In 2010, Media Lab students Karthik Dinakar and Birago Jones began a class project to create a tool for content moderation teams at major corporations like Twitter and YouTube. The technology, which generated significant interest, was presented at a White House cyberbullying summit as a mechanism for identifying concerning posts on social media platforms. During the final hours of refining the model, the duo discovered that the system was correctly identifying problematic posts, but struggled to understand teenager slang and indirect language. The discovery revealed the importance of developers understanding their data.
Recognizing this gap led Dinakar and Jones to develop Pienso, a point-and-click tool enabling non-experts to create machine-learning models. These tools are now being utilized to facilitate the construction of large language models that identify misinformation, human trafficking, illicit weapon sales, and more, without the need for coding.
The original version of the system was further improved by Cambridge-based students who trained the models, providing nuanced results.
Pienso was developed part-time until 2016 when Dinakar completed his Ph.D. at MIT and deep learning surged in popularity. MIT’s Industrial Liaison Program and Startup Accelerator aided in the connection with early partners. SkyUK was one of these partners, using Pienso to build models to identify customer problems. These models now process an average of half a million customer calls daily, saving the company over £7 million by minimizing call durations.
The tool was also used in 2020 to comprehend the emerging Covid-19 disease. Experts in virology and infectious diseases utilized Pienso to set up machine-learning models that examined thousands of coronavirus-related research articles. This data helped government officials identify and strengthen crucial drug supply chains, such as the antiviral remdesivir.
Pienso functions on both internal servers and cloud infrastructure, presenting an alternative to businesses compelled to share their data with other AI companies. The Pienso interface is composed of several integrated web apps, providing a user-friendly platform that doesn’t require coding. Through a recent partnership with GraphCore, Pienso aims to lower hurdles to access AI through reducing latency. Pienso’s broader goal is to facilitate the creation of effective AI models for specific use cases by those most familiar with related challenges.
