Long before LLMs became accessible to just about anyone with a computer and internet access, Siddhartha Dalal was working to build algorithms that used AI to solve some of the most consequential problems facing humankind.
Dalal has had a prolific career leading data and research departments. Among his many roles, he served as CTO of the RAND Corporation, chief data scientist and senior vice president at AIG, and vice president of research at Xerox, starting from Bell Labs Research, working on projects that use AI to reduce harm across national defense, construction, insurance, and beyond. Dalal has over 100 peer-reviewed publications and patents. Most recently, he was awarded a $3.2M IARPA grant (2026–27) as co-principal investigator alongside principal investigator Vishal Misra, RKS Family Professor of Computer Science and vice dean of computing and artificial intelligence at Columbia Engineering, and co-principal investigator Matthew Connelly, vice dean for AI initiatives and professor of history at Columbia's School of Arts and Sciences. The grant supports the BENGAL initiative, which works to block sensitive data extraction and identify malicious users in large language models, safeguarding national security information while enabling secure intelligence sharing.
Today, Dalal is sharing his knowledge as a professor of professional practice in the M.S. in Applied Analytics (APAN) program at Columbia University School of Professional Studies (SPS) with a joint appointment as Affiliated Faculty in the Department of Statistics.
Dalal was recently interviewed by TV Asia for a wide-ranging conversation about his career. In the follow-up discussion with SPS below, Dalal shares deep insights into the applications of AI beyond LLMs that have captured the current zeitgeist, and his approach to teaching.
In your interview with TV Asia, you mentioned that your awareness of—and interest in—AI stems all the way back to the tragic Space Shuttle Challenger disaster in 1986, which demonstrated the importance of accurate algorithms for predicting risk and outcomes. Could you tell us about that experience?
The Space Shuttle Challenger launched on January 28, 1986, on a morning with temperatures in the low 30s. The night before, NASA and its contractor, Morton Thiokol, debated the safety of launching in such cold conditions. Despite these concerns, the launch proceeded. The O-rings—meant to prevent leaks—failed, and the shuttle was destroyed, killing all aboard, including a first civilian, an elementary schoolteacher.
In the aftermath, NASA sought to determine whether the failure could have been predicted. I was asked to join the scientific team, and the models we created indicated a 16% chance of failure—a roughly 1-in-6 risk with human lives at stake. Had the launch been delayed until temperatures reached around 60 degrees, the risk would have dropped to about 1%. This was a significant finding. The predictive methods we created are now considered part of artificial intelligence, and our work is taught in high schools. From that point on, I became deeply involved in predictive analysis.
What other kinds of AI and prediction projects have you worked on?
I was funded by the U.S. government to develop algorithms that detect nuclear materials—such as those in dirty bombs—entering ports in shipping containers. I’ve also worked in construction, using AI and cameras to warn workers of potential injury risks, and helped build an automated system for car accident claims, where algorithms analyze images, assess damage, and settle claims without an adjuster. I also invented algorithms that created a new research field, Combinatorial Design Testing, now used worldwide for software and hardware testing.
I’ve also explored using AI to help people engage with ancient religious texts written in archaic languages. For example, I worked on a project that translated Buddhist and Jain texts from Prakrut into multiple languages—English, French, Hindi, and German—and enabled users to interact with them through dialogue in their own language. This work has been published and aims not just to preserve disappearing languages but to democratize access to the ideas within these traditions.
So my work spans both applying AI and inventing new algorithms—finding new ways of thinking and reshaping how we approach their use.
There have been many people warning about AI as a danger to society. What do you think about that?
AI is helpful, much like a pen is for writing—a tool that enables certain tasks. Others may disagree, but that’s how I see it: a medium, similar to how computers enable the creation and exchange of information. Yet, just as printing reshaped the world, AI will likely have a massive impact. For example, the human-oriented web in its current form is likely to die and be replaced by the one designed for interaction with AI agents.
Can you speak about how you teach AI in your lectures?
There’s a lot to learn from AI’s history, but my course emphasizes a balance of theory and practical application. For example, students work with lung image datasets to build AI classification systems that identify specific diseases. I also discuss what makes applications meaningful and push students to propose new ideas and approaches that could lead to breakthroughs. I encourage them to think beyond the classroom and define their own problems. In my Deep Learning course, final projects are student-driven: they identify a problem, design an approach, assess potential failure points, and ultimately, build and evaluate a system based on its performance.
I also want to make sure students are doing well in terms of their careers and meeting their career goals. But that starts with really understanding the current state. They need to understand the basic technology and apply it to develop applications.
What about how your students use AI to complete the work? Do you place any restrictions on how that AI is used?
I don’t restrict AI use—students would use it anyway. We’ve seen this before: slide rules gave way to calculators, along with debates about allowing them in class. What matters is that students truly understand and can explain their work. I emphasize to students that AI is like an encyclopedia: exploring one topic will lead to related ones, expanding their understanding.
But AI has limits—it depends on existing data. If a new idea isn’t represented, it won’t generate it. That’s why originality matters, and it’s what I emphasize with my students.
About the Program
Columbia University’s Master of Science in Applied Analytics prepares students with the practical data and leadership skills to succeed. The program combines in-depth knowledge of data analytics with the leadership, management, and communication principles and tactics necessary to impact decision-making across industries and organizational functions.
Learn more about the program here. The program is available full-time and part-time, online and on-campus.
