In 2010, Media Lab students at MIT Karthik Dinakar and Birago Jones began working on a project to help content moderation teams of companies like Twitter and YouTube. They developed an AI model to identify sensitive posts, but soon realized that the terminology and language used by some users was overlooked by the model. Challenged by this obstacle, they concluded that AI models should be developed by those who best understand their data. From this idea emerged Pienso, an innovative tool that enables non-experts to build machine-learning models without writing any code.
Pienso provides users with point-and-click tools to create machine-learning models for various purposes, such as detecting misinformation, identifying instances of human trafficking or weapons sales, and much more. This method allows users who are familiar with their own nuanced data and its specificities to tailor AI models to suit their needs.
Dinakar and Jones’ work initially began in a natural language processing course at MIT and later expanded. Throughout their journey, they collaborated with nearby schools, letting the students train the models, and were invited to the White House more than once to demonstrate their project. Their connections at MIT and the university’s Industrial Liaison Program and Startup Accelerator proved beneficial in helping them establish early partnerships.
One such partner was SkyUK, which used Pienso for their customer service team to comprehend common customer issues. This has reportedly saved the company over £7 million by reducing call duration in their call centers.
In response to the Covid-19 outbreaks in 2020, Pienso was used by US government officials and experts in virology and infectious diseases to sift through coronavirus research literature. This helped the government identify critical supply chains for antiviral drugs like remdesivir.
Pienso, which can operate on internal servers and cloud infrastructures, offers businesses an alternative to divulging their data when using the services of other AI companies. With its intuitive interface, users can import, refine, analyze and structure data to create fine-tuned language models in a matter of 25 minutes.
Recently, the company announced a partnership with GraphCore, an advanced computing platform, to diminish latency and improve AI accessibility. Dinakar and Jones believe their solution is pioneering a potential future where AI models are developed by the people closest to the problems they are seeking to solve. They emphasize the importance of diverse AI models that can be orchestrated together, tailored to specific needs, which they believe will lead to more effective solutions.
