Recorded at the AI for Good Global Summit in Geneva, 2026.
AI adoption is speeding up, but not evenly. This dispatch covers what several speakers argued: the gap is less about models and more about infrastructure, skills and accountability.
Key takeaways
- In many regions the bottleneck isn’t the model. It’s electricity, internet access, and whether the skills exist yet.
- Money for compute doesn’t help if people get left behind. They need training to build and deploy AI themselves, not only to use it.
- Harvard engineering professor Vijay Janapa Reddi: the model and its training are the tip of the iceberg. Engineers decide whether AI is efficient, scalable, interoperable and affordable, and we need many more of them.
- Julie Owono, an inaugural member of Meta’s Oversight Board, argued that frontier AI has to answer to people nowhere near the room where it’s built if it’s going to meet basic human rights standards.
Try this with your team
If your organization received funding for AI tomorrow, how much of it would go to people (training, time, support) versus tools? What split would you defend to your board?
Going further
Our cohort training is built on the idea that people, not tools, decide whether AI helps. If budget is the barrier, see AI Discounts and Grants for Nonprofits.
More from this series
Transcript
Lightly edited from YouTube’s automatic captions for readability. Wording and order are the speakers’ own.
Some of the things that stuck with me from today. AI adoption is accelerating, but not evenly. There are whole regions where the story isn’t the model. It’s electricity and internet access, and whether the skills exist yet.
People need investment just as much as infrastructure does. If you pour money into compute and still leave everyone else behind, that’s not very hopeful. People have to be trained, and trained to deploy it themselves, not just use it.
My favorite presentation of the day was from an engineering professor from Harvard named Vijay Janapa Reddi. What he did was a good memory jog: the AI model and the training itself are just the tip of the iceberg in terms of how AI can be deployed. Engineers are the ones doing that. They’re the holistic viewers of how AI can be planned and executed, and how efficient, scalable, interoperable and affordable it can be. So we need to train more engineers, basically.
One of the inaugural members of Meta’s Oversight Board, Julie Owono, made the point that frontier AI has to answer to the people who are nowhere near the room where it’s created and engineered, if it’s going to clear any basic bar of human rights.
Working with Meta, she would know that Facebook is essentially to blame for the genocide in Myanmar, in the way social media was deployed there. Among other things, like hate speech: if the people affected by what happens in AI are not at the table, then the harm is going to happen.