Empowering everyday experts to solve complex problems without needing to dip into coding, Pienso, an AI platform, has launched a user-friendly AI builder designed for the layperson. The innovative tech, conceived by MIT Media Lab alumni Karthik Dinakar and Birago Jones, strives to put powerful AI tools into the hands of those who understand their domains best, whether they be doctors, journalists, or anyone familiar with their unique data sets. "The people building these models should not be folks who are just machine-learning engineers," Dinakar told MIT News, emphasizing the importance of domain-specific understanding in AI development.
Initially crafted to aid content moderation on social platforms, Pienso's early demonstration was set back by a failing model unable to catch concerning teen slang until a last-minute revelation by Jones identified the glitch; then the system significantly improved once teens themselves were brought in to inform the AI, and the entrepreneurs realized the potential in enabling subject-matter experts to train machine learning models—a realization they brought to life with their point-and-click tools. Dinakar recollects, "The models those kids trained were so much better and nuanced than anything I could’ve ever come up with," marking a pivotal moment in the project's journey towards building a platform that meshes the insights of human expertise with the analytical might of AI.
The impact of Pienso extends into various sectors, from improving customer service efficiency at SkyUK to aiding government officials during the Covid-19 pandemic by analyzing infectious disease research. Dinakar reveals that, through their platform, a team lacking in deep learning expertise managed to identify key drug compounds for tackling the virus. The founders also seek to prevent businesses from being pressured into sharing their data with AI companies by offering the means to run Pienso on internal servers and cloud infrastructures.
Pienso's platform boasts of a seamless user experience likened by Dinakar to "an Adobe Photoshop for large language models, but in the web" allowing users to build and adjust AI models while bypassing the traditional barriers of deep learning and its associated coding demands. Fresh partnerships like the one with GraphCore, as the founders say, aim to boost model accessibility by further cutting down latency issues ensuring that user interaction with the AI remains fast and responsive, "If you’re building an interactive AI platform, users aren’t going to have a cup of coffee every time they click a button," Dinakar explains, indicating the need for speed in the ever-demanding landscape of AI operations.
The essence of Pienso's philosophy rests on the understanding that unique problems require unique AI models, an approach that champions the direct involvement of those who have in-depth knowledge of the problems at hand. There's an acknowledgment that the value lies not just in democratizing AI but in tailoring its capabilities to harness the insights and expertise of its users. "Everyone’s application is different, their needs are different, their data is different," says Dinakar concluding that comprehensive AI solutions will come from a diversity of models working together, directed by those who comprehend the data's intricacies.
