Skip to content Skip to footer

In 2010, Media Lab students Karthik Dinakar SM ’12, Ph.D.’17, and Birago Jones SM ’12 developed a tool intended to assist content moderation teams for platforms such as Twitter and YouTube. The tool aimed to flag harmful content, with a key focus on posts that could be linked to cyberbullying. The project was warmly received, leading to an invitation for a demonstration at a cyberbullying summit hosted at the White House. However, issues emerged when Dinakar discovered that, although the model could correctly identify offensive posts, it struggled to interpret teenage slang and other forms of indirect language. Consequently, the problem lay not with the model itself but rather with Dinakar’s lack of knowledge regarding the language being used.

This realization, just prior to the White House demonstration, prompted the researchers to highlight the importance of building machine-learning models not only to engineers but to people who are well-acquainted with the associated data. Hence, the pair went ahead to develop point-and-click tools that permitted nonexperts to create machine learning models. From this endeavor, Pienso was born, allowing users to create language models that can detect misinformation, human trafficking, and weapons sales among other issues, without any requirement of coding.

While the early version of their system was presented at the White House, the researchers subsequently worked with students from nearby schools in Cambridge, Massachusetts. Their collaboration led to significantly improved and nuanced models, suggesting to Dinakar and Jones that the most effective approach was empowering domain experts.

Although Pienso remained a part-time project until 2016 when Dinakar finished his Ph.D. and deep learning became widespread, its impact has been substantial. Early partner SkyUK relies on models created using Pienso to better understand their customer’s issues, which now help process around half a million customer calls each day, saving £7 million.

Moreover, in 2020, Pienso was pivotal in the U.S. government’s response to the emerging Covid-19 pandemic. Utilizing Pienso, experts in the areas of virology and infectious diseases could instantly build machine learning models to examine large quantities of research articles about coronaviruses. This reportedly helped identify and reinforce critical supply chains for certain drugs, including remdesivir, a popular antiviral.

Due to its ability to function on internal servers and cloud infrastructure, Pienso offers businesses an alternative to donating their data when using services offered by other AI companies. Earlier this year, Pienso also announced a partnership with GraphCore, expected to profoundly diminish latency, making the tool quicker and more efficient.

Dinakar and Jones envisage a future where specialized AI models are created by those who best know the problems they are attempting to resolve. They believe strongly in the collaborative potential of models and suggest that not one single model can be a solution for everyone, given the unique requirements and data belonging to each individual.

Leave a comment

0.0/5