On February 20, attendees at the India AI Impact Summit watched an unusual demonstration unfold. An artificial intelligence-powered device scanned a table and identified the candy bars laid out on it — Twix, Milky Way, and KitKat. Then the device did something more human: It picked a favorite. Twix.
The small, black, open-source device was unveiled by Current AI, a $400 million public-interest partnership, which brings together governments, foundations, and private companies to develop alternatives to Big Tech-dominated AI systems. The prototype, equipped with a microphone, camera, and screen, was built in collaboration with Bhashini, an Indian government-backed, voice-based AI translation project that supports 22 Indic languages.
The demo with the candy bars was targeted toward the visually impaired. The demonstrators asked questions such as “What do you see in front of me?” or “Which candy bars are on the table?” in Hindi and English, and received detailed answers in the same languages. Despite running offline on limited hardware, the system showed no drop in accuracy, according to Shailendra Pal Singh, senior general manager at Bhashini.
Current AI and Bhashini plan to release the full open-source designs and development instructions for the device on GitHub. “Our hope is that anyone could feel empowered to connect up to this device, write their own application, pull any number of models onto the device, and run inference locally in their hand,” Andrew Tergis, the Current AI engineer who worked on the debut collaborative AI build-out, said on stage.
For Current AI CEO Ayah Bdeir, the project reflects a long-held commitment to open hardware. Rest of World spoke to Bdeir about how open-source AI can serve diverse communities while reducing Big Tech dependency.
This interview has been edited for length and clarity.
Why did you create this handheld, offline AI device?
AI used to be abstract. It was something in the cloud or far away that you didn’t necessarily feel. Now, with embodied AI, we’re seeing larger companies create devices that will enter your home. They will get close to your body, enter your personal space, like glasses, butler robots, and smart home devices. These devices are recording you all the time, and they’re sending your data to the cloud. You don’t know how they work, know what they’re made of, or how they’ve been trained. Oftentimes, they’ve been trained on mostly Western languages.
That’s why we’re offering this alternative platform that is more open, customizable, and inventive, so that we can have our own version of what that looks like.
How do you ensure such AI-led public goods can scale up?
We interact in three ways: We fund and develop grants for people doing interesting work in public-interest AI. We build in areas where we think nobody else is investing because it’s not financially viable. And we invest in movement infrastructure because we believe that we can all learn from each other, and that’s how we’re going to make public-interest AI scalable.
There is already a lot of interesting public-interest AI work. It’s just decentralized, and it’s not being coordinated.
A lot of AI innovation is happening in the West. How do we make AI development more equitable?
It is a danger, and it's something that we are actively aware of and working on. The silver lining is that amid political instability and with many big tech companies overreaching, many people are seeing the risks and are disenchanted with that behavior. There's a fervent and strong desire for an alternative. That's a good thing for us. It means that when we make a call to create an alternative to collaborate towards an open and vertically integrated public interest stack, people are interested in participating in a way that they perhaps weren't before.
What are the biggest challenges in AI-powered public good initiatives, and how are you overcoming them?
There are two really important themes we're thinking about. One theme is culture preservation, which started with linguistic diversity, but we’ve expanded it to culture preservation because culture is more than just language.
The second one is around the idea of resilient AI or frugal AI. There's this continuous push for bigger, better — biggest data center, most powerful, largest model, the biggest data set — and we think that’s just not what serves the entire world. We’re focused on what it means for someone in a low-connectivity or no-connectivity area to have AI. What does it mean to create a version of AI that is small, portable, and low-resource? What does it mean to have a version of AI customized to your needs?
How can innovators outside the West survive and thrive?
By looking at what’s happened over the past 20 years, a period where we abdicated technology innovation and creativity in technology to larger tech companies. We have a lot of resulting harm, including dangerous societal impacts.
The internet, when it was first conceived, was open and free. Anybody could develop their own website, start their business, and create their own little community. It was a platform for creativity and innovation. Gradually, it started becoming a walled garden. We need to learn from that mistake and say we’re not going to let it happen in AI. ▰
Ananya Bhattacharya is a reporter for Rest of World covering South Asia's tech scene. She is based in Mumbai, India.
