Creative opportunities, independence and agency in the age of Artificial Intelligence
By Alex Gakuru | Tech Expert | Founder, ITS Academy (Integrated Technology & Society Academy)
I have spent decades promoting internet in Kenya: from advocating for the laying of the cables, universal access, affordability, provider accountability and consumer protection. So when I say the worst thing that ever happened to us might be connectivity, I do not say it lightly. I said it in João Pessoa, Brazil at an Internet Governance Forum, and I meant it: we moved from digital plantations to something worse: the algorithmic harvesting of our minds, our culture, our creativity, and our data, sold back to us as subscriptions. The question now is not whether to accept or reject artificial intelligence. The question is whether we have agency over it. Agency means we command the tools. Without agency, we are stochastic parrots, a phrase coined by Bender, Gebru, McMillan-Major and Shmitchell (ACM FAccT, 2021) to describe how AI models recombine internet text without understanding, over-representing hegemonic viewpoints while excluding smaller players. That is the trap: mimicking Silicon Valley and calling ourselves ‘Silicon Savannah’. Who told us that is what we want? I am not in that savannah. I am in the slopes of the mountain. If you are in Nandi, be innovative hills. If you are in Ukambani, be the creative plains and if in Turkana, be the generative cradle. Name your own terrain.
The binary debate (opt in or opt out of training models) misses the point. Our knowledge, traditions, and creative works have already been vacuumed into foreign models. The Creative Economy Outlook 2024 (UNCTAD, 2024) shows global creative services exports hit US$1.4 trillion in 2022, nearly double creative goods exports, yet Africa captured only US$4 billion against Europe's US$720 billion and Asia's US$359 billion (UNCTAD, 2025). The top ten exporters take 70 per cent of creative goods and 69 per cent of services. Meanwhile, Kenyan workers labelled content for OpenAI through Sama at US$1.32 to US$2.00 per hour while the outsourcing firm billed US$12.50 (TIME, 2023). That is the extractive end of the chain. Agency means refusing to be a data source for someone else's empire. It means building tools that serve our people, our languages, our realities, and keeping the value here.
The IMF estimates 40 per cent of global employment is exposed to AI: 60 per cent in advanced economies, 40 per cent in emerging markets, 26 per cent in low-income countries (IMF, 2024). The World Economic Forum projects 170 million new jobs and 92 million displaced by 2030, a net gain of 78 million, but nearly 40 per cent of existing skills will change (WEF, 2025). Creative thinking, resilience, and agility remain critical alongside AI and data skills. In Kenya, the creative economy already contributes 5.3 per cent of GDP and technology 7 per cent (Creatives Garage with Mozilla Foundation, 2024). A study of 130 Kenyan creatives found job displacement and copyright as the biggest AI threats, but improved work efficiency as the leading opportunity. The old model is dead: wait for a job, learn a software package, mimic a foreign template. Doomed are students learning software packages instead of computing fundamentals. Institutions teaching from yellowed notes are not in tandem with reality. We must stop aping. We must start solving.
Stop chasing volume. Stop writing for content mills that AI has already swallowed. Start identifying a problem your neighbour has, a filmmaker who needs a subtitle tool that understands Sheng, a musician who wants stems separated offline, a podcaster who needs transcription in Kikuyu. Ask your local AI to build it. Sell the solution and, more importantly, the service. You earn income better than platform deliveries, better than online writing gigs, better than waiting for a job that will not appear. This is not theory. With free, open-source tools you can create mobile apps for your own use that serve your need, they do not need to go online or be uploaded to app stores. No ads, no granting invasive phone permissions such as contacts, call logs or location data; no SMS access; no data harvesting. The whole industry of mining your phone just to use a text editor collapses when you build the editor yourself. Regulators meant to protect your data are hopeless against the scale of foreign-orchestrated extraction. So protect it yourself.
The first step in breaking free is establishing models, tools, and software that are open-weight, open-source, and run locally on your computer. Four reasons. First, privacy: your chats, your confidential medical or legal information, stay on your machine. They are not hosted online where the US CLOUD Act (2018) compels disclosure to US authorities regardless of where the data sits (US Congressional Research Service, 2018). Foreign jurisdictions have passed laws forcing companies to save your chats for later use in courts and processes you never consented to.
Second, resilience: when the internet is down, fibre cut, ISP failure, power outage, your system still runs. Third, cost: the median price of an entry-level 2GB mobile broadband basket in Africa is 4.2 per cent of GNI per capita versus 0.3 per cent in Europe; in low-income economies it hits 7.4 per cent (ITU, 2024). In Kenya only about 40 per cent of the population can buy that basket for under 2 per cent of actual income. Every token sent online burns money.
Fourth, sovereignty: you are not subject to random price changes, model degradation, or subscription walls. The EU AI Act (2025) exempts free, open-source general-purpose AI models with public weights from technical documentation and authorised-representative duties, though not from copyright policy and training-data summary obligations (European Commission, 2025). That exemption exists because lawmakers recognise open models as a public good; use it.
You do not need the latest GPU costing millions of shillings. Start with the hardware you have. DeepSeek-R1 weights were released under the MIT licence in January 2025 with six distilled models from 1.5B to 70B parameters that run on laptops; code and models may be commercialised freely (DeepSeek, 2025). Google DeepMind's Gemma 3n is designed to run locally on phones, tablets, and laptops: the E4B variant has 8 billion raw parameters but operates with a 4 GB memory footprint (as little as 3 GB on GPU), is privacy-first, offline-ready, and supports 140-plus languages (Google DeepMind, 2025). Chinese researchers, constrained by compute access, are doing amazing things with new AI architectures, re-engineering the engine instead of buying the biggest one. Learn from them. Ask what you can achieve with the hardware you have, not what you cannot because you lack the latest Nvidia card. That inadequacy is a marketing tactic making you an unsuspecting sales agent for trillion-dollar AI-chip corporations. Edge devices, mobile phones, modest laptops, this is where solutions come from. This is where the opportunities are.
We should not adopt names or branding that are not ours. Why not innovative slopes, creative sungura, generative hills? The history of Silicon Valley is not the pilgrimage you think it is. When you go paying pilgrimage there, you fool yourself.
For the better part of the last 100 years, this region of Northern California has had a hugely disproportionate impact on how modern humans communicate, interact, work and live. Cities around the world, from Tel Aviv to Tallinn and from Bangalore to London, have attempted to set up copycat innovation hubs by replicating the Valley’s DNA, with mixed success: commentators argue that a clone with the same scale of power, productivity and influence is not possible (History Cooperative, viewed 19 October 2026).
AI lobbying in the US surged to 648 companies in 2024 from 458 in 2023; OpenAI's federal spend rose from US$260,000 to US$2.99 million; Anthropic hit US$3.13 million; eight largest firms spent US$36 million in the first half of 2025 alone (OpenSecrets via TechCrunch, Axios, MIT Technology Review, 2025). They are well funded to maintain the status quo.
Kenya has been online for thirty years. Where is the large globally successful app locally developed? Why are we the noisiest on TikTok and Twitter yet hired only for menial data cleansing at peanuts? Are we too poor, too desperate, too misinformed? Or structurally condemned to creative surrender anchored on innovation harvesting? The backlash is at our doorstep. We must stand firm strongly asserting our prosperity interests. Create your innovation around your environment. It is more real than a foreign dream designed to fulfill others.
This is not a manifesto. It is a method. Install an open-weight model on your laptop today. Identify one person with one problem. Build the solution. Sell the service. Keep the data. Keep the revenue. Keep the dignity. Teach the next creative to do the same. That is how we move from stochastic parrots to sovereign creators. That is how we turn the creative economy from US$4 billion to its rightful share, closer to Asia's US$359 billion. That is how we stop being the source code for someone else's empire and start writing our own. The tools are free. The models are open. The evidence is verified. The only thing missing is your decision to begin. Now you know. You cannot unknow.