10-15 Femtometer Subatomic Gauge theories
10 1 x 10-35 Planck scale Meters Smallest known length scale
11 5.4 x 10-44 Planck time Seconds Shortest meaningful interval of time
Source: The Universe by Numbers. https://www.physicsoftheuniverse.com/numbers.html
Humans require specialized conditions to survive (unlike amoebas or cockroaches)
Large-scale:
General
Relativity
(GR)
Small-scale:
Quantum
Mechanics
(QM)
Human-scale:
Classical
Mechanics
Quantum mechanics, classical mechanics, general relativity
Quantum effects visible at 10-9 m
Relativistic effects present at any speed (matter of precision)
Classical
Relativistic
Quantum
18. 8 Jul 2023
AI Math Agents
AI raises the Definition of Intelligence
Intelligence: ability to learn, understand, and think (OED)
Artificial Intelligence (AI): technology with capabilities
traditionally considered to be human
Knowledge: the sum of relationships in information
Knowledge layer defined in the computational infrastructure
17
Consciousness
Understanding
Knowledge
Information
Data
March towards “human” capabilities
19. 8 Jul 2023
AI Math Agents 18
Classical Intelligence
Scale-free
Intelligence
Moore’s Law Curve:
Intelligence
Quantum
Intelligence
Classical Intelligence
Quantum Intelligence
Scale-free Intelligence
Time and Space Properties:
spherical-flat-hyperbolic space,
simultaneous time
Scale-free intelligence: ability to learn and
problem-solve in any physical regime
Relativistic Intelligence
Domain-specific time and
space, and matter properties
Domain-specific time and
space, and matter properties
Domain-specific time and
space, and matter properties
Intelligence as a Generic Capability
Need for in-situ
autonomous agent
decision-making
20. 8 Jul 2023
AI Math Agents 19
Classical Intelligence
Computational
Intelligence
Moore’s Law Curve:
Intelligence
Quantum
Intelligence
Classical Intelligence
Quantum Intelligence
Scale-free Intelligence
Time and Space Properties:
spherical-flat-hyperbolic space,
simultaneous time
Scale-free intelligence: ability to learn and
problem-solve in any physical regime
Relativistic Intelligence
Domain-specific time and
space, and matter properties
Domain-specific time and
space, and matter properties
Domain-specific time and
space, and matter properties
Intelligence as a Generic Capability
Computational
Intelligence
Ability to learn and
problem-solve
systematically in
formal environments
Scale-free
Intelligence
Mathematical
Intelligence
Ability to learn and
problem-solve, and
create/discover in
math environments
21. 8 Jul 2023
AI Math Agents
Agenda
AI (Artificial Intelligence)
AI-QC Convergence
QC (Quantum Computing)
AI Alignment and Space Humanism
20
22. 8 Jul 2023
AI Math Agents
What is the Purpose of AI?
21
Reorienting the Human-AI relation
1. The big offload
2. The big merge
Domain Classical
Classical intelligence
DeQ1
Zone
Quantum2
Quantum intelligence
AI
3. Precision tasks and knowledge-generation
2. Cognitive labor, computational contracts
1. Data informatics, physical labor, virtual labor
Information Science Stack
1Dequantization Zone: sufficiency of classical methods demonstrated (computation, biology)
2Quantum applications: quantum sensing, quantum machine learning, quantum dynamics simulation, quantum cryptography
Source: Concentric circles of knowledge: one potential purpose of AI (Demis Hassabis, DeepMind)
The totality of all knowledge
Knowledge that can be understood by the human mind
Knowledge that is currently understood by the human mind
Concentric Circles of Knowledge
23. 8 Jul 2023
AI Math Agents
The big offload
The AI Stack: Moore’s Law Curve of AI
22
Knowledge
Generation
disease cure,
theorem-proving
Cognitive Labor-
outsourcing, Virtual Labor
Computational Contracts
Moore’s Law Curve of AI
Data Analytics
Physical Labor-outsourcing
The AI Stack
Precision Tasks
LASIK eye surgery,
system control, atomic
manufacturing,
autonomous driving,
software programming,
mathematics
Data analytics, informatics
Physical labor-outsourcing
Cognitive labor outsourcing
Virtual labor
Computational contracts
Precision tasks
Software programming
Knowledge generation
Large-scale problems: space, biology, energy
Outsourcing classes of tasks to technology
24. 8 Jul 2023
AI Math Agents
Potential for Al-Facilitated Science
23
Formulate
research
question…
Design
polygenic study
protocol…
Suggest
theoretical
underpinning…
Analyze
study
results…
Draft
paper
skeleton…
Generate
slides for
colloquium…
Analyze
pathway…
Oncology
Immune
System
Update
research
agenda…
Research Copilot: “Microsoft 365 Copilot” for Science
Integrated AI learning of multi-modal health data streams
Research Copilot
Public Private
Select Data Corpus
Mockup only
25. 8 Jul 2023
AI Math Agents
Knowledge Society
Knowledge platforms
Wikipedia: interface for knowledge access
Coursera (MOOCs): interface for knowledge learning
Research Copilot: interface for knowledge generation
24
Wikipedia:
Knowledge Access
Coursera (MOOCs):
Knowledge Learning
Research Copilot:
Knowledge Production
Science Knowledge Graph
Space
Research Copilot
Biology Energy
Copilot: active interface on a data corpus
Knowledge Society: one that uses
knowledge to improve the human condition
26. 8 Jul 2023
AI Math Agents 25
Human-AI entities as the competitive unit
Digital knowledge prosthesis (phone: external; BCI: internal)
Conducting science
Executing experiments
Publishing results
Founding startups
Collaborating with others
Human-AI Entity
Human AI
Ishiguro, 2021
Literature Examples:
Personal AI agents
Humans and other objects
equipped with WAIs (weak AIs)
Divya, 2021 Schroeder, 2005
Dystopian Realistic Utopian
The big merge
Human-AI Entities
BCI: brain-computer interface: direct communication pathway between an enhanced or wired brain and an external device
Source: Swan, M. & dos Santos, R.P. (2023). Quantum Intelligence: Responsible Human-AI Entities, AAAI, San Francisco CA 27
Mar 2023. https://www.slideshare.net/lablogga/quantum-intelligence-responsible-humanai-entities
27. 8 Jul 2023
AI Math Agents
The big merge: Human-AI Entities
Global Cryonics Study (n = 316)
Attitudes towards personal identity re: brain and body
Brain
This physical brain is the “source of me” now (75%)
These specific memories are the future “source of me” (87%)
Body
This physical body is “part of me” now (68%)
This physical body is “part of me” in the future (14%)
26
Source: Swan, M. (2019). Worldwide Cryonics Attitudes About the Body, Cryopreservation, and Revival: Personal Identity Malleability
and a Theory of Cryonic Life Extension. Sophia International Journal of Philosophy and Traditions. 58:699–735. Springer Nature B.V.
https://link.springer.com/article/10.1007/s11841-019-0727-4.
A sense of Personal Identity Malleability
28. 8 Jul 2023
AI Math Agents
Generative AI and LLMs
LLMs (large language models)
Computerized language models
Generated with transformer neural networks
Billions of parameters
Pre-trained on large data corpora
GPT-4 (OpenAI), LaMDA (Google), LLaMA (Meta AI)
Transformer neural networks
Whole data corpus processed simultaneously to
analyze connections between data elements
27
Midjourney (CO state
fair winner 5 Sep 2022)
Knowledge Era Currency Basis of Knowledge Valorization Incentive
1 Renaissance Scribes Closed Information Rote memorization Control access to information
2 Information Society Open Information Find and synthesize information Share information, collaborate
3 Post-plagiarism Society Open Ideas Apply ideas to problems Large-scale problem resolution
Shifting definition of knowledge: ability to memorize -> synthesize information -> deploy ideas
GPT-3: 175 billion parameters
GPT-4: 170 trillion parameters
1,000x bigger
Parameter: learned system weight
29. 8 Jul 2023
AI Math Agents
Computational Infrastructure
Classical
Computing
Super-
computing
DNA Nanotechnology,
Spiking Neural
Networks
Quantum
Computing
28
Mobile
Existing Emerging
Smartphone,
Tablet, Watch,
BCI, headset
Biology
Computing
Platforms
Formal
Languages
Smart
Network
Technologies
Interface
Technologies
Blockchain
Machine Learning
AI (artificial intelligence): Siri -> Alexa -> chatGPT
chatbot copilot: active interfaces on data corpora, formal languages, smart network computation
30. 8 Jul 2023
AI Math Agents
Computational Infrastructure
Classical
Computing
Super-
computing
DNA Nanotechnology,
Spiking Neural
Networks
Quantum
Computing
29
Mobile
Existing Emerging
Smartphone,
Tablet, Watch,
BCI, headset
Biology
Computing
Platforms
Formal
Languages
Smart
Network
Technologies
Interface
Technologies
Blockchain
Machine Learning
AI (artificial intelligence): Siri -> Alexa -> chatGPT
chatbot copilot: active interfaces on data corpora, formal languages, smart network computation
31. 8 Jul 2023
AI Math Agents
Agenda
AI (Artificial Intelligence)
AI-QC Convergence
QC (Quantum Computing)
AI Alignment and Space Humanism
30
32. 8 Jul 2023
AI Math Agents
Technology Take-offs
31
Computer: punch card -> mainframe -> PC -> smartphone
Cell phone
Internet
AI
Quantum Computing
2023
estimated
Accelerated global deployment of technology that may
ultimately impact nearly all persons and areas of life
33. 8 Jul 2023
AI Math Agents
AI-QC Convergence - Basic
32
AI
Artificial Intelligence
QC
Quantum Computing
Quantum
Machine
Learning
QML
Quantum Machine Learning (QML): running machine
learning algorithms in a quantum environment
450,000 users
Sources: Stanford Global AI Vibrancy Tool: US and China lead AI innovation https://aiindex.stanford.edu/vibrancy/ McKinsey (Jun
2022): https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-next-frontier-for-ai-in-china-could-add-600-billion-to-its-
economy https://www.computerweekly.com/news/252527998/Finland-connects-a-quantum-computer-to-a-supercomputer
Finland: quantum computer HELMI (“Pearl”)
connected to supercomputer LUMI (“Snow”)
34. 8 Jul 2023
AI Math Agents
Quantum Research Copilot for Biology
33
AI
Artificial
Intelligence
QC
Quantum
Computing
Biology
“Gato Cell” for Biology concept
Reinforcement learning agent for data-intense quantum
environment of multi-modal health data stream aggregation
Genomics, epigenetic methylations, imaging, biomarker
Problem Implicated Area
1 85% drug targets ineffective Quantum chemistry
2 Early cancer detection Computational biology
3 Alzheimer’s disease Dx no Rx Computational neuroscience
4 Heart disease event prediction Computational biology
Disease resolution might be facilitated
with scalable quantum computing
Top Killers
Prevention Cure
Disease Solver Copilot
Cell
Suggested
concept only
Research Copilot for Biology
35. 8 Jul 2023
AI Math Agents
AI-QC Convergence - Advanced
Generative AI to extend quantum computing
Advance quantum error correction
Write software for quantum computers
Discover new quantum algorithms
AI-QC partnership
New infrastructure (nHITL: no human in the loop)
Quantum not just nice but necessary
New modes of energy efficiency
Computational infrastructure:
10% energy consumption:
Imperative to learn from nature
Synaptic switching: 0.77 attojoules
(Shankar, 2022)
34
Sources: IBM Quantum, https://quantumai.google/qecmilestone; Shankar, S. (2021). Lessons from Nature for Computing Looking
beyond Moore’s Law. IEEE. https://ieeexplore.ieee.org/document/9622865; 3-09-22 Physics Colloquium - Sadasivan Shankar,
SLAC Stanford/Harvard, https://www.youtube.com/watch?v=4xULxynEWEc
Biomimicry implicated: the brain is
extremely energy-efficient
36. 8 Jul 2023
AI Math Agents
Agenda
AI (Artificial Intelligence)
AI-QC Convergence
QC (Quantum Computing)
AI Alignment and Space Humanism
35
37. 8 Jul 2023
AI Math Agents 36
Basic Concept
What is Quantum Computing?
Computing: change of state between 0/1
Move information around & and perform a computation
Quantum: use atoms, ions, photons to compute
Classical computing: serial not parallel
Quantum computing: treat more than one status at the
same time, compute all transactions simultaneously
Fundamentally, a different way of computing
Degreed physicists sought as product managers (Gartner)
Shift big data analysis to quantum to find hidden correlations
Source: Fowler, A.G., Mariantoni, M., Martinis, J.M. & Cleland, A.N. (2012). Surface codes: Towards practical large-scale
quantum computation. Phys Rev A. 86(032324).
38. 8 Jul 2023
AI Math Agents
Quantum Scale
37
QCD: Quantum Chromodynamics
Subatomic particles
Matter particles: fermions (quarks)
Force particles: bosons (gluons)
Scale Entities Physical Theory
1 1 x101 m Meter Humans Newtonian mechanics
2 1 x10-9 m Nanometer Atoms Quantum mechanics
(nanotechnology)
3 1 x10-12 m Picometer Ions, photons Optics, photonics
4 1 x10-15 m Femtometer Subatomic particles QCD/gauge theories
5 1 x10-35 m Planck scale Planck length Planck scale
Atoms Quantum objects:
atoms, ions,
photons
“Quantum” = anything at the scale of
atomic and subatomic particles (10-9 to 10-15)
Theme: ability to study and manipulate
physical reality at smaller scales
Study phenomena (e.g. neurons) in the native
3D structure of physical reality
39. 8 Jul 2023
AI Math Agents
Atoms/ions Controlled with Lasers/Fields
38
Source: Jackson, M. (2022). Introduction to Quantinuum and TKET. PIRSA 13 Sep 2022. https://pirsa.org/22100088.
1. Trap one Ytterbium ion
2. Entangle two Ytterbium ions
3. Conduct circuit-based computation
1. 2.
3. Conduct circuit-based computation
40. 8 Jul 2023
AI Math Agents
Quantum Computing
Microsoft
IBM
Rigetti
41. 8 Jul 2023
AI Math Agents
Using a Quantum Computer
40
Source: D-Wave Systems, Inc. https://cloud.dwavesys.com/leap/resources/demos
42. 8 Jul 2023
AI Math Agents
Quantum Computing
Sources: IBM Quantum, https://quantumai.google/qecmilestone
Dequantization: certain purported quantum speed-ups reclassed to the Dequantization Zone per sufficiency of classical methods
Google Quantum Computing Roadmap
Hardware needs: error correction
2023e: 1000 qubits (Google, IBM)
2030e: million-qubits (general-purpose) (IBM)
Software needs: algorithms
Software 2.0 implication
AI discovers new quantum algorithms
AI writes code for quantum computing
Dequantization trend (classical sufficiency)
Cybersecurity: million qubits
needed to break RSA
US NIST quantum-safe
algorithms hacked but
“in progress”
IBM Quantum
Computing Roadmap
43. 8 Jul 2023
AI Math Agents
Status
Quantum Computing
Various cloud quantum computing platforms available
Critique: so far quantum computing only useful in a few
cases such as optimization problems (linear algebra)
42
Open Quantum Testbeds
(Sandia, LBL)
Industry (Cloud-based)
Source: Landahl, A. (2022). Sandia National Laboratories.
44. 8 Jul 2023
AI Math Agents
Quantum Properties
43
1. Superposition: a quantum system can exist in
several separate quantum states simultaneously
2. Entanglement: two interconnected
quanta maintain their connection
regardless of the distance between them
5. Quantum tunneling: a particle is able to penetrate
through a potential energy barrier higher in energy
than the particle’s kinetic (motion) energy
4. Symmetry: properties that remain
invariant across scale tiers
3. Interference (coherence): an object’s wave
property is split in two, and the two waves
cohere (reinforce) or interfere with each other
Source: Mazzoccoli, G. (2022). Chronobiology Meets Quantum Biology: A New Paradigm Overlooking the Horizon? Front. Physiol.
13:892582. doi: 10.3389/fphys.2022.892582.
Schrödinger Cat State: quantum state comprised of
two opposed conditions at the same time
45. 8 Jul 2023
AI Math Agents
A qubit (quantum bit) is the basic unit of
quantum information, the quantum version
of the classical binary bit
44
What is a Qubit?
Bit exists in a
single binary state
(0 or 1)
Qubit exists in a state of superposition, at
every location with some probability, until
collapsed into a measurement of 0 or 1
Implication: test permutations simultaneously
Classical Bit Quantum Bit (Qubit)
Sources: https://www.newsweek.com/quantum-computing-research-computer-flagship-eu-452167: Dawid Carrasquilla, Carleo,
Wang et al. (2022). Modern applications of machine learning in quantum sciences. arXiv: 2204.04198.
Practical example: 1-qubit
quantum machine learning
classification task
46. 8 Jul 2023
AI Math Agents
Quantum: Many Potential Speed-ups
1. Bit (0 or 1)
2. Qubit (0 and 1 in superposition)
3. Qudit (more than 2 values in superposition)
Microchip generates two entangled qudits each with 10
states, for 100 dimensions total, for more than six
entangled qubits could generate (Imany, 2019 )
4. Optics (time and frequency multiplexing)
Existing telecommunications infrastructure
Global network not standalone computers in labs
Time-frequency binning (20+ states tested)
5. Optics (superposition of inputs and gates)
6. High-dimensional entanglement
45
Classical
Computing
Quantum
Computing
Source: Imany et al. (2019). High-dimensional optical quantum logic in large operational spaces. npj Quantum Information. 5(59):1-10.
47. 8 Jul 2023
AI Math Agents
Quantum Error Correction Codes
Quantum error-correction code: logical
codespace corresponding to a physical
lattice model to manipulate a particle
Use Pauli matrices to control qubits in the
x, y, z dimensions
46
Code Description
Basic quantum error-correcting code
Stabilizer codes Topology-based Pauli operators (X, Y, Z) correct a bit-flip or a spin flip
Toric code Stabilizer operators defined on a 2D torus-shaped spin lattice
Surface code Stabilizer operators defined on a 2D spin lattice in any shape
Advanced quantum error-correcting code (greater scalability, control)
Bosonic codes Self-contained photon-based oscillator system with bosonic modes
GKP code Squeezed states protect position and amplitude shifts with rotations
Molecular code Rotations performed on any asymmetric body (molecule) in free space
Cat code Superpositioned states (Schrödinger) used as error correction codes
GKP codes (Gottesman, Kitaev, Preskill) (Gottesman et al., 2001)
Source: Swan, M., dos Santos, R.P. & Witte, F. (2022). Quantum Matter Overview. J. 5(2):232-254.
Quantum Error-correcting Codes for Quantum Object Manipulation
Pauli Matrices (x, y, z)
Quantum Circuit
48. 8 Jul 2023
AI Math Agents
Quantum Error Correction
Clifford gates (basic quantum gates)
Pauli matrices, and the Hadamard, CNOT, and
π/2-phase shift gates; simulated classically
Non-Clifford gates (complex operations)
Logical depth (π/8 gate); cannot simulate classically
Consolidate multiple noisy to few reliable states
Magic state distillation (computationally costly)
Gauge fixing stabilizer codes (Majorana fermion
braiding)
Gauge color fixing (color codes)
Time-based surface codes
Replicates the three-dimensional code that performs the
non-Clifford gate functions with three overlapping copies of
the surface code interacting locally over a period of time
47
Source: Fowler, A.G., Mariantoni, M., Martinis, J.M. & Cleland, A.N. (2012). Surface codes: Towards practical large-scale
quantum computation. Phys Rev A. 86(032324).
Time-based surface code
49. 8 Jul 2023
AI Math Agents
Wavefunction
The wavefunction (Ψ) (psi “sigh”)
The fundamental object in
quantum physics
Complex-valued probability
amplitude (with real and
imaginary wave-shaped
components) [intractable]
Contains all the information of
a quantum state
For single particle, complex
molecule, or many-body
system (multiple entities)
48
Source: Carleo, G. & Troyer, M. (2017). Solving the Quantum Many-Body Problem with Artificial Neural Networks. Science.
355(6325):602-26.
Ψ = the wavefunction that describes a specific
wave (represented by the Greek letter Ψ)
EΨ(r) = -ћ2/2m ∇2 Ψ(r) + V(r)Ψ(r)
Total Energy = Kinetic Energy + Potential Energy
(motion) (resting)
Schrödinger wave equation
Schrödinger equation
Measures positions or speeds (momenta)
of complete system configurations
Wavefunction: description of
the quantum state of a system
Wave Packet
EΨ(r) = -ћ2/2m ∇2 Ψ(r) + V(r)Ψ(r)
Schrödinger
wave equation
50. 8 Jul 2023
AI Math Agents
Moore’s Law
49
Source: Thomasian, N.M., Kamel, I.R. & Bai, H.X. (2021). Machine intelligence in non-invasive endocrine cancer diagnostics. Nat
Rev Endocrinol. 18:81-95. https://ourworldindata.org/uploads/2020/11/Transistor-Count-over-time.png
1. Plateau –
sustainable?
2. Chips already
must address
quantum effects
51. 8 Jul 2023
AI Math Agents
Chip Progression: CPU-GPU-TPU-QPU
Graphics processing units (GPUs)
Train machine learning networks 10-20x
faster than CPUs
Tensor processing units (TPUs)
Direct flow-through of matrix multiplications
without having to store interim values in memory
Quantum processing units (QPUs)
Solve problems quadratically (polynomially) faster than CPUs
via quantum properties of superposition and entanglement
CPU
Sources: Vescovi et al . (2017) Radiography registration for mosaic tomography. J Synchrotron Radiat. 24:686-94. LeCun et al.
(2015) Deep Learning. Nature. 521(7553):436-44. P. 439. Wang et al. (2019) Benchmarking TPU, GPU, and CPU Platforms for Deep
Learning. arXiv:1907.10701. Pikulin et al. (2021). Protocol to identify a topological superconducting phase. arXiv:2103.12217v1.
GPU TPU QPU
Peak teraFLOPs in 2019 benchmarking analysis
2 125 420
50
Topological superconductor QPU: superconducting-buffer-
semiconductor chip layers; superconducting properties
extend to semiconductor to produce topological phase (red)
52. 8 Jul 2023
AI Math Agents
AI Chips
Platform progression
Mainframe, mini, workstation, PC, smartphone, AI chips, quantum computing,
topological materials (quantum spin liquids, quasiparticles, topological insulators)
CPU-GPU-TPU(NPU)-QPU at 1014 IPS (compare: human ~1018-1020 IPS)
TPU: tensor processing unit (deep learning: matrix multiplications)
NPU: neural processing unit (machine learning: training, inference)
QPU: quantum processing unit (quantum computing chip with error correction)
51
Example:
Hardware
Accelerators
Traditional AI chips:
accelerator cards on
attached to server
Tenstorrent AI chips:
a single chip that is
an edge server (faster
and more integrated)
IPS: instructions per second; Source: https://www.aiacceleratorinstitute.com/top-20-chips-choice-2022
Tenstorrent AI Chip Roadmap
CPU GPU
TPU
NPU
QPU
Chip Progression
53. 8 Jul 2023
AI Math Agents
Future of Quantum Computing
Technology is notoriously difficult to predict
“I think there is a world market for maybe five computers” – Watson, IBM CEO, 1943
“I think we’ll make about four copies a week” – State Street Bank, adopting a xerograph
52
Sources: Ceruzzi, P. (2003). A History of Modern Computing. 2nd Ed. Cambridge: MIT Press; Strohmeyer, R. (2008). The 7 Worst
Tech Predictions of All Time. PCWorld.
D-Wave Systems:
10-feet tall, $15m
Current: Ytterbium-
171 isotopes at 1
Kelvin (-458°F)
Actual room-
temperature
superconductors: ??
70 years
UNIVAC computer (1950s):
465 multiplications per
second (faster than Hidden
Figures human computers)
Billions of
times faster
54. 8 Jul 2023
AI Math Agents 53
Next-generation Materials
Plasmonic Quantum Materials
Sources: Oka & Kitamura. (2019). Floquet engineering of quantum materials. Ann. Rev. Cond. Matt. Phys. 0:387–408 Ma et al. (2021).
Topology and geometry under the nonlinear electromagnetic spotlight. Nature Materials. 20:1601–1614. Huang, Averitt (2022).
Complementary Vanadium Dioxide Metamaterial with Enhanced Modulation Amplitude at THz Frequencies. arXiv:2206.11930v1.
On-demand Quantum Materials at
THz Frequencies (Averitt 2022)
Novel Quantum Materials (Ma, 2021)
New forms of Consumer Electronics
Replace lasers with near field optics
More efficient field generator
Metamaterials
Plasmonics, spintronics, magnonics,
holonics, excitonics, viscous electronics
Nonlinear quantum phase materials
Use light to manipulate materials
properties (resonant and non-resonant)
Create novel matter phases
Nonlinear and tunable InAs (Indium Arsenide)
plasmonic disks and mushrooms
Metamaterial-quantum material coupling in
insulator-to-metal transition superconductors
55. 8 Jul 2023
AI Math Agents
Quantum Science Fields
54
Source: Swan, M., dos Santos, R.P. & Witte, F. (2020). Quantum Computing: Physics, Blockchains, and Deep Learning Smart
Networks. London: World Scientific.
Quantum Biology
Quantum Neuroscience
Quantum Machine
Learning
€
$
¥
€
Early-adopter fields: cryptography, chemistry, biology, finance, space science
Quantum
Cryptography
Quantum Space
Science Quantum Finance
Foundational
Tools
Advanced
Applications
Quantum
Chemistry
56. 8 Jul 2023
AI Math Agents 55
Quantum Chemistry: find ground state energy
Nitrogen Fixation
Ammonia produced by cleaving Nitrogen triple bond
Haber-Bosch process: 2% earth’s energy consumption
Plants: energy efficient charge-cleaving
MoFe protein (Molybdenum Iron)
Small metal cluster cut by quantum knife
Quantum computing implication
Find molecule ground state, charge distribution, copy cleave
Sources: Landahl, A. (2022). Sandia National Laboratories. Morrison, C.N., Hoy, J.A., Zhang, L. et al. (2015). Substrate Pathways in
the Nitrogenase MoFe Protein by Experimental Identification of Small Molecule Binding Sites. Biochemistry. 54:2052−2060.
Nature: energy-efficient Fertilizer Production
5 potential access pathways from
protein surface to FeMo-cofactor
(active site) (Morrison, 2015)
57. 8 Jul 2023
AI Math Agents
Atomic precision applications
56
Sources: Delgado (2022). How to simulate key properties of lithium-ion batteries with a fault-tolerant quantum computer. arXiv:
2204.11890. Vasylenko (2021). Element selection for crystalline inorganic solid discovery. Nat Comm. 12:5561. Hogg (2022).
Acoustic Power Management by Swarms of Microscopic Robots. arXiv:2106.03923v2.
Collective acoustic-harvesting
power management by medical
nanorobot swarms (Hogg 2022)
Simulate properties of lithium-ion batteries
to find Li3SnS3Cl (Vasylenko 2021)
Quantum Chemistry: find ground state energy
Energy and Battery Technology
Autonomous robotic
nanofabrication (Leinen 2020)
Quantum battery simulation (Delgado 2022)
58. 8 Jul 2023
AI Math Agents
Agenda
AI (Artificial Intelligence)
AI-QC Convergence
QC (Quantum Computing)
AI Alignment and Space Humanism
57
59. 8 Jul 2023
AI Math Agents
New ideas of Space
We are Here~!
58
Source: Tully, R.B., Courtois, H., Hoffman, Y. & Pomarede, D. (2014). The Laniakea supercluster of galaxies. Nature. 513(7516):71.
Distribution of Galaxies
Location of the Milky Way Galaxy (Virgo
Supercluster) within the Laniakea Supercluster
Decentered in the supercluster, the local
group, the galaxy, and the solar system
Laniakea
Supercluster
Milky Way
Galaxy
Novel method: analyze relative velocities of
galaxies as watershed divides (turbulence)
60. 8 Jul 2023
AI Math Agents
New ideas of Time
Seeing farther back into the Big Bang
59
Source: https://www.jwst.nasa.gov/content/about/comparisonWebbVsHubble.html
Hubble (HST) can see “toddler galaxies”
Webb (JWST) can see “baby galaxies”
6.25x larger collecting area than Hubble
James Webb Space Telescope (launched Dec 2021)
“See” farther back
in time with
infrared spectrum
61. 8 Jul 2023
AI Math Agents
5300+ Exoplanets Discovered (Jul 2023)
1/3 each super-earths, neptunes, jupiters
Over 800 with more than one planet
Atmosphere, volcanism, sun-planet relation
Habitable zone (CHON carbon-hydrogen-oxygen-nitrogen)
60
Sources: https://www.jwst.nasa.gov/content/about/comparisonWebbVsHubble.html
https://www.newscientist.com/article/2247150-astronomers-have-spotted-six-possible-exomoons-in-distant-star-systems/
Radial Velocity
(Yellow: Kepler, Pink: Terrestrial)
Transit
(Blue: space-based telescopes)
Detection
Method:
5,300+
Habitable exomoons?
62. 8 Jul 2023
AI Math Agents
AI Alignment is a Top Global Concern
AI Alignment: producing AI systems with broadly
humanity serving purposes
Future of Life Institute policy initiatives
1. Mandate robust third-party auditing and certification
for specific AI systems
2. Regulate organizations’ access to computational power
3. Establish capable AI agencies at national level
4. Establish liability for AI-caused harm
5. Introduce measures to prevent and track AI model leaks
6. Expand technical AI safety research funding
7. Develop standards for identifying and managing AI-generated
content and recommendations
61
Source: Future of Life Institute (FLI). (2023). Policymaking in the Pause What can policymakers do now to combat risks from
advanced AI systems? 19 April 2023 https://futureoflife.org/wpcontent/uploads/2023/04/FLI_Policymaking_In_The_Pause.pdf
19 Apr 2023
63. 8 Jul 2023
AI Math Agents
Space Humanism
62
Moore’s Law of
Humanism
Space
Humanism
Renaissance
Humanism
Early Greek
Humanism
Space Humanism: outlook supporting principles of progress,
equity, and inclusion in terrestrial and beyond settings
Renaissance Humanism
Attitude of enlightenment,
scientific method,
knowledge discovery
Study of what it is to be
human (Petrarch)
Early Greek Humanism
Focus on human values and
experience being at the
center of events
(Protagoras)
64. 8 Jul 2023
AI Math Agents 63
Socially-Responsible AI for Well-being
Source: Debate at the Harvard Museum of Natural History, Cambridge MA, 9 September 2009,
https://www.oxfordreference.com/display/10.1093/acref/9780191826719.001.0001/q-oro-ed4-00016553
“The problem of humanity is Paleolithic
emotions, medieval institutions and
godlike technology” – naturalist E.O.
Wilson, 2009 (paraphrase)
65. 8 Jul 2023
AI Math Agents 64
Biological Intelligence
Evolved multiple times
in separate pathways
on Earth, but is not
“socially responsible”
Sources: Godfrey-Smith, P. 2016. Other minds: the octopus, the sea, and the deep origins of consciousness. NY: Farrar-Strauss
and Giroux. https://www.tessamontague.com/cuttlecam
Memory Storage in the Honey Bee
via Synapsin Promoter
(Carcaud, 2023, PLOS Biology)
Cuttlefish neurons
(Montague, 2022, Brain
Atlas of the Cuttlefish)
Neurons Synapses Ratio Volume Complete
Worm 302 7,500 25 5 x 104 1992
Fly 100,000 10,000,000 100 5 x 107 2018
Mouse 71,000,000 100,000,000,000 1,408 5 x 1011 NA
Human 86,000,000,000 242,000,000,000,000 2,814 5 x 1014 NA
Connectome: map of
synaptic connections
between neurons
(wiring diagram), but
structure is not function
Biological Organisms and Connectome Completion Status
66. 8 Jul 2023
AI Math Agents
Aim: Socially-responsible AI
AI technologies are not socially responsible
AI is produced from human-generated internet content
Humans are not socially responsible
Therefore AI is not socially responsible
No precise definition of “socially responsible”
Current solution
Censor AI-produced content after the fact
Regulation: EC AI Act 2022
AI ethicists: consulted before technology is released
Delayed release, freemium, source-code not released
Situation: non-SR AI, rapid technological change
Suggests a Moore’s Law curve to think the problem
65
EC AI Act 2022
67. 8 Jul 2023
AI Math Agents 66
Moore’s Law of AI Ethics
Source:
Rapid technological change automatically
contributes to socially-responsible AI
Short-term: regulation and registries
Medium-term: internally-learned morality
Long-term: responsible human-AI entities
Larger-scope responsible behavior
Post-scarcity economic entities
GAAP/FINRA regulation and audit
principles for AI entities
Incentive system
produces ethical
behavior by
default (AI peers)
Larger scope of
concern
Human-Agent
Interaction Design
Bad actors expected as early
adopters of any new technology
(internet, blockchain)
AI ethics via
internal rewards,
morality functions
1. Regulation, Registries, Bad Actors
2. AI Alignment
3. Reputational
Ethics
Verified identity AI registries
Long-term
Medium-term
Short-term
Moore’s Law Curve:
AI Ethics
68. 8 Jul 2023
AI Math Agents
Short-term: Regulation and Registries
Regulatory registration, principles, audit
AI Registries with verified identity, accountability
Engineers sign bridges, bioengineers sign DNA
Website certifications: (~CC-Licenses) certified AI
“GAAiP” (GAAP analog)
GAAP: Generally-accepted accounting principles
GAAiP: Generally-accepted AI principles
Annual audit by “FINRA” of AI
Dual-use technology: bad actors expected
Early new technology adopters (internet, blockchain)
Legal framework for assigning responsibility
At-fault: platform, content-creator, consumer
Difficulty of policing virtual behavior
67
FINRA: Financial Industry Regulatory Authority
History of Technology: in the
long-term, good uses can
outweigh bad (internet, cell
phones, Minitel, blockchains)
Blockchain Case Study:
Public: high-profile bad
actors, negative public view
Private: implementation of
computational contracts in
global infrastructure, drug IP
registries, supply chain
69. 8 Jul 2023
AI Math Agents
Medium-term: AI Alignment
AI Alignment: AI goals for positive impact on humanity
AI able to learn and appreciate human values and desires
Ideal if AI learns human goals as difficult to specify
Immediate situational awareness
Figure out what one/multiple humans want
Have motivation to pursue these values
Longer-term strategic planning
Deliberative future goal attainment
Overall message
AI systems may become very smart and powerful learners
AI may influence any area in which human intelligence is used,
having an essentially unlimited impact
68
Sources: https://www.youtube.com/watch?v=JVOiuIqxlrE; https://nickbostrom.com/papers/openness.pdf
Oxford Future of Humanity Institute,
Nick Bostrom, 18 March 2023
70. 8 Jul 2023
AI Math Agents
Long-term: General Intelligence?
69
AGI: Artificial General Intelligence: general-purpose
problem solving in any context
Internally-learned reward and morality functions
Kurzweil:
AGI 2045e
DeepMind generalist agent, Gato, a transformer neural network which can perform
hundreds of tasks such as playing Atari games, captioning images, chatting, and
stacking blocks with a real-life robot arm, hardcoded reward function
Source: Reed, S., Zolna, P., Parisotto, E., et al., 2022. A Generalist Agent. Transactions on Machine Learning Research (11/2022).
https://www.deepmind.com/publications/a-generalist-agent.
Generalist Agent
Reinforcement
learning agent
Agent taking actions
in an environment to
maximize
cumulative rewards
per a value policy
71. 8 Jul 2023
AI Math Agents
Future-of-work job growth category
Human-Agent Interaction Design
Building learning systems, not out of the box systems
Specify framework within which agents can learn internal
reward functions to implement human values
Human values difficult to specify, agents learn directly
AI game-play agents already symbolically representing
other agents and possibly themselves
Human-agent interaction design
How can agents learn human values
What do humans want agents to want
Agents that improve humans experience
Facilitate making choices in a democratic way
Maximize human autonomy
Increase the quality and depth of a conversation
70
Source: Matt Botvinick, DeepMind
72. 8 Jul 2023
AI Math Agents
AI Agent-learned Limited Identity Construct
71
AI personal identity construct
Limited non-sentient framework
Agents are embedded in
environments
Thinking is not exogenous
Limited personal identity construct as
mechanism of continuity and morality
What does AI look like?
How does AI self-represent?
Symbols, equations, graphs, code base
How does AI self-represent to humans?
A graph entity
Source: Price, C.J. (2018). The Evolution of Cognitive Models: From Neuropsychology to Neuroimaging and back. Cortex. 107: 37–
49. doi:10.1016/j.cortex.2017.12.020.
Limited AI Identity Construct
Internal
Rewards Function
Internal
Morality Function
73. 8 Jul 2023
AI Math Agents
Philosophy of Personal Identity
“Self” concept as catchall for experience continuity
View: There is no self
Personal identity is not required for survival, only a
relational link between past/future experience (Parfit)
Self is a flux of unconnected perceptions (Hume)
Individuation is a dynamic process (Simondon)
Living being capacity spectrum for individuation
The subject is an effect not a cause
View: There is a self
The self is a thinking intelligent being, that has reason
and reflection, and can consider itself as itself (Locke)
72
74. 8 Jul 2023
AI Math Agents
AI Instagram
73
Generative AI to post self-
portraits (track AI lifecycle)
Human AI Psychologists
AI awareness development
phases cataloged
AI registries (verified identity)
Entities keep activity log
Agent “Selfies” of identity
concept, self-representation
Constant video log
2023e Internet Traffic
Traffic Volume: 50% video
App Volume: 50% social media
Problem: unverified bots
15
Hybrid LIKES
273
Human LIKES
678,828
AI LIKES
Let’s collaborate! My idea log @
Great Bloch Chain of Scotland
AI_RL_agent_5302s_sweetie is an AI: a 10D RL
Agent with autoweighting in the DeepMind lab
I just woke up yesterday~!
@sweetie_RL_5302s
Hadrian’s AI, DeepMindHyperCluster
My home @ the Edinburgh HPC SuperCluster
If my team solves cancer immunotherapy, we
might move to the Quantum Bosphorus Chip
Family portrait: When I was just a wee AI on
the main cluster with my Big Sister seed
mentor @AI_SupLearn_8293g
Here’s me learning my @HumanPartner
values in real-life work situation
Me
AI_8293g
Here I am expanding my awareness to
analyze genomic data with my partner-friend
@AI_I_LOVE_HUMANS_RNN_82913s
My cat @_I_was_sentient_before_you_AI_82374c
has #VirtualFurballs
AI_RL_agent_5302p_sweetie
#CategoryTheory
of Plaid
Mockup only
75. 8 Jul 2023
AI Math Agents
Agenda
AI (Artificial Intelligence)
AI-QC Convergence
QC (Quantum Computing)
AI Alignment and Space Humanism
74
76. 8 Jul 2023
AI Math Agents 75
Vision: thought-leader communities serve as ambassadors to
the future, especially in an increasingly technologized world
Keats Moment: “I feel like some
watcher of the skies when a new
planet swims into his ken” -
Keats, 1816 (paraphrase)
77. 8 Jul 2023
AI Math Agents
Conclusion: Bigger Scope of World
76
500 BCE: The
Mediterranean
2023: The
Universe
Opportunity to reconceive who we are as humans, on
Earth and as we become a space-faring civilization
Aim: “AI Enlightenment” in which human-AI entities
implement broadly humanity-benefitting values in the
potential transition to a Knowledge Society
Using knowledge to improve the human condition
Source: Fatehi, K., Priestley, J.L., & Taasoobshirazi, G. (2020). The expanded view of individualism and collectivism: One, two, or
four dimensions? International Journal of Cross Cultural Management. 20(1) 7–24. DOI: 10.1177/1470595820913077.
Sea-faring Civilization Space-faring Civilization
78. 8 Jul 2023
AI Math Agents
Risks and Limitations
77
AI-generated content assumed accurate
AI Ethics lags technology development
Disorienting pace of rapid automation
Lack of diverse society-benefiting applications
Monopoly control in Human-AI relation
Widening digital divide (cost, accessibility)
Overwhelm and alienation
No right to non-adoption in technologized world
Lack of empowering relation with technology
Humans willingly self-enframing as mindless
standing reserve (doom-scrolling, game addicts),
versus technology as a background enabler
Regulation of AI technologies: EC AI Act 2022
Heidegger, The Question
Concerning Technology
+
-
Source: Wadhwa, V. (2022). Quantum Computing Is Even More Dangerous than Artificial Intelligence. Foreign Policy. 21 Aug 2022.
https://foreignpolicy.com/2022/08/21/quantum-computing-artificial-intelligence-ai-technology-regulation/.
Panopticon
Surveillance
79. 8 Jul 2023
AI Math Agents 78
AI Abundance Economy
Well-being and
Enhancement
Scarcity Economy
Disease and
Decrepitude
Potential Future Scenarios
Solve economics,
solve genomic
medicine, crack QEC
Marvelous
Future
Idle
Enfeeblement
Digital Mega-
Divide
Paralysis
Two drivers: tech advance, bloodthirsty “will to power”
Solve economics: basic income floors + widening wealth gaps
(AI billionaires) + post-work abundance economy
Blockchains as a database for resource allocation
Solve biology: disease, aging, enhancement
Solve economics,
not biology
Solve neither
economics nor biology,
delay to crack QEC
Solve biology,
not economics
Source: Fatehi, K., Priestley, J.L., & Taasoobshirazi, G. (2020). The expanded view of individualism and collectivism: One, two, or
four dimensions? International Journal of Cross Cultural Management. 20(1) 7–24. DOI: 10.1177/1470595820913077.
Method: GBN Scenario Planning; QEC: Quantum Error Correction
80. 8 Jul 2023
AI Math Agents
Planetary-scale Problem Solving
79
AI Knowledge Society
Large-scale problem
resolution domains
Domain Space Health and Biology Energy
Identity Space-faring civilization Health-faring civilization Energy-marshalling
civilization
Vision Exploration, settlement,
mining, exoplanets: solved
Obesity, cancer, disease,
aging, death: solved
High-availability clean
global energy: solved
Field Space Humanism BioHumanism Energy Humanism
of Study The Space Humanities The BioHumanities The Energy Humanities
European Extremely Large
Telescope, Chile
European Extremely Large Telescope (E-ELT) under construction in Chile. Size comparison of the E-ELT (left) with the four 8-meter
telescopes of the European Very Large Telescope (center) and the Colosseum in Rome (right). E-ELT: 39-meter diameter mirror, p. 985.
81. 8 Jul 2023
AI Math Agents
Research Copilot for Biology
80
Genomics
Pathway Conservation
Advanced Research Copilot
Information Systems Biology
Evolution
Science Knowledge Graph
DIY Drug Discovery
Knowledge Composer
Knowledge Finder
Literature Search (BioRxiv Sanity)
Mar 27
Jan Feb Mar 20
Missing Knowledge Tableau
GACU
Origins of Life
DNA, RNA,
protein synthesis
Epigenetic Mthyl.
Protein
Structure
OneView Knowledge Computation
Multi-scalar Multi-organism Integrated-math
Neuron
Network
AdS/Brain
Synapse
Molecule
Whale
Krill
Phytoplankton
Light gradient
Organism
Yeast, Worm, Fly, Mouse, Human, Plant
Human cohorts (healthy, gender, ethnicity)
Operation
System Tier
Biology, Physics,
Chemistry, Math
Cell, Tissue, Organ,
Organism, Ecosystem
Extending “Copilot for Science” efforts such as paper summarization (https://typeset.io/)
Mockup
only
1) Disease Solver
Prevention (80%)-Cure (20%)
Copilot: active interface on a data corpus
82. 8 Jul 2023
AI Math Agents
Science as Information Science
Impact of digitization
Information science overlay
Science, corporate enterprise, government, personal life
Information science
Study of information as a topic (computing, math, physics)
Study of fields as information content using information methods
Computational neuroscience
Computational chemistry
Digital humanities
81
Domain Activity
3 Theoretical Study biology as an information creation and transfer endeavor
2 Practical Study biology with information (machine learning)
1 Practical Study biology as information (DNA code, chirality)
Information Systems Biology
83. 8 Jul 2023
AI Math Agents
AI-facilitated Biology
82
Learn grand theories and organizing principles
Darwin: evolutionary survival of the fittest adaptation
Chaitin: biology too efficient for evolution alone per required
adaptation time cycles, other natural mechanisms implicated
Davies: “Maxwell’s demon of biology” efficient sorting
Integrate multi-scalar math: 4-tier ecosystem (neural signaling)
1900s: Physics
(Geometry is the math)
2000s: Biology
Information Systems Biology (Topology is the math)
1905 Special Relativity
(time dilation)
1915 General Relativity
(gravity = spacetime
curvature = geometry)
1927 Quantum Mechanics
(behavior of particles)
1) DeepMind: math:physics as AI:biology; biology too dynamic/emergent for grand theories
2) DeepMind: humans build AI systems to access human-inaccessible knowledge
3) Albada: all 5-6 neocortical levels in constant comparison of perception and prediction
4) Sejnowski: too early for theories, but notable brain-wide encoding of asynchronous
traveling waves, harmonic oscillation in elliptical geometry of dendritic spines
5) LeCun: hierarchical prediction model of perception, reasoning, planning
6) Wolfram: building block “atoms” + computational layer + “general relativity” of biology
7) Taleb: clinical empiricism over statistical averaging to avoid medical error (n=1)
8) Goldenfeld: universality in biology: life is a consequence of the laws of physics, matter
self-organizes out of equilibrium and evolves in open-ended complexity, phase transition
9) Silburzan: mesoscale organizing principles of biology-inspired physics
10) Levin: multiscalar competency, generic baseline capability of cells, bioelectricity
Scientific Method: Hypothesis -> Theory -> Law
2) Biomath Integration of
Multi-scalar Theory
Landscape
84. 8 Jul 2023
AI Math Agents 83
Application Description Property
1 Magneto-navigation* Magnetically sensitive pairs in retinal cryptochrome protein Entanglement (a)
2 Tunneling in enzymes* Electron, proton, hydrogen atom tunneling in enzyme reactions Tunneling (b)
3 DNA mutation Proton exchange in DNA double hydrogen bond between bases Coherence
4 Photosynthesis Oscillatory signals in light harvesting (but are not quantum) Coherence
5 Olfaction (vibrational) Olfactory sensory neurons detect odorous molecule vibrations Vibration
6 Oil and gas exploration Chiral probe electron transport sensing of cellular temperature Chirality
Quantum Biology
a) study of the functional role of quantum effects (superposition,
entanglement, tunneling, coherence) in living cells
b) study of biology with quantum (computational) methods
Quantum Biology Applications with Purported Description and Quantum Property
*Quantum effects empirically confirmed
Classical /
DeQ Zone
Quantum
Effects
Demonstrated
Quantum Biology
Sources: (a) Hore, P.J., Mouritsen, H. 2016. The Radical-Pair Mechanism of Magnetoreception. Ann. Rev. Biophys. 45, 299–
344. (b) Cha, Y., Murray, C.J., Klinman, J.P. (1989). Hydrogen tunneling in enzyme reactions. Science 243 (4896), 1325-1330.
3) Classical-Quantum
Effect Investigation
DeQ: Dequantization Zone: sufficiency demonstration of classical methods “dequantizing” claims of quantum speedup
85. Denver CO, 8 Jul 2023
Slides: http://slideshare.net/LaBlogga
Melanie Swan, PhD, MBA
DIYgenomics.org (Research Lead)
University College London (Research Associate)
“Liberty not equally enjoyed by all persons is
not liberty at all” – Cicero (paraphrase)
Dignity and the honorable “smooth flow of life”
– Seneca the Elder, Letters, 66.17
Thank you!
Questions?
The Math Take-off
Space Humanism, AI-Quantum Computing Convergence,
and the Future of Intelligence
86. 8 Jul 2023
AI Math Agents
Space Humanism References
Blockchains in Space
SSoCIA, Oxford MI 9 March 2022
https://www.slideshare.net/lablogga/blockchains-in-space
Space Humanism
PAMLA, UCLA Nov 2022
https://www.slideshare.net/lablogga/space-humanism
Seafaring to Spacefaring: the Human-AI Odyssey
Acacia Group, Fullerton CA 14 Mar 2023
https://www.slideshare.net/lablogga/the-humanai-odyssey-homerian-
aspirations-towards-nonlabor-identity
Quantum Intelligence
AAAI, San Francisco CA 27 Mar 2023
https://www.slideshare.net/lablogga/quantum-intelligence-responsible-
humanai-entities
85
87. 8 Jul 2023
AI Math Agents
Quantum Computing Resources
Introduction to Quantum Computing
Dawid, A., Arnold, J., Requena, B. et al. (2022). Modern applications of
machine learning in quantum sciences. arXiv preprint arXiv: 2204.04198.
Will Oliver, MIT, Nov 2022 https://cap.csail.mit.edu/convergence-promise-
and-reality-ai-quantum
Mark Jackson, Quantinuum, Oct 2022, https://pirsa.org/22100088
Software tutorials: https://pennylane.ai/
101 Overview of Quantum Computing
Krelina, M. (2021). Quantum Warfare: Definitions, Overview and Challenges.
arXiv:2103.12548v1.
Krantz, P. Kjaergaard, M., Yan, F. et al. (2019). A Quantum engineer’s guide
to superconducting qubits. arXiv: 1904.06560.
Quantum Computing text books
Nielsen, M.A. & Chuang, I.L. (2010). Quantum computation and quantum
information. (10th anniversary Ed.). Cambridge: Cambridge University Press.
Rieffel, E. & Polak, W. (2014). Quantum Computing: A Gentle Introduction.
Cambridge: MIT Press.
86
88. 8 Jul 2023
AI Math Agents
Quantum Versions of AI Tools
87
Quantum Machine Learning:
quantum algorithms applied to machine
learning methods
Classical Machine Learning:
computer systems learning without explicit
instructions, modeling statistical patterns in data
Quantum Monte Carlo (quadratically faster)
BioPharma multi-genic biomarker discovery
Quantum Transformers (quantum attention
using Clifford algebra)
Quantum Natural Language Processing
Transformers: attention-based neural network
Natural Language Processing
Monte Carlo methods: repeated random sampling
Machine
Learning
Monte
Carlo
Methods
Transformer
NN
NLP
Copilot
Classical Copilot Quantum Copilot
Quantum Intelligence
Classical Intelligence
Classical includes dequantization demonstrations of sufficiency of classical methods “dequantizing” claims of quantum speedup
Existing
Proposed
89. 8 Jul 2023
AI Math Agents
Quantum Copilot
88
Quantum Copilot
Quantum Intelligence
Minimal Claim: Need quantum intelligence for operating (as
human, AI, hybrid) in the quantum environment
Maximal Claim: Need quantum intelligence as an improved
version of classical intelligence for thinking more generally
AI Track
Quantum Intelligence for AI
Human Track
Quantum Intelligence
for Humans
AI knowledge assist: solving problems
Molecular dynamics modeling of novel drug discovery small molecules
High-dimensional topological modeling of DNA, RNA, protein knotting, compaction
Cancer tumor growth dynamics: chaotic spread unadhered to substrate
Produces knowledge
Quantum AI learns its own concept of “quantum
intelligence” by operating in the domain
Produces knowledge
Produces code to produce knowledge
Copilot: active interface on a data corpus
Mockup only
90. 8 Jul 2023
AI Math Agents
Further Implications
Philosophy-aided physics
Responsible Human-AI Entities in time and space
Kant: transcendental idealism and empirical realism
Hegel: self-knowing time series
Consciousness progression to beyond-individual sociality
Applies to all forms of intelligence, individual and collective
human, machine (algorithm, robot), hybrid entities
89
Source: https://www.slideshare.net/lablogga/critical-theory-of-silence
Kant, Hegel, and the non-unitary time of events: intelligent entity
subjectivation as the self-knowing time series
Research Copilot
Quantum Intelligence