In 2010, while studying at MIT Media Lab, Karthik Dinakar and Birago Jones developed a tool to assist in content moderation for social media platforms like Twitter and YouTube. The project, aimed at identifying concerning posts and potential cyberbullying, sparked enough interest to receive an invitation to a cyberbullying summit at the White House. However, the night before the event, Dinakar discovered a flaw in their model – it struggled to identify posts that used teen slang and indirect language.
This realization highlighted a vital factor in developing such tools. People who understand the data they’re dealing with should be the ones building these models. With this new perspective, Dinakar and Jones went on to create point-and-click tools, allowing non-experts to develop machine-learning models. These tools formed the foundation of Pienso, a platform used for building large language models capable of detecting misinformation, human trafficking, and more.
In providing a solution for those who understand the data best to assist in building models, the two founders began collaborating with local school students to refine their models. This experience led to the “Aha!” realization that their strategy should empower domain experts rather than simply democratizing artificial intelligence.
This approach catalyzed the expansion of Pienso. Not only were they invited back to the White House, but the platform itself started gaining momentum in 2016, following Dinakar’s graduation and the rise of deep learning’s popularity. Pienso has since been utilized by companies like SkyUK to build models to understand customer problems and even help process half a million customer calls per day.
In 2020, as the Covid-19 pandemic broke, government officials approached Pienso to better understand the virus. The platform facilitated experts in setting up machine-learning models to assess thousands of research articles regarding coronaviruses. This work played a crucial role in identifying and building robust supply chains for antivirals like remdesivir.
Currently, Pienso is capable of running on internal servers and cloud infrastructure, presenting an alternative for businesses reluctant to share their data with other AI companies. The platform functions like “Adobe Photoshop for large language models” which allows users to import and refine data, prepare it for deep learning, and derive structured insights, all without coding.
Pienso’s recent partnership with GraphCore is set to further lower the barriers to employing AI by significantly reducing latency. This makes the AI platform faster and more responsive to the user.
A core belief the co-founders emphasise is that the future of effective AI models lies in their development by people well-versed with the problems they are trying to solve. This approach not only brings together a diverse range of models, it also places those who best understand the data in charge, enabling them to orchestrate the process effectively.
