There is a war going on. Humanity and nature are on one side and Big Tech is on the other. The two sides are not opposed. They are orthogonal. The human side is horizontal and the Big Tech side is vertical.*

The human side is personal, social, self-governed, heterarchical, open, and grounded in the physical world. Its model is nature, and the cooperative contexts in which competition, creation, and destruction happen in the natural world.

The Big Tech side is corporate, industrial, hierarchical, competitive, mechanistic, extractive, and closed, even though it produces many positive-sum products and services that are good for people and good for nature. It is also, being competitive and rewarding toward winner-take-most outcomes, dominated by giants.

This war has been fought over many other things in the past, especially in tech. But AI is the big one right now—and perhaps the biggest one of all time.

Over the long run, both sides will win, because we need the best of what both bring to the world’s big round table. In the past, this has happened in countless markets, countries, polities, societies, and other contexts. In tech it happened with the conflict between PCs and mainframes, between the open and public Internet and closed private networks, and between open operating systems and closed ones. None of those conflicts have final outcomes, but they have arrived at stable and cooperative states. They arrived at those states because everyone recognized that people can do more with their own tools than those big entities can do on theirs, or by using people as extracted resources.

This conflict is perhaps best represented by the ironies that have been at work within OpenAI and Sam Altman, its prior and current CEO. We can also see it with sparkling clarity in something amazing that Sam said near the close of his keynote at the first OpenAI Developer Conference, in November of last year:

Lets unpack that.

Individual empowerment means exactly what it says. Agency is the ability to take action or to choose what action to take (Cambridge Dictionary). Scale means growth toward ubiquity.

You have individual empowerment and agency when you ride your bike, drive your car, mow your lawn, or button your shirt. You also have it when you work on your laptop, phone, and other smart rectangles. Scale means your power and agency will grow to the max while the same will happen for nearly everyone else. Scale for you alone won’t cut it. When most of us have it, that’s scale.

Technologies extend us. They enlarge our capacities in the world. Intelligence is one of those capacities. So is memory. Your rectangles help a lot with both. That’s why those have already scaled to ubiquity, or close enough.

AI will do the same, but only if it’s personal. Not if it’s  just “personalized.”

As Jamie Smith made clear here a year ago, “your personal AI isn’t really personal.”

ChatGPT4 doesn’t know that yet. I can tell because I just asked:

Me:
What is “personal AI”?

ChatGPT4:
“Personal AI” refers to artificial intelligence systems that are designed to serve individual users through personalized interactions and services. These AI systems learn from the specific preferences, behaviors, and needs of their users to offer more tailored assistance, recommendations, or information. The goal of personal AI is to enhance user experience by adapting to unique personal profiles.

There are various applications of personal AI, including:

1. Virtual Assistants: Like Siri, Google Assistant, and others, these AIs help with daily tasks such as scheduling, reminders, searching information, and controlling smart home devices, all customized to the user’s habits and preferences.
2. Recommendation Systems: Used in streaming services like Netflix or shopping platforms like Amazon, these AI systems analyze your past behavior to recommend movies, shows, or products you might like.
3. Healthcare: Personal AI can track individual health metrics and provide personalized advice, monitor conditions, or even alert medical professionals if needed.
4. Learning and Development: In education, personal AI can adapt learning materials to the pace and learning style of the student, offering a customized educational experience.
5. Gaming: AI in gaming can adapt to provide a challenge tailored to the player’s skill level or alter the game narrative based on choices made by the player.

The development of personal AI focuses on enhancing user engagement by creating more intuitive and responsive interactions that feel uniquely suited to each individual.

For a second opinion, I asked Google’s Gemini the same question.

Me:
What is “personal AI”?

Gemini:
Personal AI refers to artificial intelligence designed to assist and interact with a specific individual. It’s essentially your own AI companion that learns about you and tailors its services to your needs. Here are some key aspects of personal AI:

1. Personalization: Unlike general-purpose AI assistants, personal AI focuses on you. It learns from your interactions, preferences, and habits to become more helpful over time.
2. Focus on Memory and Knowledge: Some personal AI systems aim to become an extension of your mind. They can store your memories, knowledge, and experiences, making them readily accessible and even usable to help with tasks like writing or communication.
3. Varied Applications: Personal AI can have a wide range of applications. Imagine an AI that helps you:
4. Organize your life: Manage schedules, appointments, and reminders.
5. Boost productivity: Assist with tasks, research, and communication.
6. Enhance creativity: Help you brainstorm ideas or generate creative content.
7. Improve learning: Personalize learning experiences and provide targeted information.

Privacy Concerns: Storing personal data on an AI raises concerns about privacy and security.

Overall, personal AI holds promise for a future where technology seamlessly integrates with our lives, offering personalized assistance and support.

To both companies, personal AI is a personalized service—from them. It’s not something you own and control. It’s not about individual empowerment and agency. It doesn’t extend you. It’s hired prosthesis: an extension of them rather than of you.

Let’s name the category: AIaaS—AI as a Service.

But hats off. OpenAI and Gemini do amazing jobs. So do lots of other AI services by wannabe giants. I use AI every day to improve what I write, to generate and fix images, to answer questions that old-fashioned search engines can’t answer or suck at. I even pay extra for some of it (such as ChatGPT4 and Adobe’s Creative Suite). And it seems they all get better, all the time, at everything. Won’t personalization be good enough, eventually?

No. Because they aren’t you. They also aren’t yours, so they can’t extend you. They can seem to. But they don’t. They also don’t have access to most of your private life. Nor should they.

But your private life could use some AI help. For example—

• Schedules, including on your calendars, past and future
• Health data, including all your medical reports, prescriptions, appointments, insurance information, past and present providers, plus what your watch, phone, and other devices record about you
• Financial records, including bills, receipts, taxes, and anything called an account that involves money
• Travel, including all the movements your phone (and phone company), watch, and car record about where you go and where you’ve been
• Work—past and present, including whatever HR systems know or knew about you
• Contacts—all the people, businesses, and other entities you know
• Business relationships, with brokers, retailers, service providers, whatever
• Subscriptions, including all those “just $1 for the first four weeks” offers you’ve accepted, plus other forms of screwage that are stock-in-trade for companies selling subscription systems to businesses.
• Property, including all the stuff on your shelves, floors, closets, garages, and storage spaces—plus your stocks and real estate.

It’s not easy to visualize what a personal AI might do for those, but let’s try. Here’s how Microsoft’s Copilot (or whatever it’s called this week) did it for me before I got rid of all its misspellings and added my own hunks of text:

All that stuff is data. But most of it is scattered between apps and clouds belonging to Apple, Google, Microsoft, Amazon, Meta, phone companies, cable companies, car makers, health care systems, insurance companies, banks, credit card companies, retailers, and other systems that are not yours. And most of them also think that data is theirs and not yours.

To collect and manage all that stuff, you need tools that don’t yet exist: tools that are yours and not theirs. We could hardly begin to imagine those tools before AI came along. Now we can.

For example, you should be able to take a picture of the books on your shelves and have a complete record of what those books are and where you got them. You’ll know where you got them because you have a complete history of what you bought, where and from whom. You should be able to point your camera in your closets, at the rugs on your floors, at your furniture, at the VIN number of your car that’s visible under your windshield, at your appliances and plumbing fixtures, and have your AI tell you what those are, or at least make far more educated guesses than you can make on your own.

Yes, your AI should be able to tap into external databases and AI systems for help, but without divulging identity information or other private data. Those services should be dependent variables, not independent ones. For full individual empowerment and agency, you need to be independent. So does everyone else with personal AI.

Now imagine having a scanner that you can feed every bill, every receipt, every subscription renewal notice, and have AI software that tells you what’s what with each of them, and sorts records into the places they belong.

Ever notice that the Amazon line items on your credit card bill not only aren’t itemized, but don’t match Amazon’s online record of what you ordered? Your personal AI can sort that out. It can help say which are business and personal expenses, which are suspicious in some way, what doesn’t add up, and much more.

Your personal AI should be able to answer questions like, How many times have I had lunch at this place? Who was I with? When was it we drove to see so-and-so in Wisconsin? What route did we take? What was that one car we rented that we actually liked?

Way back in 1995, when our family first got on the Internet over dial-up, using the first graphical browsers on our PC, and e-commerce began to take off with Amazon, eBay, and other online retailers, my wife asked an essential question: Why can’t I have my own shopping cart that I take from site to site?

Twenty-nine years later, we still don’t have the answer, because every retailer wants you to use its own. And we’re stuck in that system. It’s the same system that has us separately consenting to what sites ironically call “your privacy choices.” And aren’t.

There are countless nice things we can’t have in the digital world today because we aren’t people there. We are accounts. And we are reduced to accounts by every entity that requires a login and password.

This system is a legacy of client-server, a euphemism for slave-master. We might also call it calf-cow, because that’s how we relate to businesses with which we have accounts. And that model is leveraged on the Web like this:

We go to sites for the milk of content and free cookies, whether we want them or not. We are also just “users.”

In the client-server world, servers get scale. Clients have no more scale than what each account—each cow—separately allows. Sure, users get lots of benefits, but scale across many cows is not one of them. And no, “login with Google” and “login with Facebook” are just passes that let calves of ruling cows wander into vassal pastures.

For individual empowerment and scale to happen, we need to be self-sovereign and independent. Personal AI can give that to us. It can do that by solving problems such as the ones listed above, and by working as agents that represent us as human beings—rather than mere users—when we engage with Big Tech’s cows.

This will be a fight at first, because the cows think they run all of nature and not just their own farms. And $trillions are being invested in the same old cattle industry, with AI painted all over the new barns. Comparatively speaking, close to nothing is going toward giving independent and self-sovereign individuals the kind of power and scale Sam Altman says he wants to give us but can’t because he’s on the big cow side of this thing.

So where do we start?

First, with open source code and open standards. We have some already. Llama 3, from Meta AI, is “your own intelligent assistant,” and positions Meta as a more open and user-friendly cow than OpenAI. Meta is still on the top-down Big Tech side of the war we’re in. But hell, we can use what they’ve got. So let’s play with it.

Here on the ground there are all these (with quotage lifted from their sites or reviews such as this one)—

• MindsDB: “an open-source AutoML framework”
• Alt.ai: “It’s an A.I. which aims to digitize users’ intentions and place it on the cloud to let our clones deal with all digital operations.”
• Keras: “a multi-backend deep learning framework, with support for JAX, TensorFlow, and PyTorch”
• PyTorch: “Python package that provides two high-level features: Tensor computation (like NumPy) with strong GPU acceleration, and Deep neural networks built on a tape-based autograd system
• Tensor Flow: “open-source framework for machine learning”
• CoreNet: a deep neural network toolkit for small and large-scale models, from Apple
• Haystack: an “open source Python framework by deepset for building custom apps with large language models (LLMs).”
• Image Super-Resolution (ISR): “(an) open source tool employs a machine learning model that you can train to guess at the details in a low-resolution image:
• Blender: “A rich interface and numerous plugins make it possible to create complex motion graphics or cinematic vistas”
• DeepFaceLab: “open source deepfake technology that runs on Python”
• tflearn: “an advanced deep learning library”
• PYTensor: “a Python library that allows you to define, optimize/rewrite, and evaluate mathematical expressions involving multi-dimensional arrays efficiently.” (Was Theano)
• LM Studio: “Discover, download, and run local LLMs”
• HuggingFace Transformers: “a popular open-source library for Natural Language Processing (NLP) tasks”
• Fast.ai: “a library for working with deep learning tasks”
• OpenCV: “a popular Computer Vision and Image Processing library developed by Intel”
• Detectron2: “a next-generation library that provides advanced detection and segmentation algorithm” and “a PyTorch-based modular object detection library”
• Ivy.ai: “an open-source deep learning library in Python focusing on research and development”
• OpenAssistant: “a project aimed at giving everyone access to a great chat-based large language model”
• PaddleNLP: “a popular open source NLP library that you can use to glean search sentiment and flag important entities”
• Delphi.AI: “Clone yourself. Build the digital version of you to scale your expertise and availability, infinitely.”
• Fauxpilot: “This is an attempt to build a locally hosted alternative to GitHub Copilot. It uses the SalesForce CodeGen models inside NVIDIA’s Triton Inference Server with the FasterTransformer backend.”
• Ray: “An open source framework to build and scale your ML and Python applications easily”
• Solid: “Solid is a specification that lets individuals and groups store their data securely in decentralized data stores called Pods. Pods are like secure web servers for data. When data is stored in a Pod, its owners control which people and applications can access it.”
• Sagen.ai: “Your very own AI Personal Assistant to manage your digital life.”
• YOLOv7: “is one of the fastest and most accurate open source object detection tools. Just provide the tool with a collection of images full of objects and see what happens next.”

—and lots of others that readers can tell me about. Do that and I will add links later. This is a work in progress.

Below all of those we still need something Linux-like that will become the open base on which lots of other stuff runs. The closest I’ve seen so far is pAI-OS, by Kwaai.ai, a nonprofit I now serve as Chief Intention Officer. I got recruited by Reza Rassool, Kwaai’s founder and chair, because he believes personal AI is required to make The Intention Economy finally happen. So that was a hard offer to refuse. Kwaai also has a large, growing, and active community, which I believe is necessary, cool, and very encouraging.

As with most (maybe all) of the projects listed above, Kwaai is a grass-roots effort by human beings on the natural, human, and horizontal side of a battle with giants who would rather give us personalized AI than have us meet them in a middle to which we will bring personal AI powers of our own. In the long run, we will meet in that middle, because personal AI will be better for everyone than personalized AI alone.

Watch us prove it. Better yet, join the effort.

---

*I am indebted to Lavonne Reimer for introducing and co-thinking the horizontal vs. vertical frame, and look forward eagerly to her own writings and lecturings on the topic.

Elon Musk wants to colonize Mars.

This is a very human thing to want. But before we start following his lead, we might want to ask whether death awaits us there.

Not our deaths. Anything’s. What died there to make life possible for what succeeds it?

From what we can tell so far, the answer is nothing.

To explain why life needs death, answer this: what do plastic, wood, limestone, paint, travertine, marble, asphalt, oil, coal, stalactites, peat, stalagmites, cotton, wool, chert, cement, nearly all food, all gas, and most electric services have in common?

They are all products of death. They are remains of living things or made from them.

Consider this fact: about a quarter of all the world’s sedimentary rock is limestone, dolomite and other carbonates: remains of beings that were once alive. The Dolomites of Italy, the Rock of Gibraltar, the summit of Mt. Everest, all products of death.

Even the iron we mine has a biological source. Here’s how John McPhee explains it in his Pulitzer-winning Annals of the Former World:

Although life had begun in the form of anaerobic bacteria early in the Archean Eon, photosynthetic bacteria did not appear until the middle Archean and were not abundant until the start of the Proterozoic. The bacteria emitted oxygen. The atmosphere changed. The oceans changed. The oceans had been rich in dissolved ferrous iron, in large part put into the seas by extruding lavas of two billion years. Now with the added oxygen the iron became ferric, insoluble, and dense. Precipitating out, it sank to the bottom as ferric sludge, where it joined the lime muds and silica muds and other seafloor sediments to form, worldwide, the banded-iron formations that were destined to become rivets, motorcars and cannons. The is the iron of the Mesabi Range, the Australian iron of the Hammerslee Basin, the iron of Michigan, Wisconsin, Brazil. More than ninety percent of the iron ever mined in the world has come from Precambrian banded-iron formations. Their ages date broadly from twenty-five hundred to two thousand million years before the present. The transition that produced them — from a reducing to an oxidizing atmosphere and the associated radical change in the chemistry of the oceans — would be unique. It would never repeat itself. The earth would not go through that experience twice.

Death produces building and burning materials in an abundance that seems limitless, at least from standpoint of humans in the here and now. But every here and now ends. Realizing that is a vestigial feature of human sensibility.

Take for example, The World Has Plenty of Oil, which appeared in The Wall Street Journal ten years ago. In it, Nansen G. Saleri writes, “As a matter of context, the globe has consumed only one out of a grand total of 12 to 16 trillion barrels underground.” He concludes,

The world is not running out of oil any time soon. A gradual transitioning on the global scale away from a fossil-based energy system may in fact happen during the 21st century. The root causes, however, will most likely have less to do with lack of supplies and far more with superior alternatives. The overused observation that “the Stone Age did not end due to a lack of stones” may in fact find its match.

The solutions to global energy needs require an intelligent integration of environmental, geopolitical and technical perspectives each with its own subsets of complexity. On one of these — the oil supply component — the news is positive. Sufficient liquid crude supplies do exist to sustain production rates at or near 100 million barrels per day almost to the end of this century.

Technology matters. The benefits of scientific advancement observable in the production of better mobile phones, TVs and life-extending pharmaceuticals will not, somehow, bypass the extraction of usable oil resources. To argue otherwise distracts from a focused debate on what the correct energy-policy priorities should be, both for the United States and the world community at large.

In the long view of a planet that can’t replace any of that shit, this is the rationalization of a parasite. That this parasite can move on to consume other irreplaceable substances it calls “resources” does not make its actions any less parasitic.

Or, correctly, saprophytic; since a saprophyte is “an organism which gets its energy from dead and decaying organic matter.”

Moving on to coal, the .8 trillion tons of it in Wyoming’s Powder River Basin now contributes 40% of the fuel used in coal-fired power plants in the U.S. Here’s the biggest coal mine in the basin, called Black Thunder, as it looked to my camera in 2009:

About half the nation’s electricity is produced by coal-fired plants, the largest of which can eat the length of a 1.5-mile long coal train in just 8 hours. In Uncommon Carriers, McPhee says Powder River coal at current rates will last about 200 years.

Then what? Nansen Saleri thinks we’re resourceful enough to get along with other energy sources after we’re done with the irreplaceable kind.

I doubt it.

Wind, tide, and solar are unlikely to fuel aviation, though I suppose fresh biofuel might. Still, at some point, we must take a long view, or join our evolutionary ancestors in the fossil record faster than we might otherwise like.

As I fly in my window seat from place to place, especially on routes that take me over arctic, near-arctic, and formerly arctic locations, I see more and more of what geologists call “the picture”: a four-dimensional portfolio of scenes in current and former worlds. Thus, when I look at the seashores that arc eastward from New York City— Long Island, Block Island, Martha’s Vineyard, Nantucket, Cape Cod—I see a ridge of half-drowned debris scraped off a continent and deposited at the terminus of an ice cap that began melting back toward the North Pole only 18,000 years ago—a few moments before the geologic present. Back then, the Great Lakes were still in the future, their basins covered by ice that did not depart from the lakes’ northern edges until about 7,000 years ago or 5,000 B.C.

Most of Canada was still under ice while civilization began in the Middle East and the first calendars got carved. Fly over Canada often enough and the lakes appear to be exactly what they are: puddles of a recently melted cap of ice. Same goes for most of the ponds around Boston. Every inland swamp in New England and upstate New York was a pond only a few dozen years ago, and was ice only a dozen or so centuries before that. Go forward a few thousand years and all of today’s ponds will be packed with accumulated humus and haired over by woods or farmland. In the present, we are halfway between those two conditions. Here and now, the last ice age is still ending.

As Canada continues to thaw, one can see human activity spark and spread across barren lands, extracting “resources” from ground made free of permafrost only in the last few years. Doing that is both the economic and the pestilential thing to do.

On the economic side, we spend down the planet’s principal, and fail to invest toward interest that pays off for the planet’s species. That the principal we spend has been in the planet’s vaults for millions or billions of years, and in some cases cannot be replaced, is of little concern to those spending it, which is roughly all of us.

Perhaps the planet looks at our species the same way and cares little that every species is a project that ends. Still, in the meantime, from the planet’s own one-eyed perspective, our species takes far more than it gives, and with little regard for consequences. We may know, as Whitman put it, the amplitude of time. We also tend to assume in time’s fullness all will work out.

But it won’t.

Manhattan schist, the bedrock anchoring New York City’s tallest buildings, is a little over half a billion years old. In about the same amount of time, our aging Sun, growing hotter, will turn off photosynthesis. A few billion years later, the Sun will swell into a red giant with a diameter wider than Earth’s orbit, roasting the remains of our sweet blue planet and scattering its material out into the cosmos, perhaps for eventual recycling by stars and planets not yet formed.

In a much shorter run, many catastrophes will happen. One clearly is what our species is already doing to the planet during what geologists correctly call the Anthropocene. I suppose that’s a good reason for Elon and crew to “save” a few members of our vain little species. But why fuck up Mars before we’re done fucking up Earth, when there’s still some leverage with the death we have at home and that Mars won’t begin to have until stuff dies on it?

I’ve always been both an optimist and a realist. Specifically, I’m an optimist for at least the short run, by which I mean the next few dozen years. But I’m a pessimist for our civilization — and our species. Death is always a winning bet.

But hey, maybe nature knows better what to do with us than we do.

The first ancestor of this piece appeared in blogs.harvard.edu on 4 March 2008. The second is here on Medium.

I just returned to DALL-E 3 after using its Microsoft version (currently called Copilot | Designer) for a while. But I can’t get in.

See how it says “Try in ChatGPT︎?” When I do that, it goes to https://chat.openai.com/. After I log in there, it offers no clue about where DALL-E 3 is. So I asked, “How do I get to the DALL-E 3 page after I’m logged in here?”

ChatGPT answered, “As of my last update in January 2022, DALL-E 3 is a hypothetical iteration of the DALL-E model, and there’s no information about its existence or how to access it. If it were to be released, you might typically find access through OpenAI’s website or platform if they make it available to the public. However, without further updates or announcements, it’s challenging to provide specific instructions. If OpenAI releases DALL-E 3 or any similar model in the future, you can check their official website or announcements for information on how to access it.”

Additional useless dialog followed. So now I’ll ask the Lazyweb. Does anyone know how to get into DALL-E 3?