In 2010, MIT Media Lab graduate students Karthik Dinakar and Birago Jones collaborated on a project aimed at helping Twitter and YouTube content moderation teams detect problematic content. One daunting challenge they initially faced was understanding the nuanced and coded language that posters were using. This hurdle sparked an understanding that successful model development required more than just machine learning engineers; it needed people deeply familiar with the data they were analyzing.
This recognition led them to create Pienso, a program providing point-and-click tools that enable nonexperts to build their own machine-learning models without coding. Pienso can detect misinformation and provide help in other fields such as human trafficking and weapons sales. The platform’s early iterations involved collaboration with local students for the training of the models.
Jones and Dinakar began their collaborative work at MIT’s Media Lab. The pair worked part-time on Pienso until 2016, when Dinakar completed his PhD and interest in deep learning began to surge. To date, they remain connected to many people around the MIT campus, which they credit for shaping their approach to human-computer interfaces.
The team also acknowledges MIT’s Industrial Liaison Program and Startup Accelerator for connecting them to early partners, including SkyUK. The latter used Pienso to create models that enabled understanding of their customer’s common issues and managing around half a million customer calls daily. The implementation is said to have saved SkyUK over £7 million by reducing call times.
In 2020, Pienso was used by government officials to better understand Covid-19 by mining thousands of research articles about coronaviruses. As an effective, accessible option for businesses, Pienso negates the need to donate data to third-party AI companies. Dinakar likens its interface to Adobe Photoshop for large language models. With a series of stitched web apps, users can point and import, refine, analyze and structure data, and generate a fine-tuned language model within 25 minutes.
A recent partnership with computer hardware company GraphCore has further enhanced Pienso’s efficiency and reduced latency. Looking forward, its founders envision its use in creating a vast array of AI models capable of addressing different needs by the people who understand their data best.
