As Media Lab students in 2010, Karthik Dinakar SM '12, PhD '17, and Birago Jones SM '12 teamed up for a class project to create a tool that would help content moderation teams at companies like Twitter (now X ) and YouTube. The project generated enormous enthusiasm and the researchers were invited to demonstrate at a cyberbullying summit at the White House; They just had to get it up and running.
The day before the White House event, Dinakar spent hours trying to put together a working demo that could identify concerning posts on Twitter. Around 11 p.m., he called Jones to tell him he was giving up.
So Jones decided to look at the data. It turned out that Dinakar's model was pointing to the right types of posts, but the posters used teen slang terms and other indirect language that Dinakar didn't catch. The problem was not the model; it was the disconnect between Dinakar and the teenagers he was trying to help.
“Then we realized, right before we got to the White House, that the people who build these models shouldn't just be machine learning engineers,” Dinakar says. “They should be the people who best understand your data.”
This idea led researchers to develop point-and-click tools that allow non-experts to create machine learning models. Those tools became the foundation for Pienso, which today helps people create great language models to detect misinformation, human trafficking, gun sales, and more, without writing any code.
“These types of applications are important to us because our roots are in cyberbullying and understanding how to use ai for things that really help humanity,” Jones says.
As for the first version of the system shown at the White House, the founders ended up collaborating with students from nearby schools in Cambridge, Massachusetts, to allow them to train the models.
“The models those kids trained were much better and more nuanced than anything I could have come up with,” Dinakar says. “Birago and I had this big 'Aha!' “At which point we realized that empowering domain experts, which is different from democratizing ai, was the best way forward.”
A project with purpose
Jones and Dinakar met when they were graduate students in the MIT Media Lab's Software Agents research group. Their work on what became Pienso began in course 6.864 (Natural Language Processing) and continued until they earned their master's degree in 2012.
It turned out that 2010 was not the last time the founders were invited to the White House to demonstrate their project. The work generated a lot of excitement, but the founders worked at Pieso part-time until 2016, when Dinakar finished his PhD at MIT and deep learning began to gain popularity.
“We are still connected to a lot of people on campus,” Dinakar says. “The exposure we had at MIT, the fusion of human and computer interfaces, expanded our understanding. Our philosophy at Pienso could not be possible without the vitality of the MIT campus.”
The founders also credit MIT's Industrial Liaison Program (ILP) and Startup Accelerator (STEX) for connecting them with early partners.
One of the first partners was SkyUK. The company's customer success team used Pienso to create models to understand their customers' most common problems. Today, those models help process half a million customer calls a day, and the founders say they have saved the company more than £7 million to date by shortening the length of calls to the customer's call center. the company.
“The difference between democratizing ai and empowering people with it comes down to who understands the data best: you, a doctor, a journalist, or someone who works with clients every day? Jones says. “Those are the people who should create the models. “That’s how you get insights from data.”
In 2020, just as Covid-19 outbreaks began in the US, government officials contacted the founders to use their tool to better understand the emerging disease. Pienso helped virology and infectious disease experts set up machine learning models to mine thousands of coronavirus research articles. Dinakar says they later learned that the work helped the government identify and strengthen critical supply chains for drugs, including the popular antiviral remdesivir.
“Those compounds were discovered by a team that didn't know deep learning but could use our platform,” says Dinakar.
Building a better future for ai
Because Pienso can run on internal servers and cloud infrastructure, the founders say it offers an alternative for companies that are forced to donate their data by using services offered by other ai companies.
“The Pienso interface is a series of web applications tied together,” explains Dinakar. “You can think of it as an Adobe Photoshop for large language models, but on the web. You can point and import data without writing a line of code. You can refine the data, prepare it for deep learning, analyze it, give it structure if it's not labeled or annotated, and you can end up with a large, fine-tuned language model in a matter of 25 minutes.”
Earlier this year, Pienso announced a partnership with GraphCore, which provides a faster, more efficient computing platform for machine learning. The founders say the partnership will further lower barriers to leveraging ai by dramatically reducing latency.
“If you're building an interactive ai platform, users won't grab a cup of coffee every time they click a button,” Dinakar says. “It has to be fast and responsive.”
The founders believe their solution is enabling a future where people who are more familiar with the problems they are trying to solve develop more effective ai models for specific use cases.
“No model can do everything,” says Dinakar. “Everyone's application is different, their needs are different, their data is different. It is highly unlikely that a model will do everything for you. “It’s about bringing together a set of models and allowing them to collaborate with each other and orchestrate them in a way that makes sense, and the people doing that orchestration should be the ones who best understand the data.”