Artificial Intelligence (AI) is no longer a futuristic concept confined to research labs and high-tech corporations. It’s now woven into the fabric of our everyday lives, shaping how we work, communicate, and solve complex problems. Yet, as AI’s influence grows, so do the disparities in who has access to its transformative potential. “Democratizing AI” is about breaking down barriers—technical, economic, and social—so that everyone, not just the privileged few, can harness its power. This article explores how open-source tools, accessible education, and ethical frameworks are driving the movement to make AI an inclusive force for innovation, equity, and global progress; and introduces a new project Spin I undertook to address certain gaps in the ecosystem for developing a democratic future in AI.
1. The current state of AI
In its early days, AI was more of a novelty than a tool. At its early best, generating coherent language responses, but entirely lacking the intelligence and performance required to tackle real-world tasks and challenges. Over time, several breakthroughs by technology companies, researchers and startups in the field have made these tools significantly more accessible and practical. However, the cutting-edge of AI research still remains in the hands of a few with the resources to invest in supercomputing power, specialized talent, and proprietary data. For the majority of individuals and organizations, the barriers to entry are still daunting and AI innovation can feel like a distant opportunity , a guarded fortress, threatening their jobs, businesses and livelihoods. Without substantial funding or access to influential networks, the prospects of gaining a foothold in this transformative technology field remain difficult.
The financial hurdles despite the hype and propaganda, remain immense. Building AI systems requires access to cutting-edge hardware at large scales, which only the wealthiest of players can afford. Training and running inference at scale for even a single AI model can demand computational power costing hundreds of thousands to millions of dollars. Additionally, the intricate nature of AI creates a steep expertise gap, accessible only to a small cadre of highly skilled engineers and researchers. For startups, nonprofits, and innovators in underfunded institutions, AI was—and often still is—a distant dream, a transformative technology locked behind gates they cannot open, with only glimpses of hope given with a few freebies designed to turn people into consumers for the technology than participants in the fruits of labor, investment and AI advancement sitting on horizon.
2. Why is democratization important?
Research in AI is generally committed to ensuring fairness, equity, and innovation throughout the world. There are however challenges posed by the social, political, economic and legal struggles on how the commercialization of AI will affect jobs, social harmony, psychological and financial well being of communities across the planet. Such profound effects call for democratic participation by all stakeholders.
There are four aspects to the democratization and commercialization of AI: access to Data, Compute, IP (technical knowhow) and Finance.
On the Data front: Proprietary AI models often harness vast amounts of private and public data without explicit user consent. The creators and owners of this data rarely receive equitable recognition or share in the revenue and growth generated by these models, raising significant ethical, legal and financial concerns. Democratizing ownership over AI and data assets is an important step towards empowering individuals and communities to take ownership of their contributions, ensuring that the benefits of AI are distributed more fairly across society. Companies like Google, Meta, Microsoft and OpenAI have already gotten into hot waters with organizations and government bodies across the world and examples of such legal battles can be seen in New York times’ lawsuit against OpenAI [1, 2], objecting to its use of NY Times’ content in training its models, or Australia’s legal imposition on Google and Meta to pay Australian news agencies a monetary compensation when showing such content as part of its search results [3] (which shows how the battle over data and content monetization has been playing out even before modern LLMs were on the scene). Such legal and social battles show the importance of putting in place monetary and legal mechanisms for sharing the data under which AI advances are made.
[1] https://harvardlawreview.org/blog/2024/04/nyt-v-openai-the-timess-about-face/
[2] https://www.theguardian.com/media/2023/dec/27/new-york-times-openai-microsoft-lawsuit
[3] https://www.france24.com/en/live-news/20241212-australia-to-force-tech-titans-to-pay-for-news
On the Compute front: The concentration of computing resources in the hands of a few well-funded entities stifles human ingenuity and creates perverse incentives to displace the labor developing AI assets with the same AI assets in the future. Many talented innovators are unable to design, create, or own new AI models simply because they lack access to the computational infrastructure required to develop them. This disparity perpetuates a system of “haves” and “have-nots,” where only the privileged few can actively participate in shaping the future of AI [4, 5, 6]. Democratization is required to bridge this gap by providing access to resources and model outcomes, leveling the playing field for individuals and organizations across the globe. AI assets further have an outsized effect on productivity and reduction in labor force and job requirements. As a result, creating financial and legal shareholding mechanisms to share in the fruits of labor, financing, compute, data, and IP development that leads to AI advances is vital in ensuring fair access to AI generated capital and revenues in the future.
[4] https://venturebeat.com/ai/ai-research-finds-a-compute-divide-concentrates-power-and-accelerates-inequality-in-the-era-of-deep-learning/
[5] https://www.nature.com/articles/d41586-024-03792-6
[6] https://www.signalfire.com/blog/ai-compute-shortage
On the Finance front: Access to Finance, underpins the ability of individuals and organizations to invest in the development of AI and become shareholders in the future of AI. While on the input side, financial resources are required to access compute, data and talent pools for developing and running AI, a more important financial consideration is on the output side of the equation. AI models tend to displace human labor effectively, creating opportunities for individuals and organization to start and run their own businesses and generate revenue from their operations. However, this privilege if restricted by financial constraints to only a few, comes at the cost of lost jobs and a skewing in power dynamics toward oligarchic business structures due to a disparity between those with access to finance to develop and run AI based businesses vs those without finances, depending on skill, time and labor in a shrinking pool of opportunities to claim their piece of the pie as more automation kicks in. What is important is how shareholding and revenue shares in this increasingly automated world gets distributed between those who finance such development vs those who invest with time, skill and efforts in developing such businesses. Furthermore, how financial investment opportunities and revenue generation opportunities are provided on a persistent basis in the operation of such AI-based businesses beyond the initial wins and compensation, is an important consideration to avoid future conflicts from arising from such disparities.
True democratization must extend beyond access to tools and infrastructure to the sphere of finance, equity and income opportunities. It requires a commitment to ethical training, the use of diverse datasets, and the establishment of equitable governance structures. By addressing biases in AI systems and ensuring accountability in their use, we create technology that is not only accessible but also responsible and inclusive. Democratizing AI is not just a technical goal; it is a societal imperative to ensure that AI serves as a force for shared progress rather than division and conflict.
3. What is being done to address democratization in AI?
A lot of opensource AI models and agentic frameworks are now available through platforms like Ollama, HuggingFace, Letta, LangChain, to name a few (you can refer to Letta’s summary of AI agent stacks [7] to get a brief idea of what the open and closed API technology stacks look like). In short, a lot of open source and open weight AI models are already shared online that creators and companies can use as a starting point and develop their products on top as long as they have the financial, computational, data, and technical resources to execute on a business plan. That is where the catch lies however, there are limitations to the availability of the above four resources, and each one of these can cause an organization to falter or give up on their AI.
[7] https://www.letta.com/blog/ai-agents-stack
On the subject of data and IP
As of the writing of this article, there is still no standardized legal and financial mechanism in place to share and monetize data by individuals and companies as data suppliers for the development of AI at scale. Yann LeCun, VP and Chief AI Scientist at Meta recently posted on the subject of data and urged institutions, governments and communities to make data freely available for training under the condition that such models be made freely available with weights and inference code. Such a data sharing model offers the implicit deal of getting more representative AI models to the public while allowing companies to gain access to representative data sets, making an implicit social contract for social good while allowing companies to retain certain IP and business interests on the training and modification of such models.
On the other extreme end is the usage of public data by companies for the training of closed API models, providing paid access to the public to the resulting AI models but no ability to host or modify or run the trained models.
A mid-path that has often been proposed but as yet not implemented is in providing data for training with certain compensation through shareholding mechanisms within the model. Examples of such proposals can be found in [8, 9].
[8] https://www.forbes.com/councils/forbestechcouncil/2024/06/26/20-tools-and-strategies-for-safe-and-efficient-ai-system-data-sharing/
[9] Sharing Data With Shared Benefits: Artificial Intelligence Perspective, National Library of Medicine, https://pubmed.ncbi.nlm.nih.gov/37642995/
[10] Data sharing in the age of deep learning, https://www.nature.com/articles/s41587-023-01770-3
A technical hurdle here is in implementing zero-trust mechanisms to enforce the contracts under which the data is to be shared and enforce access to the AI outcomes as dictated by the terms of a contract. Web3 smart contracts provide a natural fit for implementing such data sharing mechanisms, however, a solution has to be customized keeping in mind the privacy concerns and without sacrificing legitimate competitive advantages [10] that companies and stakeholders seek within certain application domains through privatized data and AI model environments.
As a first pass solution, I address the simpler paradigm of data sharing that Yann LeCun refers to in his post, where data sharing in the context of social good is targeted. With a Web3, zero-trust mechanism, I add an additional layer to this proposal, where people can be fairly compensated through NFT tokens for their data contribution to the training of a model and become financial participants in the process of AI based investment and growth.
On the subject of compute
As highlighted above, availability and affordability of compute resources is often a limiting factor for individuals and organizations. Often high end GPUs are unaffordable or locked in through contracts and thus unavailable in the open market. There is however an abundance of underutilized consumer grade GPUs in the hands of individuals and organizations. Further, consumer grade GPUs like the RTX3090/4090 have already been shown to provide comparable performance to high-end GPUs in training and inference with mid-sized models like Llama3 and above (with AI model performance comparable to OpenAI GPT4 or higher). This opens up an opportunity to offload some inference and training costs to in-house hardware at a cheaper rate than commercially available GPUs in the cloud. This also allows individuals and organizations to become hosts for AI model training and inference resources to provide democratized access to their peers within the organization or across the internet, creating an AirBnB-like setup for AI infrastructure.
On the subject of finance
On the finance side, cloud service providers like Google offer certain free cloud credits to startups developing AI. There are also other Cloud Service Providers offering high end GPU servers for rent at competitive prices (often still too expensive for individuals, and smaller companies). The problem is somewhat addressed through private equity investments into startups building AI (creating a boost to solving financial problems for startups but still leaving out a lot of talent pool out of reach of the investors and vice versa). Spin provides an alternative means for startups and individuals to acquire compute resources at a sponsored or more affordable rate from cloud providers, institutions and peers across the internet.
Beyond compute, the problem of financing data acquisition also persists and NFT/smart contract based data sharing provides an alternate means to connect data to AI systems for training and RAG based applications. On the output side of the value generation through AI, Spin also provide the resource providers (hosts for data – like artists, writers, creators, and compute – like cloud service providers, enterprises and individual hosts), a pathway to monetize their resources and become direct, active and persistent participants and beneficiaries in the AI value chain, providing a more democratic framework for how value gets distributed in an increasingly automated world.
4. How can you help democratize AI?
There are three steps essential to democratizing AI:
1. Enforcement of data and IP laws within the AI ecosystem
- Providing shared access to AI training and inference infrastructure
- Providing shared access to finance and revenue streams generated by AI opportunities
1. Enforce data and IP laws
As individuals and organizations, we need to start by protecting our rights against misuse of data in AI training and AI generated (inference) outputs. This requires significant work towards establishing legal and technical mechanisms for enforcing data protection and detecting copyright infringements by AI. The problem is a challenging one, both for copyright enforcers and for AI developers and companies who may inadvertently infringe on copyrights without realizing they have done so, given the massive amounts of data that an AI model churns through. If you are a writer, a journalist, an artist, a lawyer, a stakeholder in the legal and technical challenges in enforcing data protection and copyrights for yourself or your organization, then seek out solutions to enforcing such mechanisms.
If you are a company or individual developing AI, find out more about how you can legally access public and private data without infringing the owners rights.
2. Become compute hosts for shared AI training and inference
Large language and other AI models require a massive amount of compute for both training and inference. You can and should become a participant and an investor in this compute ecosystem. Currently millions of individuals and startups around the word are bottlenecked by the amount of compute available to them to run and experiment with AI. You can become a host for such AI using just the consumer grade GPU’s you have at home and the open source AI models released regularly by the community. By providing such compute you can monetize your compute power and reduce the bottlenecks for millions of innovators, researchers and startups around the world, while earning a bit of money on the side.
3. Provide shared access to finance and revenue streams generated by AI opportunities
As AI developers, we have to be cognizant of the social impact our work will have on the financial wellbeing and livelihoods of billions of families and individuals around the world. Enabling individuals access to knowhow and access to data, compute and finances to become shareholders and stakeholders in the revenue streams generated with AI is an important consideration to maintain social harmony and avoid social conflicts from arising in the future. Talks of universal basic income in the face of AI driven job loses are already being considered, however a more justified premise for providing such universal income in proportion to data, innovation and compute contributions by individuals makes people receive such income with greater dignity and with a well justified reason as they contribute their data, compute and model IP to the ecosystem. Giving the AI developers, data contributors and compute contributors, all, a persistent income source in a future -more automated with AI, with fewer opportunities for persistent work earning a living, is a vital consideration.
As a startup founder in this space, I am testing out a PoC for Spin at https://synaptrix.org and would appreciate your feedback.
What do you think, what would you like the future of AI to look like?
