This talk was given to the Tech User Group, Central Oregon on November 15th, 2023.
~~
AI revolutionizes the way we interact with information. It will change the way we work. In this talk, we will introduce the fundamental concepts and the latest research and policies in the field. We will then explore numerous opportunities and societal challenges related to technology adoption, work augmentation and identity. Finally, we will introduce Personal AI and the mission of Kwaai Lab.
Kwaai Lab, a non-profit, volunteer-based open-source AI lab, is composed of researchers, architects, developers, and philosophers. Our goal is to design and implement the tools, fundamentals, and policies that empower us all to own our own Personal AI assistants.
1. AI and the
Future of Work
Presented by Matthew Small, Kwaai
www.mattasmall.com | www.kwaai.ai
Tech User Group, Central Oregon
November 15, 2023
BendTECH Coworking
2. About Me - Matt Small
Advisor - Kwaai Lab, Open Source Personal AI Lab
www.kwaai.ai
Consultant - GTM, PLG, Co-Sell, CX | Cloud, AI, Open Source
www.mattasmall.com | in/mattasmall
Organizer - CloudBeers Bend, Tech Social Hour
www.linkedin.com/company/cloudbeers-bend
Organizer - Bend Guys for Good, Philanthropy and Community
www.linkedin.com/company/bend-guys-for-good
Host, Fundraising - Lucid Cradle, Psilocybin Services
www.lucidcradle.com
Connect on
LinkedIn
3. About Kwaai - Making AI Personal
Kwaai is a non-proļ¬t, volunteer based, open source development organization
focused on democratizing access to these tools by building a free Personal AI.
Tools
Everything required to build and maintain your own AI assistant.
Fundamentals
Deep research on eļ¬ciency, frameworks and making AI more accessible.
Policy
Advocate for transparent, open, AI development, training and licensing.
www.kwaai.ai
Join us Fridays, 10 AM
Weekly Public Meeting
View Past Meetings
on Youtube: @kwaai-ai
[Wikipedia] āKwaaiā - Adjective
1. bad-tempered, aggressive, ļ¬erce
2. (South Africa) (slang) A term of approval, equivalent to great or fantastic; cool; excellent.
[Urban Dictionary] You must possess some really cool gearā¦Kwaai is the top form of "Coolness"
4. Terminology
Artiļ¬cial Intelligence (AI) Machines performing tasks like reasoning and learning.
Learning AI's process of improving decision-making from patterns in data.
Machine Learning (ML) Algorithms that learn from data to improve decisions.
Deep Learning (DL) Learning complex patterns using multi-layer neural networks.
Model Computational representations for analyzing complex systems.
LLM (Large Language Model) AI that processes and generates human language.
GPT (Generative Pre-trained Transformer) AI that processes and generates large bodies of text at once.
GPTs/Agents AI software that is tuned to perform tasks, aka ābots,ā ātwins.ā
RAG (Retrieval Augmented Generation) Retrieving additional information prior to generating an output.
Computer Vision AI interpreting information from visual data.
Natural Language Processing (NLP) AI that processes and interacts using human language.
Hallucination AI generating incorrect or unrelated information.
5. If youāre not paying for it, youāre the product.
AI is a cloud utility service,
for intelligence.
6. Credit: Edoardo Querci della Rovere
https://www.linkedin.com/in/edoardoquercidellarovere/
7. Data ā Knowledge ā Action
Data Sources
Stored Context and
Knowledge Model
I/O with Systems
and Services
8. Classiļ¬er AI
Fat Input ā Knowledge ā Thin Output
Generative AI
Thin Input ā Math ā Fat Output
VS
9. Some Moments in AI History
1942 Asimovās Three Laws of Robotics
1950 The Turing Test
1956 āArtiļ¬cial Intelligenceā - The Dartmouth Conference
1997 IBMās Deep Blue beats Kasparov at Chess
2011 IBMās Watson wins Jeopardy! | Appleās Siri released.
2016 Googleās DeepMind Alphago defeats Sedol at Go
2017 Attention is all you Need [GPTs] - Ashish Vaswani
2020 OpenAI releases GPT-3 (175B parameters, GPT-2 was 1.5B)
11. Some Moments in AI History
December 2022 Cohere Enterprise private model, Google Med-PaLM
February 2023 Microsoft Bing connects LLMs to the Internet
March 2023 Meta LLaMA OSS, GPT-4, Microsoft Private GPT-4, Google
Bard, Anthropic Claude, BloombergGPT,
Open Letter for Pause in Development
May 2023 WGA Strike Begins, 'Abucin' drug discovery
June 2023 EU AI Act (December)
July 2023 Books3 Lawsuit Filed, White House AI Safety Directive
October 2023 WGA Strike Ends, US Executive Order on AI
November 2023 Open AI GPTs, xAI, G7 AI Principles
12. What is it?
Why is it so good/bad at what it does?
What more will we calculate?
Will this replace my job?
What will jobs become?
Will it replace my identity?
What does the future hold?
13. Eļ¬ective Accelerationism
All tech is always good. (e/acc)
Techno-Optimism
Tech is a force for good.
Transhumanism
Transform humans with technology.
Post-Scarcity Economy
Goods will be plentiful.
Cyberpunk
Systems are getting oppressive.
Techno-Feudalism
Tech power is political might.
Techno-Dystopianism
Tech will dehumanize us.
futures
21. The
Arrival
Mind
Paradox
AI Persuasion and Hive Minds
Louis Rosenberg, PhD
TED 2017: https://www.youtube.com/watch?v=Eu-RyZt_Uas
Venture Beat, October 2023: https://venturebeat.com/ai/snoop-dogg-sentient-ai-and-the-arrival-mind-paradox/
23. Reality Check
Generative AI changed the
future of work.
Regulation is urgently needed,
but unlikely to shift much.
We must pursue responsible use
and learn as we go.
We must be cautious of
technology companies.
24. AI Policy (as of Nov ā23)
IN FORCE
ā Authors Guild v Google - SCD (2015)
Google Book Search is fair use because itās public service.
ā Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith - SCD (May 2023)
The art was not transformed enough for fair use.
ā US Copyright Oļ¬ce Guidance for AI Generated Works (March 2023)
AI generated art is not copyrightable because it lacks human authorship.
ā WGA Strike Concessions
AI is a tool for the writers, not the studios to replace them.
ā GDPR AI Addendum
Mandates data minimization, human rights, and the explainability of AI decisions.
ā Self Regulatory Frameworks
Partnership on AI, AI Now Institute, OpenAIā¦
an image of Prince inspired by Andy Warhol
Created by
MidJourney
25. AI Policy (as of Nov ā23)
ONGOING
ā Regulatory Capture Eļ¬orts from OpenAI, Microsoft, Google, Meta, etc.
ā USPTO: 10,000+ public comments have been expressed and reviewed.
ā FTC Testimony to USCO: Weāre mad. Unfair competition, piracy, and consumer fraud is our problem.
150+ legal processes in progress worldwide, including:
ā Sarah Silverman v. Meta & v. OpenAI: Battle over popular source library Books3āRestatement required.
ā Internet Archive vs 4 Large Publishers: Battle over the COVID Emergency LibraryāPublishers have edge.
Rightsholders, Creatives, Prof Organizations globally have spoken up or issued cease and desist orders including;
ā 100,000+ European Creatives across 13 Guild Organizations
ā 15,000+ Authors
ā 2,200+ US News Publishers
ā UNESCO - Ethical Impact Assessment
26. PROPOSED
ā USA White House An Executive Order on AI
Aims to promote AI research and ethical standards in the US.
ā The EU AI Act
Proposed regulation for high-risk AI applications.
ā G7 AI Principles
International principles for ethical AI use.
ā Global Partnership on AI (GPAI)
An initiative for global cooperation on responsible AI development.
ā OECD Principles on AI
Framework to foster AI that respects human rights and democratic values.
ā IEEE Ethically Aligned Design
Guidelines to ensure AI aligns with ethical principles and human values.
AI Policy (as of Nov ā23)
28. Free, distributable software.
Your own hardware and data.
Take the time, or pay for the help.
Code and development is open, modiļ¬able, and reproducible.
Security through transparency and collective accountability.
Build a commercial ecosystem around an open core.
Ethics - Open Source Development
29. Alignment - AI Mortality - A Modern Law
How can we ensure that the motivations of an
artificial intelligence align with our own?
An AI entrusted with a human life
should die if the human dies.
If the passenger dies, the driver dies.
If the trainee dies, the ļ¬tness trainer dies.
If the patient dies, the doctor dies.
If the accused dies, the lawyer dies.
If the executive dies, the executive assistant dies.
SPD-13
āRunaroundā
Asimov
30. Intelligence is not what one knows,
itās what one does when one doesnāt know.ā
Jean Piaget, Theory of Cognitive Development
Noise ā Signal ā Data ā Knowledge ā Decision
31. Machine Learning (95%)
Data Driven Pattern Recognition
Deals with Correlation
Fast, Reactive
āAnimal Instinctsā
Machine Reasoning (5%)
Knowledge Relationship Driven
Deals with Causation
Slow, Logical
āDeveloped Brainā
AI Research and Engineering
34. Latest: Machine Reasoning
Neural networks canāt
learn and think
concurrently.
Causal relationships
cannot be discovered
solely by observational
data.
35. Latest: Machine Reasoning
Logically
Analogies
Induction
Deduction
Fuzzy Logic
Causation
Probabilities
Constraints
Case Based
The beauty of knowledge
is that itās diverse.
Each node is a reasoning engine itself.
Cloning the behavior of human cells.
Determine causation vs correlation.
Khai Pham, Kwaai Advisor,
Life Sciences
36. Latest: Fluid Dynamics for Causal Networks
āWhat if each node in a neural
network is a volume element in a
ļ¬uid, and the interactions were basic
on the physics of ļ¬uid mechanics?ā
Fluidic model could represent a causal network.
There is a maximum speed of causality, c.
It could explain the limits of determinism.
It could help build better reasoning AI.
Reza Rassool, Kwaai Chair
37. Latest: āEmergent Behaviorsā
Self-organizing, collective behaviors emerge when a large collection of things acts as one.
ā āSpontaneousā behaviors in super large LLMs
ā Swarming / Hive behavior in Robots, Traļ¬c, Cells
ā Survival instinct.
āWhen an information system
becomes aware of its existence, itās
likely going to behave in ways
evolution has repeatedly favored in
organic systems.ā
Jeanette Small
Kwaai Member
38. Fundamental Research Needed
ā Develop causal networks.
ā Enable learning and thinking concurrently.
ā Make models run on anything.
ā Train models on less data.
ā Network models as a swarm/hive mind.
ā Identify physical and biological patterns.
ā Determine and align motivations.
39. Considerationsā¦
Ethical and Societal
Ethical and Moral AI Decision Making, Cultural and Social Impact,
Impact on Democracy and Public Discourse, Threat to Personal Identity
Economic and Employment
Threat to Employment, Over Reliance on AI, Digital Divide and Inequality
Monopolization by Big Tech
Privacy and Data Security
Threat to Our Data and Privacy, Dependence on Data and Data Quality
Governance, Legal Issues
Regulatory and Legal Challenges, Global Governance and Cooperation,
Unclear Motivations and Transparency
Technological Risks
Security Risks, Bias in AI (Included or Excluded), Manipulation with AI
Environmental and Health Concerns
Threat to Climate, Threat to Our Psychological Health
40. Will I lose my job?
Maybe. Not exactly.
You may lose your job to
someone using AI.
āProductivity increasesā
will be measured
Jobs will change.
On a more nuanced task, which involved analysing
quantitative evidence only after a careful reading of
qualitative materials, AI-assisted consultants fared worse.
GPT missed the subtleties.
ā
ā
41. Cyborgs intertwined with the AI,
constantly molding, checking and reļ¬ning
its responses.
Centaurs divided labour, handing oļ¬
more AI-suited subtasks while focusing
on their own areas of expertise.
Harvard Business School
Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Eļ¬ects of AI on Knowledge Worker Productivity and Quality: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4573321
42. Adoption Shifts in Work
Switchboard ā VoIP
Typing Pool ā Word Processing
Mailroom ā Email
Stock Traders ā Electronic Trading
Travel Agents ā Online Booking
Reprographics ā Desktop Publishing
Cutting Room ā Digital Video
Drafting Room ā CAD
47. Personal AI (PAI)
Your āExecutive Assistant,ā
that understands your context,
to interface between you and
external cloud/AI services.
Learns your from your data.
Authenticated and authentic to your Persona.
Prompts/validates other AIs and responses.
Generates outputs owned by you.
Vested in your success.
48. AIs
APIs
PAI POD
ā¢ āPersona, Objects, Dataā
ā¢ Self-Sovereign identity
ā¢ Private, secure AI model and data
ā¢ Personal Service Authentication
PAI Network
ā¢ Identity-Based, Blockchain
ā¢ Service Authentication
ā¢ Direct/Group Communication
PAI OS / Apps
ā¢ Conversational OS and UX
ā¢ Health, Financial, Logistical...
PAI I/O
ā¢ Context Included/Validated
ā¢ Least Privilege Info Shared
ā¢ Best-Fit Service Selected
PAI Training
ā¢ Continuous Model Creation
ā¢ On Device or On Cloud
External AIs and API Services
eg. ChatGPT, Claude, Google Search, MS Docs, Uber
49. Suggestions for Use
Avoid āfreeā AI services.
Make āopaque promptsā to public AIs.
Donāt outsource your executive function.
Use your AI as an assistant, not your proxy.
Parent your AI like a toddler.
Become a Cyborg Centaur.
Stay in the loop.
Be nice.
50.
51. Humans are superior.
We are highly evolved biological computers.
We think with more than the brain.
We sense more than visuals.
We are hive animals.
We are creating our ādigital twinsā in our image.
52.
53. Join the Movement!
Who are you?
ā An AI Thinker
a data scientist, developer, project manager, entrepreneur, philosopher, teacher, leaderā¦
ā High-Energy, Curious, Concerned, Motivated
ā Eager to collaborate internationally with like minded individuals
across technology, science or policy
Whatās the ask?
Volunteer 8 hours/month (Weekly Meetings, Workgroups)
Contribute to development and technical papers.
Assist with organizing and project management.
www.kwaai.ai
Join us Fridays, 10 AM
Weekly Public Meeting
View Past Meetings
on Youtube: @kwaai-ai