In 2010, MIT Media Lab students Karthik Dinakar SM ’12, PhD ’17 and Birago Jones SM ’12 embarked on creating a tool to assist content moderation teams at companies like Twitter (now X) and YouTube. Their demo, which was presented at a cyberbullying summit at the White House, identified troublesome posts through machine learning. However, the model initially failed to recognize certain posts due to the use of teenage slang and other forms of indirect language. This realization provoked a shift in their approach, asserting that those who best understand the data should be the ones building machine learning models, not necessarily the machine-learning engineers.
This new perspective paved the way for the creation of Pienso, a tool designed to make building large language models for identifying misinformation, human trafficking, weapons sales among other issues, easier for non-experts. The tool works without writing any code and is based on a point-and-click system. The researchers worked in collaboration with local students to train the models gaining more nuanced results.
Dinakar and Jones’ work despite beginning as part-time advanced to feature in the White House multiple times and gained attention from various sectors. One of their early collaborators was SkyUK, which successfully utilized Pienso to manage customer problems. The models constructed with Pienso now process approximately half a million customer calls each day.
They believe that creating valuable insights from data is more feasible when the ones building the models have a deep understanding of the data. In 2020, Pienso aided experts in virology and infectious disease to organize machine-learning models during the Covid-19 pandemic.
The fundamental premise of Pienso is to cater to businesses wary of sharing their data. Serving non-AI experts, Pienso operates as a web app. Within minutes, users can import, refine, prepare and analyze data without any coding involved, resulting in large language models.
In partnership with GraphCore, Pienso aims to further reduce barriers with AI by significantly lessening latency and providing faster and more efficient computing platforms for machine learning. Dinakar and Jones are convinced that more effective AI models are possible when those most familiar with the problems at hand are at the forefront of developing solutions.
