On the Economic and Political Aspects of Open Source. Q&A with Franz Kiraly

Q1. You’ve been a strong advocate for open source AI software. How would you explain to policymakers why open source AI infrastructure is as critical to the international economic landscape as physical infrastructure like roads or telecommunications networks? What are the economic costs of overlooking this?

FK: Let’s start with an economic metaphor, where instead of AI we build cars. Imagine a world where everyone has easy access to car parts, e.g., via a sci-fi replicator such as in Star Trek that magically produces things on a button press. Cars themselves, however, do not come out of replicators, and have to be assembled from parts. The car part blueprints need to be maintained and regularly updated to newest technologies, but remain abundantly available.

This car-world seems alien to us, but from an economic perspective it is exactly how the AI value and supply chain works.

Open source software – the car parts in the metaphor – are the basis for all commercial AI systems. For example, all chatbots and agentic AI use neural networks, which are mostly built using the open source library torch or similar components. Machine learning applications typically use the open source library scikit-learn, and so on. These libraries are free to download and use, available on a button press.

From an economic perspective, the value chain for AI, and that for “replicator world” cars is the same. In terms of abundance, this is similar to the real world value chain for tea – which can use water that comes out of the tap. The water supply, the car-world replicators, and open source software are all commodity infrastructure in the sense of abundant, low-cost availability and reliability. Unlike water however, significant innovation work must go into the car parts (or AI components), constantly.

In all three scenarios (tea, replicator world cars, AI), it should be an obvious idea that the infrastructure itself (water, car parts, open source) must be maintained, developed, and protected – possibly even as public infrastructure guaranteed by the state.

A Harvard Business School paper estimates the overall demand value of the open source ecosystem at 8.8 trillion dollars.

In the tea example, this number would be “total water demand”.

8.8 trillion dollars sounds like a lot, and it is. For comparison: the GDP of the EU is around 20 trillion dollars currently! We are not talking about peanuts, but goods in the order of entire nation state economies. 

This also means: open source software can make and break the future of nations. Others have also taken note – such as the democratically unaccountable tech oligarchs, who want to force their ideas upon the world. In the car-world: whoever controls the replicators will corner the car market.

Let’s try to unpack some common misunderstandings now, that often confuse – and are sometimes intentionally used to confuse – decision makers:

• In the fundamental layers of the AI space, there is no competition anymore between closed or open software, like “windows vs linux”. In the car-world metaphor: there are only replicator car parts, no one builds car parts to sell them anymore. The confusion here: cars are still being sold, so people think this also applies to car parts; and car parts are still being developed, so people think they must be sold.
• The misconception to attribute the entire value to the “car” assembly, while ignoring value creation in the “car parts” like engine or battery.
• the term “infrastructure” is overloaded, it is also used for hardware and cloud systems on which AI runs. In the water example, this is a confusion of water pipes and water itself. Both are important, but it is an obvious mistake to focus only on water pipes and not water production.

The reason for the above: end users, and the typical decision maker, never interact with the “car parts”, only the “cars”. While everyone has an exposure to AI nowadays, only experts are aware of deeper parts of the value chain.

Overall, decision makers must understand the AI supply chains and economic value creation chains in play. Failing to understand the AI supply chain – with open source as crucial value creator – leaves one open to adversaries and fraud.

For example, dodgy AI vendors may sell a very thin wrapper around (cost-free) open source. Hyper-scalers may intentionally destroy open source infrastructure to widen their own competitive advantage. Not being able to “see” either is a form of incompetence. It is, obviously, bad for buyers of AI, and nation state economies alike.

Q2. Despite the growing importance of open source AI, many politicians seem to treat it as a niche technical issue rather than a strategic priority. What do you think explains this disconnect between the technical community’s understanding and political recognition? What would need to change for this to shift?

FK: To start, I indeed agree with the premise in the question. Politicians generally do not seem to understand the importance of open source and the AI supply chain.

This is, as many roots of big problems, a complex topic on the interface of technology and human nature.

First, democratic politicians tend to mirror their electorate in their selection of “topics of the day”. So, if the electorate does not understand the AI supply chain, and does not feel there is a need for action, democratic politicians will not put it on the agenda. Classically, technology topics go over the heads of the general populace, so tend to show up not proactively, but reactively, often when it is already too late to act.

The public perception is also frequently distorted by fearmongering (“AI will destroy us”), which distracts from real issues like socio-economic impact, data and intellectual property protection, or the pressing technological sovereignty questions. This distortion may even be intentionally placed by our adversaries to exacerbate the first problem.

Second, even well-meaning and ethical politicians have a hard time to come up with sound policies without a very good technical understanding of the AI supply chain. It is also difficult to find competence outside top tier tech companies, certainly this does not exist in the usual politicians’ office. Hence, politicians will have to resort to external input, which typically leads to two channels:

• Academic experts. Unfortunately (yet not widely known), most academics are also incompetent in the AI topic. This can be gleaned from the general lack of footprint of academic institution in the AI supply chain. Talking here about software and usage footprint, not number of papers or citations (irrelevant metrics nowadays). Of course, to make matters worse, academics are often endowed with huge egos, so this channel will produce useless noise at best, and confusion with completely misplaced priorities at worst.

• Corporate lobbyists. The AI oligarchs are sending their best lobbyists to confuse and misdirect politicians away from socio-economic or supply chain issues. In many places, professional communications people masquerade as open source spokespeople, looking like plausible points of contact but ultimately misdirecting funds or neutralizing effort. Sometimes these lobbyists even subvert a country’s sovereign technology institutions – I have seen institutional capture by an unholy alliance of lobbyists and opportunists, not only once.

Sadly, what politicians want is rarely found – a hypothetical, proper lobby for the communities that are historically and really building open source AI software. This is due to how the socio type has historically operated – not as professional communities that with a political voice, but mostly technical communities of practice, comprised of technologists, below the radar of every public or political perception. 

Politicians must recognize this if they want their AI policies to be effective – ensure that those who create and maintain the technology are heard, rather than giving attention only to those who shout loudest (or have the financial backing to subvert political attention to the detriment of society).

Third, naturally, there is also a lot of corruption and mis-financing. The fewest citizens or politicians actually understand the AI supply chain, yet everyone agrees that huge efforts are necessary.

This means a lot of money floating around, but without a clear consensus on how accountability looks like! This makes AI a paradise for fraud – starting from soft corruption such as crony funding, to hard corruption such as direct embezzlement. Academics have also discovered this as a funding opportunity, for senseless academic projects led by politically influential players who are “professor for AI” only since when they discovered that there was money in it. Personally, I would guess that more than 95% of “AI” money in academia is completely misspent, and maybe more than 80% in an industrial or startup setting.

Overall, we are looking at a significant political cluster fudge – tackling this will not be easy.

A first step for misallocation of funding must be to make it strictly track record based – e.g., using KPI that reward traction in the AI supply chain. AI is software, and use can be measured in downloads. No one should ever give money to academic “AI” groups who have never produced anything that anyone has ever used, or AI startups where none of the staff have ever written a line of code. The simplest of track record checks will prevent most of it and the worst of it.

As conceptually simple as this suggestion is, it will be practically hard to implement: the masses of corrupt beneficiaries of the current system will not just let their hard stolen loot disappear – hence, any real move towards this will elicit massive resistance and retaliation, which politicians will be reluctant to be the target of.

Q3. We’re seeing increasing consolidation of AI development within a handful of US tech companies. From your perspective, what are the implications of this concentration of control over AI systems and infrastructure? How does this compare to the role open source could play in creating a more balanced landscape?

FK: Again I agree with the premise – yes, a few US companies are increasingly controlling AI development. We also see increasing control attempts directed at the full AI supply chain, including the open source communities that historically were, and still are, mostly, open. There are intentional attempts to destroy software projects in the open source space, or to take operational control with the aim to restrict access.

These attacks on individual projects and individual maintainers are often massive, but are invisible to the public eye, with plenty of plausible deniability built in, for instance mediated through allied non-profits or academics. There is clearly a playbook, I suspect that this playbook is not new and has been deployed during earlier tech booms such as operating systems, databases, or the internet.

A prominent early example of this adversarial behaviour towards open source is openAI – the company behind ChatGPT – dropping support for openAI gym (now gymnasium). Unknown to many, openAI in its early days indeed relied on open source AI software projects like openAI gym. More recently, openAI has mostly abandoned the idea of being “open” in the sense of open source or open supply chain – even though they still carry the idea as a relic in their brand name. 

Less prominent examples involve orchestrated direct attacks at key personnel in open source projects, to replace them with corporate loyalists, or to weaken a project that competes with one under corporate control.

Another neuralgic point are package distribution pathways, that is, the servers via which the software components are made discoverable and installable. For the Python programming language, this is the Python package index (PyPI), administered by a US non-profit, the Python Software Foundation. In my opinion, there has been an authoritarian takeover in the last years – dissenters are bullied or banned from elections for flimsy reasons, control is tightened around a small set of loyalists, some directly employed by US big tech. There is confusing public messaging in the “culture wars” space but, in my opinion, this is just a distraction from the core issue of US control.

The implications are in my opinion dire – there is a clear trajectory aimed at complete dominance over the AI supply chain by a small set of US based actors. With politicians unaware of “deeper” parts of the AI supply chain and value chain, the campaign is currently not meeting much resistance, to a point that the loss may soon become irreversible, perhaps within the next years, if we do not act.

Socio-economically, it means that we will move from a situation where we basically would have to reinvent or reconstruct the entire technology stack in the AI supply chain to do anything meaningful in the AI economic space. That is, we may end up at a technology disadvantage, which is no longer realistically possible to catch up.

Open Source currently plays a crucial role: it is an ecosystem of technology and people which naturally resists centralization and control – since open source, as a technology, is openly available, decentralized, and openly governed. Its very nature is an anathema to the architects of an authoritarian future fueled by AI.

This also means that adversarial actors would have to effectively destroy or marginalize open source as it currently exists, in order to meet their goals of control and centralization – we are indeed seeing these dynamics picking up speed rapidly.

But on the positive side: it also means that democracies need only embrace the technology and its economic ecosystem as it is currently structured, as an operational principle for their AI sector, in order to get ahead quickly and be able to steer the direction that the AI revolution takes.

Q4. Technological sovereignty’ has become a buzzword in policy circles, but what does it actually mean in the context of AI? How does open source software serve as a foundation for genuine technological independence, and what would be at stake if European or other nations don’t invest in this capacity?

FK: Technological sovereignty must mean that we – as democratic societies – can control direction, speed, and goal of the newest technological revolution. Absence of tech sovereignty means that control lies with others, and currently the trajectory is increasingly veering towards a techno-fascist, corrupt imperium led by oligarchs.

By now technological sovereignty is a matter of survival, of entire societies and governance systems, such as democracy itself. That democracies are slipping, or even collapsing, across the world is, in my opinion, a phenomenon that is absolutely not independent of the rise of social media and AI-fueled algorithms.

It is fascinating to behold, that China have rebuilt the entire AI supply chain for itself, starting about 10 years ago, resulting in what in essence looks like total sovereignty in the AI tech sphere. For instance, China has set up completely independent package distribution pathways for Python. They are no longer relying on the Python Software Foundation’s PyPI that the entire rest of the world uses without significant alternative. China is even maintaining fallback copies of the open source packages for core AI technologies, and has built up the requisite communities behind them so they could instantly replace the US technology stack if, say, all US-bound internet connections to China were suddenly severed.

It think the democratic societies of the world must learn from this example. In the same way, resilience and sovereignty must be looked at in the entire tech supply chain and value creation chain. For instance, some unpleasant questions:

• If the Python Software Foundation restricts PyPI, or the US puts export tariffs on “crucial software components” (as currently in the threat to China), would our AI systems still work?
• If Microsoft restricts the use of GitHub, would our software development infrastructure still work?
• Do we have significant and sufficiently professionalized developer communities in Europe to build and maintain key open source AI software component infrastructure if the need arises tomorrow?

These are crass scenarios of course, but I do not think any reasonable politician would anymore dare to say that these are impossible – or even implausible over the next 10 years. And similar but less extreme questions will be asked in partial variants, e.g., increasing soft control being exerted instead of hard cuts.

Beyond resilience, there is capability. If we – as societies – do not grasp the technology stack and the AI value creation chain, how could we hope to innovate on it? Already now, most successful AI startups are located in the US and in China – we have fallen behind and our competitors are using their advantage brutally.

Open source communities are perhaps the most powerful tool that we have in ensuring resilience and capability – knowledge is shared, and development is decentralized. There is strength in numbers and collaboration across the world. And, as the example of openAI shows: it is very much possible to come out of nowhere, to leverage open technologies to reach a worldwide leadership position quickly.

Q5. Looking ahead, how do you see the connection between open source AI development and the future of democratic governance? Beyond the economic and technical arguments, what’s the democratic case for supporting open source AI as a matter of public interest?

FK: This is a very interesting question.

It is probably undisputed that the AI revolution will rapidly transform our societies in hard to predict ways, over the next years – so technology governance will be key, and it will be a societal responsibility. As democratic societies we also have different goals and viewpoints than the typical Silicon Valley company.

Societal governance is inextricably linked to technology governance, hence.

As mentioned before, the nature of open source communities is decentral and openly governed. Open source governance forms a natural system of technological governance that extends democratic governance to AI technologies.

Contrast this with typical US corporate governance for AI, which follows the autocratic or oligarchic template instead.

Follow me on a thought experiment: imagine if in 2010, the EU had funded a successful project to build a European open source social media platform, through decentralized communities of open source developers, with governance by representatives of the people, with servers in Europe, and all algorithms being available to public scrutiny.

How different would the world now be?

Or, imagine, if, in the 2000s a community of open source enthusiasts would have collaborated to build an open system to collect and share all knowledge of the world? Well, ok, that is what happened so far – it is called Wikipedia. But imagine, if Jimmy Wales would then have moved to monetize Wikipedia, e.g., by selling product placement and later hidden political ads in Wikipedia articles.

These two examples show that going down one route or another may not be as inevitable as we may think – and politicians as well as individuals can help with creating the right conditions.

Now we are faced with a similar storm, a technology shift uprooting the world even more rapidly and fundamentally than social media did back then.

We have to set the course quickly and decisively.

To conclude, I would like to outline where the journey could go – a better world, or one that is even worse than it already is. To prepare the canvas, we need to stop and realize:

Software, by nature, is post-scarcity technology.

Software can be copied indefinitely. When publicly managed, it can simply be copied and given to anyone and everyone without a significant additional cost – after production and design, and given ubiquitous hardware, of course.

This is different from, say, apples, or furniture, that cannot be “copied” once produced and designed.

Since AI is software, AI is also a post-scarcity technology (if hardware is ubiquitous).

Key point number one: the core technology for the current technological revolution – AI – is post-scarcity.

Many companies now add licensing and subscription models, and so on, to introduce artificial scarcity and map the technology on classical commercial distribution models. But this does not change the fact that AI tech is infinitely replicable and shareable.

Coming back to the numbers from the very beginning: 8.8 trillion dollar worldwide sector demand marks a clear and accelerating trend: the worldwide economic value of the open source “post-scarcity” sector. To repeat: more than the GDP of any EU state, including sums attributable to actual furniture or car parts production.

Also, AI software components in turn were not built on Windows, but on Unix based operating systems, also largely open source. Observe: the “post-scarcity” sector has been already growing and self-reinforcing for decades by now.

Key point number two is governance. As already mentioned, open source communities are decentralized and typically open governed, along democratic ideals. It is only a small – but powerful – set of bad actors that try to force the sector back to scarcity economics, back to autocratic corporate control.

On a side note related to governance, to avoid misunderstandings: I want to emphasize a strict difference between post-scarcity technology and ideas of (Marxist) Communism. in Communism, the proletariat jointly owns (in theory at least) a scarce resource, e.g., furniture factories. In contrast, in a full post-scarcity economy the ownership of furniture factories becomes irrelevant, because everyone owns a Star Trek style replicator that can produce furniture (or car parts).

When ownership becomes irrelevant, only governance matters, e.g., who can use the replicators for what – is it totalitarian (like real world Communism), or democratic.

So, suddenly, direct control becomes crucial as a means to exert power, rather than control of a scarce resource. Closely related is the concept of public ownership of goods or services, such as (in many countries) water, public transport, or healthcare.

Understanding AI software as a post-scarcity technology highlights that tech companies are incentivized to exert political control, simply since the technology becomes increasingly uncontrollable, “post-scarce”. Artificial scarification is a logical reaction; as is undermining of democratic governance structures – around tech, but also more widely.

For democratic societies to survive and thrive, this suggests a recipe: support, integrate, and embrace the “post-scarcity sector” in its current form – already governed by democratic principles of technology governance.

In the long-term outcome, this decision may then indeed mark the difference between social media that unites and social media that divides; between knowledge that belongs to all, and knowledge controlled by a few; or, between Star Trek replicators and 1984 book burnings.

…………………………………………

Dr. Franz Kiraly is the director of the German Center for Open Source AI, Germany’s largest non-profit for open source AI software (gcos.ai). He is also the founder of the widely used software package “sktime” for AI based forecasting, and a contributor to numerous open source projects. Previously, Dr Kiraly has held principal roles in data/AI departments in industry, academic faculty positions at UCL and ELTE, and a fellowship at the Alan Turing Institute.