1 of 26

Planning (AI) innovation

Dr. Cecilia Rikap

UCL’s Institute for Innovation and Public Purpose; CONICET; COSTECH, Université de Technologie de Compiègne

c.rikap@ucl.ac.uk @CeciliaRikap

2 of 26

The innovation myth

Throw things at the wall and see what sticks

Move fast and break things

From the garage to the Nasdaq

Serendipity is something that is planned, not at the dictating level of the particular item but in the sense of putting great people in contact with great resources and identify needs and let them go, that is how you plan for serendipity and get results. (IBM former VP)

1 and 2 are innovation as anti-planning (serendipity) // 3 is everyone can have a successful company (also associated with individual merits) // 4 is technology progresses by itself (no one is pushing the Accelerator pedal) // 5. assumes that innovators are rewarded for their innovation

La iteración (con datos y AI) resulta en minor innovations. Hay convencimiento de que los senior managers vinculados a R&D y MKT crean futuros más allá de lo predecible
El fracaso como prueba de que siguen siendo innovadoras // pero en verdad la inestabilidad no promueve la innovación, para ser disruptivo se necesita estabilidad en el sistema (es más fácil fracasar si sos Google o Amazon, tenés red). Silicon valley es la red de innovación más estable que existe

- P(doom) Blaming it all on the technology is not only useful for diverting attention from those developing it. AI scientists and engineers working within the AI frontier innovation systems controlled by these giant corporations will feel less responsible

La innovación es una relación social de producción. No hace falta ser el primero ni hacer algo tecnológicamente nuevo (aunque el desarrollo tecno y el claim de ser el primero se usen como señales (señuelos) de la innovación). La innovación es una relación de unicidad. La sociedad tiene que percibir un producto como único para considerarlo una innovación, pero no único como la obra de arte porque tiene que ser industrialmente reproducible. Quien la ofrece es visto como único. Hay una sola Google.

“es una zapatilla Nike” / y cuando no nos gusta decimos que la propia Nike no hizo una zapatilla Nike. Los fracasos no son “verdaderos” Nike

Pero estos no son los creadores sino los apropiadores. “By blurring the distinction between creative co-production and the appropriation of outputs, the true creator, the innovator, has been erased in plain sight as intellectual monopolies attribute authorship”

3 of 26

The US NSS Planning S&T

US National Security State: US DoD + DARPA + NIH + CIA (Weiss, 2014)

Innovation for military supremacy

Dual-use strategy: military + commercial (+control)

DARPA

Building an S&T network for computing science Internet + connecting people, organizations, ideas

4 of 26

The demise of the US NSS

Since the end of the Cold War 🡪 vacuum in planning

This space was conquered by Big Tech

AI is dual-use, but Big Tech do not depend on military contracts

They do rely on the US DoD and military as allies

Military expenditure (% of GDP) – Source: World Bank data

5 of 26

Planning AI

Internal R&D

Amazon Grand Challenges Division

Google’s Manhattan project: DeepMind

Microsoft’s bet on OpenAI + Centralized AI team

Knowledge and innovation network

Regulation & narratives

Source: Scopus Top 14 AI Conferences (2018-2020)

Source: WoS (2014-2019)

6 of 26

Controlling�Open Source

Keep employees happy
Access reusable code for private software
Get users’ feedback in advance
Profit from developers’ free work
Offer adjacent or complementary services
Create and control an ecosystem
Become the norm and standards

open sourcing improves our models, and because there's still significant work to turn our models into products, because there will be other open source models available anyway, (…) more specifically, there are several strategic benefits. (…) more compute efficient to operate due to all the ongoing feedback, scrutiny, and development from the community. (…). Second, open source software often becomes an industry standard, (…) on building with our stack, that then becomes easier to integrate new innovations into our products. That’s subtle, but the ability to learn and improve quickly is a huge advantage and being an industry standard enables that. Third, open source is hugely popular with developers and researchers. (…) so this helps us recruit the best people at Meta, which is a very big deal for leading in any new technology area. And again, we typically have unique data and build unique product integrations anyway, so providing infrastructure like Llama as open source doesn't reduce our main advantages.

And my first point here is that Big Tech control and profit from the open source space. There are many reasons but, to me, the most relevant ones are (click) profit from developer’s free work and you have here an indicator that compares people working for each company registered on GitHub and the total number of contributors to some of these companies most popular open-sourced projects. But also and particularly important for AI (click and click), open sourcing is a strategy used to creat and control what they describe as ecosystems that basically add many co-produced pieces/modules to the larger corporate innovation system of these companies and make their solutions the norms and standards. And also in relation to AI code and models…

So their capacity to monopolize knowledge even extends and conquers initiaves conceived to counterbalance their power

7 of 26

Source: Crunchbase February 2024.

Controlling (AI) start-ups

Double dip: Direct line to the CEO + technological lock-in 🡪 OpenAI

Cloud Credits: “compute vouchers”

8 of 26

The public cloud

GAM: 65% of the global market (+ cables)

Supermarket

50,832 services in AWS (Apr-2025)

“Partners”: users & sellers

Digital (control) technologies are produced, exchanged & consumed in their clouds

“As AI gets more efficient and accessible, we will see its use skyrocket, turning it into a commodity we just can’t get enough of.” (CEO Microsoft, LinkedIn post Jan -25)

All these services are designed following the rules –the computing protocols- of cloud giants and to use the services employees need to get GAM certifications. GAM thus shape how the technology evolves. They define the categories of services, how the APIs are designed to interconnect services and AWS even approves (or not) price increases in its marketplace.

In the cloud, software (including AI) and data are offered as black box services and even storage in an outsource datacentres limits learning opportunities because you never acquire the machine in your factory and can do reverse engineering and so on.

But the main problem is not how they dictate the production of this or that service from formally independent but de facto dependent companies. The main problem is their control of the whole ecosystem through their clouds. Adding all the pieces together, because they control the cloud…

9 of 26

The public cloud

GAM: 65% of the global market (+ cables)

Supermarket

50,832 services in AWS (Apr-2025)

“Partners”: users & sellers

Digital (control) technologies are produced, exchanged & consumed in their clouds

Black boxes: use without access

🡪 Public & corporate dependences

Amazon has a lot of SaaS and tries to increase the stickiness of the system. (…) It is not in your control to switch to other companies easily. (…) If you want to move from one cloud provider to the other, it takes months of planning

(AWS senior software engineer).

All these services are designed following the rules –the computing protocols- of cloud giants and to use the services employees need to get GAM certifications. GAM thus shape how the technology evolves. They define the categories of services, how the APIs are designed to interconnect services and AWS even approves (or not) price increases in its marketplace.

In the cloud, software (including AI) and data are offered as black box services and even storage in an outsource datacentres limits learning opportunities because you never acquire the machine in your factory and can do reverse engineering and so on.

But the main problem is not how they dictate the production of this or that service from formally independent but de facto dependent companies. The main problem is their control of the whole ecosystem through their clouds. Adding all the pieces together, because they control the cloud…

10 of 26

Control Technologies

Governing requires information

Citizens/workers

Organizations & networks

GAM sells to other leading corporations & states (control) technology for ruling

GAM still rule by controlling the digital sphere

You see your business; you have data from everywhere and even on paper. How you rationalize all the data is what makes Microsoft stay at the top of the trend and function efficiently with AI. (Microsoft operations manager).

In the end of the day, it is just about predictions, inferences, so the whole point of supply chain management is about forecasting, how many units of an item will be demanded, how much data you can use and how much of the accuracy game it gives you. Machine Learning is in a big part used for forecasting. (Amazon, research director)

We ”see” what they want us to see and how they want us to see it.

What they have monopolized are the technologies to measure, to quantify. And the effects are directly connected to society not only because those measuring rule and now those selling the quantification technologies are a handfull of companies but also because of their epistemic totalitarianism

While ir extendes the ruling power of other IMs over their respective spheres of control, at the same time cements a power relation among leading corporations.

And a related effect is that the same is happening with states that were supposed to be sovereign –yet we know that this was not valide everywhere, because most of the peripheral countries’ states were already technologically dependent on tech from the North. Anyway, this is a form of further subordination for these states to a handfull of corporations and the same can be said of core states like those in Europe.

11 of 26

”

AI needs to be regulated in a way that balances innovation and potential harms.

We do not have to wait for regulation to have standards or adopt the misstandard, I call it the start of any self regulation. Then, on top of that, if we talk about regulation, maybe we can unpack it from the application domain, because after all context in which something is being applied, in education, healthcare, retail and we can have the regulatory frameworks that already exist

Innovation as progress 🡪 regulating only the harmful/risky outputs

Present themselves as the guardrails of society in its digital journey

Blame China as the thread

Corporate planning of AI regulation

12 of 26

Before OpenAI released ChatGPT I didn’t see Chinese companies doing this but the night after that everyone started talking on doing it for China and a lot of companies and funding for this and as of today have developed very similar or even better models, with similar ideas. (…) In China, it would have been much more difficult to do something as unprofitable as OpenAI, but after you see is doable, then the funders and everyone is there but it is hard to pioneer. (Megvii AI scientist)

US tech companies in Silicon Valley are more innovative and create new things. They are more entrepreneurial and make 0 to 1, create things from nothing to something. The Chinese companies can make large size based on things that were already found, going from 1 to 100 in size, speed. Other than the AI technologies, there are also smart vehicles and other tech products that also have similar trends. The US is more innovative and creates something and China further develops models. (Netease engineer)

“I think for now everyone is seeing and waiting, and we want more people to test it to make the model better, but the critical thing of the AI models is how can you actually commercialize it or whether you want to make it open source and create a community for AI and everybody can contribute to it. (…) I am not sure because I don’t know the company decision whether we want to set up a standard for AI like GitHub community where Google or Azure are trying to set up a standard too or not.” (Alibaba International Cloud)

An example of this strategy

13 of 26

And Chinese Big Tech?

Source: Scopus Top 14 AI Conferences (2018-2020)

My thesis is that for Chinese capital being a fast follower/imitation w scale of US frontier innovations is the best strategy and will not move from there

So it is more like Chinese concentrated capital is happy to arrive second (they are risk averse and fast in copying and scaling and because of the size of China plus the delay in expansion of US capital to certain parts of Asia, their market is super big. And this generates clashes of power with the Chinese state)

Proof 1: a progressively more detached system. Communicating channels now mostly depend on Chinese AI scientists going back to work for Chinese companies + Microsoft + the few R&D that Chinese tech giants still do in the US BUT smaller companies have separated the business (Pony.ai)

We have in this presentation 4 players + the AI adopters

14 of 26

AI in China: Imitation

Geopolitical constraints

Disciplinary cases: Huawei, TikTok

Limited access to processors

Internal protected market (+DSR)

Mostly Chinese data 🡪 Chinese models

Favours US Cloud giants

More AI uptake

More competition for AI start-ups

Regulatory shield

15 of 26

Corporate planning of AI regulation

Further favoured by the Trump administration

“Let us be clear: when it comes to AI, America is a global leader. It is American companies that lead the world in AI innovation. It is America that can catalyse global action and build global consensus in a way that no other country can.” Kamala Harris

A more secure & resilient supply chain

Revival of industrial policy: chips act

US Executive AI order: “Safe, Secure, and Trustworthy Development and Use of AI”

A statement on US technological leadership underpinning geopolitical power (vs China)
No dismantling of IM: data + models remain secret & no regulation on users’ data
Focus: harmful AI uses, applications

16 of 26

Intangibles extractivism: S&T from the peripheries monetized by a few giants from the core.

Epistemic totalitarianism: imposing topics, methods and ultimately a way of thinking.

Summing up, planning innovation…

Knowledge (S&T & experiential)

Data / Information

Narratives

Social relation that enables value capture

Intangibles

17 of 26

Thanks

Dr. Cecilia Rikap

(University College London; CONICET; COSTECH, Université de Technologie de Compiègne)

c.rikap@ucl.ac.uk @CeciliaRikap

Capitalism shall be seen not only as a society based on depriving the many from the material means of production but also depriving the many of the knowledge/intellectual capacities to control/organize/plan a production process. Intellectual monopolization is the latest step in this conflict. Democratic eco-socialist planning requires democratic control technologies.

18 of 26

The AI value chain

AI Chip Design

Foundry

Processor

AI Foundation Models

AI-Powered Apps

Data Centres

Data

AI Talent

Micro Tasks

Cloud Network power

Chokepoints (lock-ins)

Panopticon

Agenda setting

They control most of the AI –and other digital tech- value chain. And here you have a very simplified depiction.

And network power abuses are to be found not in rising prices of their and others’ services on the cloud –which they sometimes do it actually- but in the way they control the whole ecosystem, which can be reduced to 3 forms of cloud network power:

Chokepoint because everyone that wants to sell a computing service –from software to consulting services- must do it on their clouds. Even DeepSeek or Meta.

Panopticon: they oversee the global economy from their clouds

Agenda setting: the AI that we get is the AI that evolves hand in hand with their clouds and they push through their cloud for massive AI uptake for everything.

If we zoom into this exercise of power, there are intertwined / related networks, such as the AI Research network as well as the open source community:

19 of 26

20 of 26

The Rulers

Cybernetics: processing information and using scientific prediction for advancing control, governing and steering

Second-order cybernetics: hierarchical multilevel control

Each controlling system is controlled by another system

21 of 26

Intangibles extractivism: S&T from the peripheries monetized by a few giants from the core.

Epistemic totalitarianism: imposing topics, methods and ultimately a way of thinking.

States

Hospitals

Subcontractors (value chain)

Summing up, planning innovation…

22 of 26

Deskilling creative labour

Fastest adoption ever: individual users AND organizations

McKinsey: “3/4 of genAI’s expected value will be in R&D, software engineering, marketing & customer service”

Creative labour 🡪 fact checkers

GenIA: shapes how we think and what creative labour is

23 of 26

States

Hospitals

Subcontractors (value chain)

Twin extractivism: intangibles & nature

Intangibles extractivism & epistemic totalitarianism

24 of 26

Digital’s twin extractivism

Massive consumption of electricity

Deepened by AI

Training & use

+ Water consumption

In 2022, Microsoft +34% & Google’s +22%

25 of 26

Green-tech

IA-solutionism

Influencing how the ecological crisis is being measured

Deepening states’ dependence on BT

We were an official partner at COP-27 in 2022, where we participated in over 50 events and moments throughout the conference with public sector leaders from the United States, Europe, Africa, the Middle East, and Asia (…) to showcase the role that the technology sector can play in enabling climate mitigation and adaptation. (Google, 2023).

26 of 26

The AI research agenda (organizations & topics)

(2018-2020)

Source: Top AI Conferences retrieved from Scopus

Topics directly linked to US & Chinese Big Tech

Tsinghua University 🡪 Reinforcement Learning

Pekin University 🡪 Large Language Models

Division of AI labour in China’s NIS? Or is it that they are producing frontier AI with (for) Microsoft?