In 2010, Karthik Dinakar and Birago Jones began a project to develop a tool helping content moderation teams at companies like Twitter and YouTube. The aim was to assist these teams in identifying inappropriate or harmful content. The project created considerable interest and the researchers were invited to present their work at a cyberbullying summit at The White House.
The initial idea was to identify concerning posts on social media platforms such as Twitter. However, Dinakar realized that his model had trouble understanding the complexity of language, particularly the usage of slang and indirect language used primarily by teenagers, rendering it less effective.
Understanding that the people building machine-learning models should be able to comprehend their data, and, most importantly, the context it was derived from, the two MIT graduates created point-and-click tools allowing non-experts to create their own machine-learning models. The tools later evolved into a venture called Pienso.
Pienso helps build substantial language models to detect misinformation, human trafficking, weapon sales, and more. This is achieved without writing any code. Instead of trying to democratize AI, the researchers focused on empowering the experts in each domain to build their own models for these different use-cases. Students from Cambridge, Massachusetts were invited to help train the models, producing results that were much more nuanced and capable.
The founders continued developing Pienso until deep learning gained traction in 2016. They partnered with SkyUK to build models that can understand customer problems. The models reportedly process around 500,000 customer calls daily and have saved the company over £7 Million by reducing call lengths.
During the initial phase of the Covid-19 pandemic, Pienso was also used to analyze thousands of research articles about coronaviruses. The results helped the government identify and strengthen the supply chains for key drugs, including popular antiviral remdesivir.
Pienso operates on internal servers and cloud infrastructure, which makes it an alternative for businesses that don’t want to share their data with other AI service providers. The interface provides a seamless experience allowing users to refine data without any coding experience.
A partnership with GraphCore made the machine-learning platform faster and reduced latency, creating a more efficient ecosystem for users. The founders envision a future where users well-versed with their problems can use Pienso to develop AI models capable of effectively solving those issues.
