Written by Deb Roy, CEO of Cortico & Director of MIT Center for Constructive Communication (MIT CCC)

ChatGPT has taken the world by storm. This new artificial intelligence (AI) technology has begun to impact our lives overnight. Increasingly, people who had little or no experience with AI chat technology are using ChatGPT to help them brainstorm, read, write, and be entertained. Companies are instructing their staff to integrate ChatGPT into their work. And of all things, Microsoft’s search engine has given us a peek into a machine that seems truly social – or perhaps more accurately, socially deranged. 

The disruption is hard to miss. Educators are discovering that their exams can be aced by a tool that every student can freely access. Some people are having the AI tool rewrite their emails to land better with recipients. Others are forming “friendships” with AI companions. Elected representatives will not know if they are hearing from constituents or bots. It feels like we have arrived at a watershed moment in AI. 

With a reported 100 million people already using the ChatGPT service, many of us are now wondering: How do we manage the power, promise, and peril of this new technology? What jobs will we be able to automate? What new human endeavors become possible? What are the risks? What harms might this technology usher in? What ethical issues will arise? We are actively exploring these questions at Cortico in collaboration with the MIT Center for Constructive Communication (CCC). In fact we have had a running start.

As someone who has worked with AI for decades, I was stunned in my first encounters with ChatGPT for two reasons. First, the underlying language model has shown sophisticated linguistic competence that even many in the field did not anticipate (including the engineers who built the model). Second, the team that built ChatGPT made an easy-to-use chat interface and put in place a scalable infrastructure that enabled millions of people to directly experiment with this new form of technology. This has sparked a conversation about AI like no other in the history of the field.

I am excited about how the advances in language modeling on display in ChatGPT – and increasingly available more generally – will supercharge Cortico’s platform. There are clear opportunities on the near horizon for improvements that include better search, automatic summarization, efficient sense-making, and AI-assisted design of conversation guides. 

The heart of ChatGPT is a language model, in this case a collection of over 175 billion parameters (decimal numbers) that capture incredibly nuanced patterns of language extracted from hundreds of billions of words of text scraped from the Internet. My personal connection with language modeling runs deep. My PhD thesis in the late 90s was focused on language modeling, and much of the research I lead today at CCC is based on language modeling (including a class we are teaching this semester). 

But we are also aware of risks including the introduction of bias and blind spots. An AI model trained on Internet data tends to reflect the biases of the Internet. Voices and perspectives that are underheard on the Internet will tend to be underrepresented in the AI models. If that model in turn is used to automatically interpret community voices, this could easily yield systematic reinforcement of structural biases and defeat the purpose of our work. 

We know how important it is for our partners to build trust with community members as conversations are recorded to better understand how shared experiences can spark change. Our goal is to move as fast as possible to leverage new technologies while taking the time to understand, evaluate, and avoid the possible downsides. The best way we know how to do this is in collaboration with our partners.

At MIT we already have a number of experiments underway to develop new AI capabilities that promise to advance the capabilities of the Cortico platform. In the coming months CCC will work with Cortico to identify learning partners who are interested in piloting, evaluating, and refining new prototype tools and methods. Let us know if you want to get involved. And as we find ways to responsibly introduce new AI capabilities into our platform, the Cortico product engineering team will integrate them into our product roadmap so that they become generally available.

Our mission is to leverage the power of technology to foster real talk and real listening between people. With the rise of artificial intelligence that is able to mimic human-level linguistic behavior, we will experience a tsunami of artificial talk and artificial listening. This means our mission will be more valuable than ever. Our resolve is to leverage the power of AI in ways that keep us firmly on this path.