Open-source developers are catching up. Open-weight models now match the capabilities of closed corporate systems almost immediately upon release. An independent global network iterates faster than a single corporate lab.
Small businesses and independent researchers download open-source code and run it locally. They bypass the massive enterprises that charge per token for server access. People are running machine learning on their own hardware. This redistributes computing power.
Hardware is the main bottleneck for machine learning. Massive data centres full of graphics processing units gave major tech companies an advantage. Decentralised compute networks have changed this. Developers now rent distributed GPU capacity from independent operators worldwide. This bypasses the cloud computing giants by pooling idle resources.
Federated learning allows people to train models across thousands of local devices. The data stays with the user. A London hospital and a Madrid clinic can train a diagnostic tool together. They never share a single patient's confidential record. In a world of data breaches, this approach protects sensitive information.
Artificial intelligence manages supply chains and handles healthcare triage. It filters information and executes transactions. Relying entirely on proprietary models means outsourcing logic to unaccountable corporations. We cannot audit corporate systems for bias or systemic flaws. We give them our data, and they sell the resulting models back to us.
Open-source development makes the underlying code public. Independent researchers and journalists can inspect the software. Communities outside the US and China can build digital systems that reflect their own languages and cultures. They can avoid importing homogenised corporate tools.
Governments and institutions need to fund open-source ecosystems. Current legislation often targets these open projects while attempting to regulate safety. True safety requires transparency and public scrutiny. The public must control these tools.