In 2010, Karthik Dinakar and Birago Jones, students at the Media Lab, collaborated to develop a tool that could assist content moderation teams in identifying concerning posts on platforms like Twitter and YouTube. Their innovation received extensive attention, leading to an invitation to present their creation at a cyberbullying seminar at the White House. However, they encountered difficulties while preparing a working prototype, specifically around slang and indirect language not being picked up by the model. This led to a key insight – the creators of such models needed a deep understanding of the data they were working with, rather than just engineering expertise.
This realization inspired the development of “point-and-click” tools that enable non-experts to construct machine-learning models, forming the basis for Pienso. This tool assists users in crafting efficient language models for a variety of applications, including the detection of misinformation, human trafficking, firearm sales, and more, without the need for coding knowledge.
Together, Dinakar and Jones understood that empowering the data experts, rather than simply democratizing AI, was the more effective approach. They sought to involve those who were best equipped to understand the data, like doctors, journalists, and customer service workers, in the creation of the models.
The founders expanded on the initial system by partnering with students from surrounding schools to enhance the training of the models. Later, they were able to leverage networks at MIT, the university’s Industrial Liaison Program (ILP), and the Startup Accelerator (STEX) to connect with early collaborators. Pienso was deployed by one such partner, SkyUK, to develop models to comprehend common customer issues, processing half a million calls daily while saving the company over £7 million by reducing call lengths.
Pienso’s capabilities were further showcased during the COVID-19 outbreak as government officials sought their assistance to comprehend the emerging disease. The platform facilitated the creation of machine learning models to analyze a vast quantity of research articles about coronaviruses, contributing to the identification and fortification of vital supply chains.
Another advantage of Pienso is its ability to operate on both in-house servers and cloud infrastructure, offering an alternative solution to businesses that would otherwise have to share their data with other AI companies. This user-friendly interface requires no coding knowledge, and users can develop a fine-tuned language model in under half an hour.
Recently, Pienso partnered with GraphCore, a provider of a more efficient and faster computing platform for machine learning. The collaboration aims to further minimize barriers in deploying AI technology by significantly reducing latency. The founders envision a future where effective and unique AI models are developed for specific uses by those best acquainted with the problems they are resolving. Ultimately, it is about creating a collection of models that can collaborate effectively, orchestrated by those who understand the data best.
