When most people think of AI's potential to help humanity in 10, 100, or 1,000 years, they think much too small.
They think about what it does today, and then just imagine "more of that." Like AI agents providing more employees, or tutors, or companions.
All tremendously useful, but still tactical vs. strategic.
The best use case I've thought of so far is something far more general. It's a pattern for solving problems generally, as opposed to just a set of one-off solutions.
One way to characterize the universal challenge for anything alive is how to go from its current situation to its desired situation.
This applies to finances, employment, starting a business, relationships, raising children, health, and so on. You can imagine this as the eternal problem for anything with goals.
Ultimately it's a world-model problem. Here are the pieces:
In other words, one end-tier capability for any intelligence—human or otherwise—is the ability to turn any problem into a current-to-desired-state transition problem—at any scale.
This requires extraordinary understanding of how things work and significant intelligence, but what it needs most is what scientists call working memory. This is like the size of the space that a given intelligence can operate on at any given time.
It needs to hold all these things in its mind at once and then apply its intelligence, creativity, etc. These current and desired states will be summarized, compressed, and otherwise efficiently minified, but they will still be vast, so working memory (context size) matters a lot.
One of the primary use cases for this model of problem solving comes down to the most common questions ever.
You ask these questions when you're looking for a life partner, when you're trying to hire employees, trying to raise a child, and a thousand other things.
Being human means constantly making decisions like these.
If you squint your eyes a bit, you'll be able to see that even these types of questions are just another way to think about state transition.
While we are asking technically different questions, about different situations, and at different times, they can all be reframed as:
Which of these choices will get me closer to my desired state?
A State Managing AI will then model a dozen or a billion futures where you did this vs. that, and then advise you which one to choose.
It's like asking a human life coach the same thing, except the life coach can't hold petabytes of world knowledge in its mind at the same time and then apply beyond-human intelligence to the analysis.
To illustrate the point further, let's look at some examples of how this model can be applied to common problems in personal life, business, and other everyday challenges.
The eternal problem of finding the right person to build a life with.
Problem | State-Transition Implementation |
---|---|
Which of these guys should I choose? | Model out a million lives with him vs. the other guy, using 200 billion life parameters, compare those outcomes to ideal state, provide recommendations |
Should I still be dating her? | Model out a million lives with her vs. someone more ideal, calculating for current level of bonding and shared experience, compare that to desired state, and advise if you should stay or go |
Where Should I Go to Meet People? | Model the top meeting places, including online forums, dating apps, bars, churches, hobby spots, etc., and the people likely to hang out in those places. Take representative people and model out a life with them and compare to ideal situation |
One of the most common problems in business: seeing why people are cancelling or not renewing.
Problem | State-Transition Implementation |
---|---|
How do I keep this customer from cancelling? | Create a desired state for the customer and compare that to their current state with your company. Advise on what you can say, or change about their expeirence to make their situation best match their ideal state. |
Which of my current customers are likely to cancel? | Create a current state model for all current customers, create an ideal state where they're perfectly happy with your product/service, analyze the difference between the two, recommend changes / take actions |
This is for a startup that takes the context from any situation, like two friends riding bikes together in nature, and plays the perfect song (or soundtrack) as background music
Problem | State-Transition Implementation |
---|---|
What's the perfect song to be playing in the background at this moment? | Build the current state maps for all human listeners, build their ideal state maps (e.g., feeling more connected to each other in this intimate moment), look at their individual and shared pasts, pick the perfect song to play to make that moment one they'll remember forever |
These were just a few random examples of the types of problems that can be addressed using this framework. Here is a short list of others that I'm sure you could add 20 more to without much effort.
Problem Space | State-Transition Model |
---|---|
Vulnerability Management | Current systems, current vulnerabilities, lack of the worst ones on the most important systems, transition plan, action |
Tutoring | Current curiosity, self-confidence, knowledge, and curiosity level, ideal self-confidence and competence level, transition plan, incoming questions / challenges, response interactions |
Starting and Managing a Family | Current state without a partner, determining ideal state with partner and future family, break the steps into pieces, finding a partner, finding a place to live, schools, saving money for college, etc. |
Career Growth | Current career with title, salary, responsibilities, ideal career with title, salary, responsibilities, transition plan, action |
Starting a Business | Current life/business state, business idea, ideal business state, ideal life state with that business, transition plan, recommended set of actions, taking action for them |
Nearly all human challenges can be viewed within this lens of, "I am here, but I wish I were there.", and the more powerful AI becomes the more capable it will be at helping people perform those transitions.
One of the biggest technology implications of this model is that it's a replacement for software.
Everything we do, and everything we build, is designed to help us change the world to a more desired state. Plumbing is designed to get water from here to there. Cars take people from A to B. Businesses change customer state from unhappy to happy, and your financial state from struggling to comfortable.
Software is the same. It's technology that helps you understand and manage state changes.
When you have sufficient context about current and desired state for a given problem set, you don't actually need different software.
You may need different data, and perhaps different types of UI/UX, but ultimately you're dealing with questions and actions—either point in time or continuously running—that operate against those different states.
Think about hiring and managing employees inside of a company, and making sure they're working on the right things for the business.
How many software verticals is that?
Why have all these different software packages when you could just have their source data as part of a unified context that AI can hold in its mind all at once?
The advantages of this model are many, but the most obvious one is the most powerful.
These pieces all know about each other.
Here's are some examples of common business questions that are excruciatingly manual and time/talent intensive:
Answering these types of questions is a full-time job for dozens, hundreds, or thousands of people within a company. And it's highly manual, requiring people to organize meetings, query dozens of systems, collect and analyze data, come up with conclusions, and then do another set of meetings to communicate those conclusions.
And that's just one point in time!
In 15-60 days the company has changed so much it needs to be done again. These types of analysis are perpetually stale, and they require many of your best people to try to maintain them.
AI State Management turns this into questions you ask the system as often as you want, and the answers easily cross all these "software" domains and return better results than if you'd spent weeks doing it manually (which costed millions of dollars in human work).
As you might have noticed, that work above is often handled by Project Managers. You need multiple, dedicated people to try to do what comes naturally to AI—looking at all the moving pieces at once so you can manage them.
In this model, Project Management becomes questions asked to the system, and answers are returned in seconds rather than weeks or months. In any format you want.
Oh you need charts for that? Tables organized this way or that way? Reports on current state of the various projects? For different audiences? Those all take seconds to generate.
The same applies to figuring out what to build next, and how to manage the product development process.
These don't remove the need for a Product Manager, actually, which I'm quite happy about. Because certain work—like Project Management—requires one to be opinionated about what to build and why.
You can't just listen to customers, and you can't just listen to the engineers. You have to listen to all of them, and executive leadership, and the market, and the competition, and the future, and the past, and the present.
What this model does, however, is turn the entire endeavor into a current state and desired state problem.
The vision that comes from leadership, and the product manager, all goes into the desired state. Sentiment and feedback from users goes in as well. The roadmap. The current state of development. The engineers we have working on the team. All the support issues. Etc. Everything is part of the product's context (which is also aware of company goals as well).
Now the Product Manager can manage the entire thing by interaction with the system.
These types of questions will evaluate years of code changes, support tickets, internal meeting notes, countless design documents, investor comments, social media posts, etc., and respond back in seconds.
Today that's a full-time job for a team of people, takes days or weeks of research across multiple software products, and it's instantly stale as soon as you finish.
That's the best part. The human part.
create_art_prompt
Fabric prompt on the text of the article.