Practical AI (Sep 2024)

A four-week course with Stephen Reid

Zoom link: https://us06web.zoom.us/j/9082171779

Week 1

AI (Artificial Intelligence), ML (Machine Learning) and DL (Deep Learning)

See Machine Learning Algorithms In Layman’s Terms, Part 1 and What is Deep Learning? - MachineLearningMastery.com to learn more about the different fields of machine learning/deep learning (Unsupervised Learning, Supervised Learning, Reinforcement Learning)

The state of DL-powered AI

AI already beats the average human at a number of key tasks…

…and is expected to surpass humans on a bunch more over the coming years

Recommended newsletters and podcasts

"With so many new AI newsletters popping up, it can be overwhelming to figure out which you'd actually enjoy reading. To choose the right AI newsletter, consider your level of expertise in the field. If you're a beginner, AI Breakfast. If you know a bit more about tech, you might want a newsletter that talks about daily software and tools you could implement today, if so – Ben's Bites. And lastly, if you want a newsletter that dives deep into technical discussions and emerging research you should go with The Batch. You really won't go wrong with any of them, they're all extremely interesting reads about technology that is actively changing the world."

My favourites:

Co-Intelligence: Living and Working with AI by Ethan Mollick | Goodreads

AI tool directories

Top AI Tool Directories:

The best case

"If we can safely harness the power of AI for human betterment,

then we can paint a utopian future our ancestors could hardly fathom.

A future free of disease and hunger,

where biotechnology has stabilised the climate and biodiversity.

Where abundant clean energy is developed in concert with AI;

Where breakthroughs in rocketry and materials sciences

have propelled humans to distant planets and moons;

And where new tools for artistic and musical expression

open new frontiers of beauty, experience, and understanding."

– THE HUMAN FUTURE: A Case for Optimism

Reducing toil:

Catalysing creativity:

Promoting health:

Improving education:

History of deep learning

– A brief history of AI - Raconteur

1950s and 1960s: Birth and Initial Excitement

Photo of Frank Rosenblatt from Professor’s perceptron paved the way for AI – 60 years too soon

Perceptron Research from the 50's & 60's, clip

The concept of a perceptron, an early type of neural network, was introduced by Frank Rosenblatt in the late 1950s.
This period saw the first wave of excitement about the potential of these models to mimic brain-like computation and solve complex problems.

1970s: First Winter

Marvin Minsky and Seymour Papert published "Perceptrons" in 1969, which highlighted the limitations of perceptrons, especially their inability to solve non-linearly separable problems (like the XOR problem). This led to a decline in interest and funding for neural network research.

“What Rosenblatt wanted was to show the machine objects and have it recognize those objects. And 60 years later, that’s what we’re finally able to do,” Joachims said. “So he was heading on the right track, he just needed to do it a million times over. At the time, he didn’t know how to train networks with multiple layers. But in hindsight, his algorithm is still fundamental to how we’re training deep networks today… He lifted the veil and enabled us to examine all these possibilities, and see that ideas like this were within human grasp.”

1980s: Revival with Backpropagation

The introduction of the backpropagation algorithm in the 1980s by David Rumelhart, Geoffrey Hinton, and Ronald Williams allowed for training of multi-layer perceptrons, overcoming the limitations pointed out by Minsky and Papert.
This led to a resurgence in interest in neural networks. The availability of more powerful computers also helped fuel this interest.

1990s: Second Winter

– What Was Actually Wrong With Backpropagation in 1986?, slide by Geoffrey Hinton

Neural networks, especially deeper ones, suffered from issues like vanishing and exploding gradients, making them difficult to train.
Other machine learning techniques like Support Vector Machines (SVMs) and simpler methods became more popular due to their better performance in various tasks and clearer theoretical foundations.

2010s: Deep Learning Revolution

This is when neural networks, especially deep neural networks, began achieving state-of-the-art results in various tasks such as image recognition (e.g., ImageNet competition), speech recognition, and machine translation. Several factors contributed to this resurgence:

Data: With the advent of the internet, there's been an explosion in the amount of data available. Neural networks, especially deep ones, perform better with more data.
Hardware/'compute': The rise of Graphics Processing Units (GPUs) for parallel computation made it feasible to train very large neural networks.
Software/algorithms: Techniques like dropout, batch normalisation, and advanced activation functions were developed, making it easier to train deeper networks.

Late 2010s to Present: Generative AI

The late 2010s marked the rise of generative models within the deep learning community. Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), diffusion models and other generative techniques began to demonstrate the ability to create realistic images, sounds, and videos.
OpenAI's GPT series and models like BERT showcased the power of generative models in the realm of natural language processing, producing human-like text and achieving state-of-the-art results in numerous NLP tasks.
The potential of generative AI has expanded the horizons of what's possible in fields like art, music, entertainment, and design. AI-driven content creation tools, style transfer techniques, and deepfakes are notable outcomes of this era.
The advancements in generative AI have also brought about new challenges and ethical considerations, particularly in areas like misinformation and the authenticity of digital content.

The Triple Exponential

The notion of the "triple exponential" in AI progress refers to the rapid growth in three key areas: data, hardware, and software. Each of these areas is experiencing exponential growth, and the combination of all three has catalysed the rapid advancement of AI in recent years:

Data: The digital age has led to an explosion in the amount of data available. Every online interaction, transaction, sensor reading, etc., generates data. The availability of big data is crucial for training sophisticated AI models, especially deep learning models that require vast amounts of labelled data. The exponential growth in data availability has been a key driver in the success of modern AI.
Hardware/'compute': This refers to the exponential growth in computational power. Moore's Law famously predicted that the number of transistors on a microchip would double approximately every two years, leading to an exponential increase in processing power. While the original formulation of Moore's Law has seen some challenges, other hardware innovations like specialised AI accelerators (e.g., TPUs or GPUs) and advancements in quantum computing continue to drive rapid growth in computational capabilities.
Software/algorithms: AI algorithms and models are becoming increasingly sophisticated. Deep learning, which was conceptualised decades ago, has seen a resurgence in the 2010s due to the availability of large datasets and powerful hardware. The exponential progress in software includes not only the development of new algorithms but also the refinement of existing ones to achieve better performance.

What is a Large Language Model (LLM)?

What are Large Language Models (LLMs)?

Introduction to large language models

What are Large Language Models? | NVIDIA

Highly recommended:

Large Language Models from scratch

Large Language Models: Part 2

A Very Gentle Introduction to Large Language Models without the Hype | by Mark Riedl | Medium

LLMs

Probably folks in this course are interested in the balance between quality, speed (output tokens/s) and context window.

From LLM Leaderboard | Artificial Analysis, sorting by Quality:

Sorting by context window:

The leaders in the field at this point are:

OpenAI (o1 and GPT)
Anthropic (Claude)
Mistral (Large 2)
Meta (Llama – the only open source model on this list, can be run locally)
Google (Gemini)

Official apps

ChatGPT MacOS app

Different models
Upload files
Edit prompts
Stop generating
Regenerate
Microphone
Voice chat
GPTs
I have Memory and Web Browsing turned off

Claude, Gemini, Mistral, Llama: 'Install page as app'

Claude

Projects
Artifacts

Gemini

Flash is extremely fast
Use AI Studio for long context tasks

Mistral Chat
Meta AI (Argentina, Australia, Cameroon, Canada, Chile, Colombia, Ecuador, Ghana, India, Jamaica, Malawi, Mexico, New Zealand, Nigeria, Pakistan, Peru, Singapore, South Africa, Uganda, United States, Zambia and Zimbabwe)

Alternatives to official apps

Chatbot UI - BYO API key for:

OpenAI
Anthropic
Google Gemini
Mistral
(Groq, Perplexity, OpenRouter)

Working with Gemini's long context window: Google AI Studio
For Llama: the blazing fast SambaNova Cloud

Working with Markdown

Typora is fantastic! Paste & 'Copy without theme styling'.

Running LLMs locally

Prompt engineering

Prompt Engineering 101 - Crash Course & Tips

Master the Perfect ChatGPT Prompt Formula (in just 8 minutes)!

Prompt Engineering Tutorial – Master ChatGPT and LLM Responses

The ULTIMATE Beginner's Guide to Prompt Engineering with GPT-4 | AI Core Skills

I Discovered The Perfect ChatGPT Prompt Formula

Prompt databases

'Hallucinations'/Fabrication/BS

What are AI hallucinations—and how do you prevent them? | Zapier

Recommended:

Context windows

Google's NEW Gemini 1.5 Flash SHOCKS the Industry (2M context window!)

Gemini 1.5 Pro now offers a 2M token context window for devs

Notable LLM-based apps

NotebookLM

https://notebooklm.google/

Personalized AI research assistant powered by Google's most capable model, Gemini 1.5 Pro

Insanely good podcast generator: NotebookLM’s new ‘Audio Overview’ turns your notes into a mini podcast
Excellent citations

Perplexity

https://www.perplexity.ai/

Let's ask Perplexity: How is Perplexity AI different?
Much better at citations than ChatGPT
Nice follow-up questions suggests

Poe

https://poe.com/

Access a variety of LLMs on one platform
Compare answers across LLMs: Poe adds feature to use multiple AI bots in a single chat

Cursor

https://cursor.sh/

Code in natural language!
See also:

Lex

https://lex.page/

AI writing assistant ("Never write alone again")
Good example of 'co-intelligence'

GPT for Work

https://gptforwork.com/

GPT for Sheets is particularly interesting, as an easy way of running many prompts in parallel
Install via https://gptforwork.com/

Getting an OpenAI API key

Go to https://platform.openai.com/
Click Dashboard
Go to API keys on the left

Getting an Anthropic API key

https://console.anthropic.com/

Bonus: Turning Kindle books into text with Calibre/DeDRM

Go to Manage Your Content and Devices

Go to Books and select Download & transfer via USB

Select any device and click Download. The book should download to your computer.

Install Calibre
Install DeDRM

Download the zip file here
Once downloaded, create a new folder and name it whatever you like
Extract the zip file into that folder
Go to Calibre, then Preferences > Advanced > Plugins > Load plugin from file > New folder you created > Select DeDRM_plugin.zip
Plugin should successfully load into Calibre

Add and convert books:

Once downloaded, go to Calibre and select Add Books. Select the books you wish to convert into EPUBs/other formats and they should load onto Calibre
Once downloaded, select the book(s) and press Convert Books
When the new menu pops up, ensure the Output Format on the top right is what you require, and press OK
Voila! It should remove the DRM from your Kindle book

(Last parts copied from this guide)

Week 1 recording

https://www.youtube.com/watch?v=OIkqbDHv3Fk

Week 2

AI image generation

The current leaders: FLUX, Ideogram, Midjourney, Playground

Note how cheap Playground and FLUX.1 [schell] are

FLUX.1 [schell] is also blazing fast!

Midjourney is the most expensive, and by far the slowest, and now not even the highest quality! However, it has by far the most sophisticated UI:

Midjourney's web editor

I highly recommend reading/watching the web editor docs in full. See also the docs for character references, in particular note the cw parameter:

"Use the character weight parameter --cw to set the strength of characterization. --cw accepts values from 0 to 100. --cw 0 focuses on the character's face only. Higher values use the character's face, hair, and clothing. --cw 100 is default."

Other notable image tools

AI voice

AI video

17 Best AI Video Generators in 2024 - With Video Examples

"There are very different use cases for AI video generators. Let's break them down into 3 subcategories because it wouldn't be fair to compare a diffusion model to an AI avatar generator."

AI film

The World's First AI Filmmaking Course — Curious Refuge

"Teaching the future of storytelling"

Nobody & The Computer - YouTube

Video-to-video:

The Best AI Video Tools Compared: KLING vs LUMA vs Runway Gen 3

Kling vs. Runway Gen 3 vs. Luma Dream Machine vs. Pixverse

AI music

A.I. Sampling and how the Music Industry will change forever

Live from Latent Space (Album made with Google MusicLM)

Suno
Udio

Week 2 recording

https://www.youtube.com/watch?v=Lps8_ZWs9Ck

Week 3

fal.ai

Fal.ai, which hosts media-generating AI models, raises $23M from a16z and others | TechCrunch

Notable models:

Flux LoRA:

Fast training: https://fal.ai/models/fal-ai/flux-lora-fast-training/playground
Slower training: https://fal.ai/models/fal-ai/flux-lora-general-training/playground
Use the LoRA: https://fal.ai/models/fal-ai/flux-lora
Training Your Own Flux LoRA Model
Train FLUX lora on anyone or anything

AuraSR upscaler: https://fal.ai/models/fal-ai/aura-sr
Wizper: https://fal.ai/models/fal-ai/wizper
Realtime camera: https://fal.ai/demos/camera

Replicate

Background removal:

https://replicate.com/pollinations/modnet

ControlNet:

ControlNet Revolutionized How We Use AI To Generate Images

NEW ControlNet for SDXL!

How ControlNet v1.1 Is Revolutionizing AI Art Even Further

Runway

It can be useful to get familiar with the language that is being used around AI image and video, so you know what to search for or ask for:

Adobe Express's generative AI glossary

Week 3 recording

https://www.youtube.com/watch?v=P76qPC40MYQ

Week 4: AI Futures

Books

Key figures

Veterans:

Geoffrey Hinton

"Godfather of AI" Geoffrey Hinton: The 60 Minutes Interview

Prof. Geoffrey Hinton - "Will digital intelligence replace biological intelligence?" Romanes Lecture

Nick Bostrom

What happens when our computers get smarter than we are? | Nick Bostrom

Prof. Max Tegmark

The Impact of chatGPT talks (2023) - Prof. Max Tegmark (MIT)

'Doomers':

Eliezer Yudkowsky

Will Superintelligent AI End the World? | Eliezer Yudkowsky | TED

Connor Leahy

AGI in sight | Connor Leahy, CEO of Conjecture | AI & DeepTech Summit | CogX Festival 2023

Skeptics:

Gary Marcus (Marcus on AI)

The AI Bubble: Will It Burst, and What Comes After?

Brian Merchant (Blood in the Machine)

Why Big Tech is Ruining Our Lives with Brian Merchant - Factually! - 250

Bosses:

Demis Hassabis (Google)

Unreasonably Effective AI with Demis Hassabis

Dario Amodei (Anthropic)
Sam Altman (OpenAI)
Yann LeCun (Meta)
Jensen Huang (Nvidia)

Organisations

Life 3.0 (Max Tegmark)

Max Tegmark lecture on Life 3.0 – Being Human in the age of Artificial Intelligence

Life 3.0: Being Human in the Age of AI | Max Tegmark | Talks at Google

How to get empowered, not overpowered, by AI | Max Tegmark

Max Tegmark on Life 3.0: Being Human in the Age of Artificial Intelligence

On the Lex Fridman podcast:

Max Tegmark: Life 3.0 | Lex Fridman Podcast #1

Max Tegmark: AI and Physics | Lex Fridman Podcast #155

Max Tegmark: The Case for Halting AI Development | Lex Fridman Podcast #371

Glossary

Life 1.0 (biology): Biological life that evolves based on genetic information and cannot significantly redesign itself during its lifetime. Examples include bacteria.
Life 2.0 (culture): Life forms like humans that can learn and adapt, essentially having the ability to redesign their software (or neural patterns).
Life 3.0 (AI/biotech): A life form that can design not only its software (like Life 2.0) but also its hardware. In other words, these life forms would have the ability to drastically change or improve themselves both mentally and physically.
Intelligence: The ability to acquire and apply knowledge and skills, solve problems, and adapt to new situations. It encompasses reasoning, understanding, learning, and cognition.
Artificial Intelligence (AI): The simulation of intelligence in machines. It involves creating algorithms that allow computers to perform tasks that would typically require human intelligence. This includes tasks like visual perception, speech recognition, decision-making, and translation between languages.
Narrow Intelligence/Weak AI: An intelligent system designed and trained for a particular task. Examples include virtual personal assistants, such as Apple's Siri or image recognition software. They operate under a predefined set of rules and do not possess general problem-solving capabilities beyond their specific domain.
General Intelligence: Often associated with humans, this type of intelligence reflects the ability to understand, learn, and apply knowledge in multiple domains, reason through problems, and be adaptable to varied situations.
AGI (Artificial General Intelligence)/Strong AI: AI systems that possess (general) intelligence comparable to human beings, allowing them to perform any intellectual task that a human being can do.
ASI (Artificial Superintelligence, Superintelligence): An intellect that is much smarter than the best human brains in practically every field, including creativity, reasoning, and social skills.

Superintelligence: Science or Fiction? | Elon Musk & Other Great Minds (Elon Musk, Stuart Russell, Ray Kurzweil, Demis Hassabis, Sam Harris, Nick Bostrom, David Chalmers, Bart Selman, Jaan Tallinn)

Friendly AI: An AI that is designed to be beneficial to humanity and safe.
Utility function: A mathematical representation of what an agent (like an AI) wants or values. It is used to describe the agent's goal or objective.
Consciousness: The subjective experience of being. Tegmark discusses various perspectives on consciousness and its potential relevance to AI.
Technological unemployment: The phenomenon where automation and AI displace jobs faster than new jobs are created.
Existential risk: A risk that threatens the entire future of humanity.
Recursive self-improvement: The process by which an AI system could potentially improve its own architecture or algorithms, leading to rapid increases in intelligence.
Intelligence explosion: The idea that once we create an AI smarter than us, it might assist in building an even smarter AI, leading to a rapid chain reaction of increasing intelligence levels.

The intelligence explosion: Nick Bostrom on the future of AI

Singularity: A hypothetical point in the future when technological growth becomes uncontrollable and irreversible, leading to unforeseeable changes to human civilization. Often, the singularity is associated with the moment when an AI undergoes recursive self-improvement, rapidly advancing its own intelligence beyond human comprehension.
The Alignment Problem: The challenge of ensuring that the goals and behaviours of AI systems align with human values and intentions. As AI systems become more autonomous and potentially more powerful, there is a concern that they might act in ways that are not beneficial or even harmful to humanity. The alignment problem emphasises the need to develop AI in such a way that it understands and respects our values and does not inadvertently cause harm or operate in unintended ways.

Summary

Definition of Life 3.0: Tegmark categorises life into three stages based on how it processes information. Life 1.0 (like bacteria) is life that evolves but doesn't learn during its lifetime. Life 2.0 (like humans) can learn and adapt but is limited by biological evolution. Life 3.0 can redesign not only its software (learning) but also its hardware (body), breaking free from evolutionary shackles.
The Promise and Peril of AI: Tegmark discusses the immense potential benefits of AI, such as curing diseases, solving complex problems, and ushering in an era of abundance. However, he also warns of the risks, including the possibility of AI surpassing human intelligence (superintelligent AI) and becoming uncontrollable.
Cosmic Perspective: He speculates on the broader cosmic implications of AI, suggesting that the transition from Life 2.0 to Life 3.0 may be a recurrent cosmic event, and that our universe might be filled with Life 3.0 civilizations.
AI Development Scenarios: Various scenarios are explored regarding how superintelligent AI might come about, including a slow development where society has time to prepare and a rapid development that could catch humanity off guard.
Values and Goals: One central concern is ensuring that superintelligent AI aligns with human values. Small misalignments could lead to unintended negative consequences. Tegmark discusses the challenge of defining these values and ensuring they are instilled in AI.
The Future of Jobs and Economy: The impact of AI on the job market is discussed, considering both the potential for job displacement and the emergence of new professions. The book also touches upon the implications for wealth distribution and possible solutions like universal basic income.
Consciousness and Identity: Tegmark delves into the philosophical questions of consciousness and identity in the age of AI, contemplating if machines could ever be conscious and what that means for our understanding of self.
The Role of Physics: Being a physicist, Tegmark relates the development of AI to the laws of physics, suggesting that understanding the cosmos can inform our understanding of intelligence and cognition.
Collective Decision Making: Tegmark argues that the future of AI should be decided collectively, emphasising the importance of global cooperation to navigate the challenges and capitalise on the opportunities of the AI revolution.
Call to Action: The book concludes with a call to action for researchers, policymakers, and the general public to engage in the conversation about the future of AI, ensuring that its development benefits all of humanity.

Aftermath scenarios

Superintelligence in SF. Part I: Pathways to Superintelligence | Thore Husfeldt
Superintelligence in SF. Part II: Failures | Thore Husfeldt: "Given the highly disruptive and potentially catastrophic outcome of rampant AI, how and why was the Superintelligence released, provided it had been confined in the first place? It can either escape against the will of its human designers, or by deliberate human action."

Summary of AI risks (beyond alignment)

High level: The A.I. Dilemma - March 9, 2023

Lethal Autonomous Weapons (LAWs): LAWs can independently identify and attack targets without human direction. Their introduction into warfare and policing could lead to rapid escalation of conflicts and unintended civilian casualties. The ethical concerns surrounding the removal of human judgement from life-or-death situations make these weapons highly controversial.

Slaughterbots - if human: kill()

Slaughterbots

The ideas behind 'Slaughterbots - if human: kill()' | A deep dive interview

Surveillance Systems: AI-driven surveillance technology, such as facial recognition, enhances the ability of entities, especially governments, to monitor populations. This technology can be misused by authoritarian regimes to suppress opposition, monitor dissent, and violate privacy rights, leading to a dystopian loss of privacy and freedom.

Artificial intelligence study decodes brain activity into diaglogueAI Mind Reading Experiment!

Loss of Jobs: As AI and automation technologies advance, they threaten to displace human jobs in several sectors, from manufacturing to services. While new job roles might emerge, the transition could lead to economic hardships, societal unrest, and a need for new skills training and social safety nets.

Artificial Intelligence responsible for 5% of jobs lost in May

How AI Is Already Reshaping White-Collar Work | WSJ

Is AI coming for your job? | DW Business

Wealth Concentration: AI could empower a handful of mega-corporations with significant competitive advantages, leading to monopolistic behaviours, stifling innovation, and concentrating wealth and power, further widening societal inequalities.

Yanis Varoufakis: Welcome to the age of technofeudalism
How AI will make inequality worse

Algorithmic Bias: AI systems often learn from historical data. If this data contains societal biases, the AI will replicate or even amplify these biases, leading to unfair or discriminatory decisions. Such biases can manifest in various areas like hiring, lending, or law enforcement, and can perpetuate racial, gender, or socio-economic disparities.

How AI Image Generators Make Bias Worse
ChatGPT Has A Serious Problem

Malfunction or Unexpected Behaviours: Advanced AI models can sometimes act unpredictably, especially when confronted with situations they weren't trained on. Such erratic behaviours in critical areas like healthcare, transportation, or finance could have dire consequences.

The Most Unsettling Records From the AI Incident Database
Epic AI fails

Progress on interpretability:

Security Vulnerabilities: AI systems can become targets for cyberattacks, or even worse, be used by malicious actors to find and exploit vulnerabilities in other systems. With AI-driven cyber warfare, the scale, speed, and potential damage of attacks could be unprecedented.

Hacking with ChatGPT: Five A.I. Based Attacks for Offensive Security

Dependency: A societal over-reliance on AI, especially in sectors like energy, transportation, or healthcare, could be perilous. Any malfunction, deliberate shutdown, or cyberattack could lead to systemic collapses.

The Real Danger Of ChatGPT

Artificial Escalation

Misinformation & Propaganda: AI can be used to craft persuasive narratives or generate misleading content at scale, undermining truth and facilitating propaganda campaigns.

'Artificial Intelligence could pollute the world with misinformation'
What will the future of AI-powered disinformation look like? | The Stream
Fake image of explosion near Pentagon stirs concerns over artificial intelligence
AI's Disinformation Problem | AI IRL

Deepfakes: Deepfakes are the result of advanced AI models that manipulate media, often replacing one person's likeness with another. These AI-generated creations can be almost indistinguishable from genuine recordings. Their primary danger lies in spreading misinformation, creating fake evidence, or even blackmailing individuals. In the realm of politics, journalism, or legal proceedings, the presence of deepfakes can erode public trust and threaten democratic processes.

The Incredible Creativity of Deepfakes — and the Worrying Future of AI | Tom Graham | TED

Deep Fakes are About to Change Everything

Deepfake audio of Sir Keir Starmer released on first day of Labour conference

MrBeast and BBC stars used in deepfake scam videos - BBC News

AI Content Flood: As AI becomes more adept at generating content – from articles and videos to art and music – we are witnessing a surge in the volume of AI-created content. This influx has the potential to overwhelm traditional content, making it challenging for individuals to discern between human-created, genuine content and AI-generated, potentially inauthentic content.

AI Just Killed YouTube

Filter Bubbles & Echo Chambers: AI-driven platforms can trap users in information silos, exposing them only to similar viewpoints and reinforcing pre-existing beliefs. This can polarize societies and weaken the shared understanding of reality.

Beware online "filter bubbles" | Eli Pariser
How news feed algorithms supercharge confirmation bias | Eli Pariser | Big Think

Emotional & Psychological Impact: Relying on AI for companionship or replacing traditionally human roles can affect our emotional health, potentially diminishing genuine human interaction and altering our social fabric.

The Depressing Rise of AI Girlfriends
The Rise of A.I. Companions [Documentary]

AI, Copyright, and Intellectual Property: The rise of AI-generated content presents complex challenges to traditional notions of copyright and intellectual property (IP). When an AI creates a piece of music, a work of art, or a novel, who owns the rights to that work? Furthermore, AI can be used to replicate styles or mimic human creations, potentially infringing upon original works without directly copying them.

ChatGPT and Generative AI Are Hits! Can Copyright Law Stop Them?

Can artists protect their work from AI? – BBC News

AI-created artwork sparks copyright debate

A.I. Versus The Law

Environmental Impact: Training AI models, especially the larger ones, requires significant computational resources. This can have a sizable carbon footprint and further strain our planet's resources.

AI's hidden climate costs | About That

Peter Henderson: Environmental Impact of AI (and What Developers Can Do)

Statement on AI Risk

Statement on AI Risk | CAIS

Elon Musk calls for artificial intelligence pause

AI pioneer calls to stop before it’s too late | Stuart Russell

AI 'godfather' quits Google over dangers of Artificial Intelligence - BBC News

Limitations of AI

Papers:

AI and regeneration

AI and climate

Can AI Help Solve the Climate Crisis? | Sims Witherspoon | TED

AI Is Poised to Slow Climate Change. Here Are the 6 Industries Where It Can Pack the Biggest Punch (Inc, Jan 2024)

Greenhouse Gas Emissions Monitoring
Power Sector
Manufacturing
Materials Innovation
Food Systems
Road Transport

4 ways AI can help with climate change, from detecting methane to preventing fires (NPR, Jan 2024)
8 ways AI is helping tackle climate change (WEF, Jan 2024)
2024: Is this the year AI helps us fight climate change? (Sifted, Jan 2024)
The AI Revolution in Climate Science (Project Syndicate, Jan 2024)
Crowdsourcing AI Solutions for Climate Change: AI-Startup Winners at COP28 (AI for Good, Dec 2023)
Accelerating climate action with AI (Google, Nov 2023)
AI for Climate Action: Technology Mechanism supports transformational climate solutions (UNFCC, Nov 2023)
Explainer: How AI helps combat climate change (UN, Nov 2023)
Tackling climate change with machine learning (MIT Sloan, Oct 2023)
How To Fight Climate Change Using AI (Forbes, Jul 2022)

Key papers

by Climate Change AI

Tackling Climate Change with Machine Learning - A Summary

by Innovation for Cool Earth Forum

Artificial Intelligence for Climate Change Mitigation

Sounds good? We should also be aware of the Jevons Paradox…

Jevons Paradox & The Rebound Effect

AI & drones for reforestation

Drones and AI team up to reforest Rio de Janeiro | Technology

These seed-firing drones plant thousands of trees each day | Pioneers for Our PlanetUsing Drones to Plant 20,000,000 Trees

Tree-Planting Drones 🌳🌱 | WWF-Australia

Startup's seed-dropping drones can plant 40,000 trees a day

In Brazil, drones take flight in Rio in high-tech reforestation push (Reuters, Jan 2024)
New Tree Tech: AI, drones, satellites and sensors give reforestation a boost (Mongabay, Jul 2023)
Drones and AI: an alternative in strategic reforestation (TecScience, Apr 2023)
Seed-Planting Drones Are Reforesting Canada With Lightning Speed (Reasons to be Cheerful, Feb 2023)
How can AI help protect and sustain our global forest ecosystems - AI for Good (AI for Good, Oct 2022)

AI for talking to animals and plants

How artificial intelligence is helping scientists talk to animals - BBC News

Using AI to Decode Animal Communication with Aza Raskin

How Scientists Are Using AI Tech To Communicate With Animals

Get Ready for AI Religions: Sam Altman, Transhumanism and The Merge

How AI is decoding the animal kingdom (FT, Jan 2024)
How AI could help us talk to animals | The Week UK (The Week, Jan 2024)
New wave of startups uses AI to decode the secret language of plants (Fast Company, Nov 2023)
Artificial Intelligence Could Finally Let Us Talk with Animals (Scientific American, Oct 2023)
How to Use AI to Talk to Whales—and Save Life on Earth | WIRED (Wired, Aug 2023)

Podcasts:

Podcasts of the Year: Talking to Animals using Artificial Intelligence | Scientific American (Science, Quickly; Dec 2023)
Talking With Animals... Using AI (Centre for Humane Technology, May 2023)
Tech Tonic presents: Can AI help us speak to animals? Part one & Part two (FT, Sep 2023)

AI to accelerate innovation

Could artificial intelligence power the future of fusion? (U. Rochester, Sep 2023)
Fast-tracking fusion energy’s arrival with AI (MIT News, Feb 2023)
DeepMind Has Trained an AI to Control Nuclear Fusion (WIRED, Feb 2022)

How Google Solved Nuclear Fusion's Big Problem

AI Cracked the Code of Nuclear Fusion to Destroy Oil and Gas

Can AI solve nuclear fusion? | Demis Hassabis and Lex Fridman

Could AI discover new mathematics/physics?

Multi-Agent Hide and Seek

Will artificial intelligence ever discover new laws of physics? (New Scientist, Nov 2022)
A.I. Is Coming for Mathematics, Too (The New York Times, Jul 2023)

Scientists begin building AI for scientific discovery (University of Cambridge, Oct 2023)
Roll over Oppenheimer: a new AI trained on the laws of physics could unlock the universe (Sifted, Aug 2023)

What is xAI, Elon Musk’s new AI company, and will it succeed? (New Scientist, Jul 2023)
What Elon Musk is really building inside his ChatGPT competitor xAI (Fortune, Nov 2023)

This AI just figured out geometry — is this a step towards artificial reasoning? (Nature, Jan 2024)
Google DeepMind’s new AI system can solve complex geometry problems (MIT Technology Review, Jan 2024)
DeepMind approaches gold standard in complex maths in latest AI breakthrough (FT, Jan 2024)
AI Matches the Abilities of the Best Math Olympians (Scientific American, Jan 2024)
DeepMind's AI finds new solution to decades-old math puzzle — outsmarting humans (The Next Web, Dec 2023)
AI scientists make ‘exciting’ discovery using chatbots to solve maths problems (The Guardian, Dec 2023)

AI to redesign economies

Design goals for a new economic system from Daniel Schmachtenberger's New Economics Series:

Lasting global peace – Conflicts prevented and solved non-violently when needed.
Thriving physical and psychological well-being for everyone. Robust health optimized, disease prevented, and where health issues do arise, they should be cured as completely as possible, as quickly as possible, addressing all causal dynamics, utilizing all the tools available, with minimum side effects.
A transparent, open, information sharing world. Where all the information that could empower people is readily available; all interests are aligned with what is true and systemically positive; disinformation is identified and discarded, etc. Choice making (governance) can only be as good as the relevant information fed into the process (sense-making). [Partial and/or corrupted information make good choicemaking impossible.]
Abundance of all meaningful goods and values for everyone in the system. Where scarcity is intentionally, progressively engineered out of the system as an essential design goal. Where economic valuation is rigorously connected to real value.
A thriving diverse ecology and biosphere. Where new products are made from old products, obsoleting waste and environmental damage from resource acquisition, in a closed loop, upcycling materials economy. With nutrient and microbiome rich soils. No industrial pollutants in the environment. Healthy coral, large fish populations, old growth forests, protected natural areas and nature integrated with the human built world, etc.
A system that supports the maximum freedom of individuals and encourages their unique self-actualization… while encouraging the greatest depth and breadth of interpersonal intimacy and synergy. All people having access to the best resources of health care, education, and creativity that are technologically possible. People incented to create and to support others to create…and to connect meaningfully with other humans…to appreciate the beauty of the world and to add beauty to it.
Good systems of choice making, not damaged by vested interest. Choices that require the participation of many people, and/or that will affect many people, that need maximum integrity and minimum bias. Processes for resolving conflicts that are structurally oriented to prefer optimal conflict resolution.
Anti-fragility and full richness of all complex systems: ecology, physiology, psychology, culture. Resilience, antifragility, health, and aliveness are proportional to self-organizing complexity. Both the safety and real value of a civilization depends on its alignment with these fundamental complex systems.
Antifragility in the presence of exponential technology. Developing the power of gods requires developing the wisdom and care of gods.

How the CIA Destroyed the Socialist Internet: Cybersyn, Part 1 | Kernel Panic | MashableThe British Guru Who Wired Chile’s Cybernetic Socialism: Cybersyn, Part 2 | Kernel Panic| MashableHow an Insurrection Strangled Chile’s Digital Utopia: Cybersyn, Part 3 | Kernel Panic | MashableCybersocialism: Project Cybersyn & The CIA Coup in Chile (Full Documentary by Plastic Pills)

There were almost two internets. Then, the CIA destroyed one (Mashable, Feb 2022)
Enter third-wave economics (FT, Oct 2021)

Hayek promoted markets over central planning on the basis that successful economies need to be run via some form of decentralised collective intelligence. At the time, the best (or least-worst) form of this decentralised collective intelligence was the market. In 2024, however, is it possible that AI and blockchain technology together constitute a new and improved form of economic decentralised collective intelligence, superior to the market mechanism - what some are calling third-wave economics, or Cybersyn 2.0?

Intelligence vs wisdom

What is the Philosophers Stone? Introduction to Alchemy - History of Alchemical Theory & Practice

"We believe Artificial Intelligence is our alchemy, our Philosopher’s Stone – we are literally making sand think."

– The Techno-Optimist Manifesto

From Sand To Silicon: The Making of a Chip | Intel

How To: Turn Sand Into Silicon Chips

"The philosopher's stone is a mythic alchemical substance capable of turning base metals such as mercury into gold or silver. It is also called the elixir of life, useful for rejuvenation and for achieving immortality; for many centuries, it was the most sought-after goal in alchemy."

So You Want to Be a Sorcerer in the Age of Mythic Powers... (The AI Episode) - The Emerald

Josh Schrei "Mythic Powers in the Age of AI"

"The rise of Artificial Intelligence has generated a rush of conversation about benefits and risks, about sentience and intelligence, and about the need for ethics and regulatory measures. Yet it may be that the only way to truly understand the implications of AI — the powers, the potential consequences, and the protocols for dealing with world-altering technologies — is to speak mythically…"

Wise AI:

"It’s up to liberal democracies to demonstrate institutional co-evolution as a third-way between degenerate anarchy and an AI Leviathan."

…similar to Daniel Schmachtenberger's Third Attractor framework:

In Search of the Third Attractor, Daniel Schmachtenberger (part 1)

In Search of the Third Attractor, Daniel Schmachtenberger (part 2)

AI & Moloch: We already have a misaligned superintelligence, it's called humanity

Daniel Schmachtenberger | Misalignment, AI & Moloch | Win-Win with Liv Boeree

Daniel Schmachtenberger: "Artificial Intelligence and The Superorganism" | The Great Simplification

Who is Moloch and What is the MetaCrisis?

AI, Moloch and the Genie's Lamp