In 2010, Karthik Dinakar and Birago Jones began a project while at MIT’s Media Lab, aiming to build a tool to aid content moderation teams at companies like Twitter and YouTube. This tool aimed to identify concerning posts, but the creators struggled to comprehend the slang and indirect language commonly used by posters. This led the pair to a valuable insight: those who best understood the nature and nuances of the data should build machine learning models.
This experience motivated Dinakar and Jones to develop a tool capable of assisting non-experts in developing machine learning models. Their creation, named Pienso, allows users to build large language models to detect misinformation, illegal weapon sales, and human trafficking, among other uses, without writing code. The algorithm is so effective due to its method of training; students from nearby schools were enlisted to train the models, leading to more nuanced and effective results.
During their time at MIT, Dinakar and Jones started work on what would become Pienso. Despite holding part-time attention until 2016, Pienso was twice invited to the White House for demonstrations and has gathered considerable interest. Following Dinakar’s completion of his MIT PhD in 2016, increased attention and development on Pienso coincided with a boom in deep learning’s popularity.
One of Pienso’s early partners was SkyUK, whose customer success team used the tool to build models to understand common issues raised by customers. These models reportedly process half a million customer calls per day and have saved the company over £7 million by reducing call length.
The founders of Pienso were contacted by US government officials in 2020 to assist with the understanding of Covid-19 at the outset of the US outbreak. The tool was used to mine thousands of research articles on coronaviruses, assisting in identifying and strengthening critical supply chains for drugs, including remdesivir.
The co-creators of Pienso see a future where AI models are developed for specific uses by the individuals best equipped to understand the problems they intend to solve. Because of the tool’s flexibility – it can run on internal servers and cloud infrastructure without needing data donations – and efficiency offered by its point-and-click interface, Dinakar and Jones believe it offers an important alternative to existing AI solutions and is well positioned to continue making a significant real-world impact.