My AI Prediction: Employment will transform into micro-entrepreneurship + Tax compliance will become effortless

For me the future of AI is clear: If everyone can do everything, why would anyone still want to do something for someone else? All employees will be automated away, so of course the people will not stay unemployed forever, but at some point the governments would recognize this and start big education campaigns for making people learn entrepreneurship, creative thinking, emotional self-reflection and planning to realize their dreams.

I believe that we will see a massive mobilization of micro-entrepreneurs, of people becoming independent from any employer and following their own passion.

For an economy to be inclusive to truly everyone, the cost of running any kind of company must be zero. Founding should not cost anything, making the annual report shouldn’t cost anything, a tax advisor should not be necessary for regular operations, knowing a thousand bookkeeping accounts should not be necessary, a proper ERP should just be “built-in” and be the open source market standard. Only that’s what I would call actual economic inclusivity, where founding and running a company properly must be accessible to anyone, so that you can actually concentrate on running your business instead of all the administrative stuff around it.

That kind of software is of course exactly what I’m working on.

What I’ve been working on intensely over the last about 1.5 years. I could give a long lecture about what I’ve done in this direction already. It’s very theoretical at parts, but I’m slowly making progress to turning it into real code.

The purpose of the system we’re creating is to be able to turn any contract, statutes document, policy document or law into a digital administrative structure.

Basically, what COBOL always wanted to be: Business people formulating what they want to build. But while COBOL quickly became clunky, with LLMs we get another shot at solving this problem.

But the most important issue is to migrate existing systems into this more harmonized structure, and to always keep them upgradable.

Schemas must be decided upon cooperatively, but when an upgrade has been decided upon, it must also be rolled out to all network participants quickly.

Employment will, in large parts, not exist anymore.

So basically I think that AI will give back humans their humanity.

AI will be a very thin veil: the easy input layer. Within the machine, everything must be according to formally verified rules. But to get data into it, at some point the translation of messy human speech into a more structured, legal form must happen. And it should best happen at the very beginning (i.e., at the input layer), so that issues in this translation process from human language into this more formal type of language can be detected by the end-user immediately.

Computers will not become human, but humane.

They will not consist of windows and little icons and big and small messy buttons anymore. No one will want commercial software anymore if it doesn’t talk the universal language of computers.

The solution is not one shared language, but the intentional interoperability of every language.

First we identify common concepts that are just obvious facts of reality: Persons, Coordinates, Time, Documents, SI units, these kinds of things.

Then we identify more high-level concepts. For example a German GmbH (limited liability company) is constructed in the following two steps:

• a law: the “Regulation on the establishment and maintenance of the company register” (https://www.gesetze-im-internet.de/gesrv/anlage_1.html) → this defines the schema of the German company register

• a certain type of document (in German: the “Registerblatt”, which is the proof that a certain type of company actually exists and what its status is), which you can get from https://www.handelsregister.de/rp_web/welcome.xhtml?cid=1

Before you have this document, you cannot reliably assume that the GmbH exists at all.

Same thing (extracting schema from law or contract → instantiating schema with data) will happen all the time. In both parts LLMs are needed at the input layer:

• to read out what the structure is

• to structure messy input data according to such a structure

Once the structure has been determined and approved by experts, data should of course preferably be input in a way where it doesn’t even become unstructured in the first place, so that no LLMs are needed to handle input, as the user prompt has already been maximally specific and unambiguous. The opposite of this desirable case would be, for example, receiving documents as scans. But, as always, real things are the ground truth, so that when data is read out from a scanned document, and at some point someone detects a mistake made by the OCR, the document will always be immediately accessible, as the chain can always be traced back, no matter how far into the different data processing steps.

We will just need to build this structure once, do it properly, and then we’ll have a super valuable asset for all of humanity.

What did Apple want in the beginning? Making computers more beautiful, more easily accessible. Make them do exactly what we want, as quickly as possible. And consequently, radically reducing time people spend on computers. I think it’s necessary to return to these roots.

Making computers beautiful makes it necessary to have a fundamental shift of how computing is organized. From centralized to peer-to-peer and end-to-end-encrypted. From managed by corporations and governments to self-managed. Collaboratively standardized schemas. Personal data repositories. Data sovereignty/Self-Sovereign Identity. Auto-sync with all data silos which need to be attached. Event-driven architectures, real-time by default. Smooth workflows, simply because “all systems are go.”

It’s gonna feel sooooo good.