As students at the Media Lab in 2010, Karthik Dinakar SM ’12, PhD ’17 and Birago Jones SM ’12 embarked on a project to build a tool for moderating content on platforms such as Twitter and YouTube. The project received so much attention that they were invited to demonstrate it at a cyberbullying summit at the White House. However, the night before the presentation, Dinakar realised that their model was not identifying problematic posts correctly due to a misunderstanding of teenage slang language in the posts. The team learnt that the best people to build machine-learning models were those who had a thorough understanding of the data being used.
This realisation led to the development of Pienso, a tool that allows non-experts to create machine-learning models without writing any code. Today, Pienso is helping to build larger language models that can identify misinformation, human trafficking, selling of weapons and more. Dinakar and Jones have stayed connected with MIT, crediting the university’s Industrial Liaison Program (ILP) and Startup Accelerator (STEX) for introducing them to early partners.
One of these initial partners was SkyUK, a company whose customer success team used Pienso to create models to understand their customers’ common issues. These models process half a million customer calls daily and have saved the company over £7 million by reducing the duration of calls to their call center.
In 2020, during the onset of the Covid-19 pandemic in the U.S, health officials used Pienso to understand the disease better. Experts in infectious diseases and virology used the platform to investigate thousands of research articles about coronaviruses. This research helped the government identify and strengthen crucial supply chains for drugs, including the widely used antiviral remdesivir.
The Pienso team believes that their solution will enable the development of more effective AI models for specific applications by those who are most familiar with the issues they are trying to address. They stress that one model cannot do everything due to variations in application, needs and data. They envision a future of orchestrated collaboration between a variety of models, guided by those who understand the data best.
