Deep Fridays #1: Sanjay Rajagopalan's Vision for Vianai
Human-Centered AI
AI isn't just algorithms and code. It’s design, responsibility, and vision. I spoke with Vianai Systems' Chief Design and Strategy Officer, Dr. Sanjay Rajagopalan, for an extended conversation spanning empathy in design, truth in output, and why enterprises need this focus for their generative AI journeys.
I genuinely enjoyed it. Sanjay is easy to talk to, and even shares an outside-of-tech passion of mine: theatre. He’s one of the original founders of Naatak, the largest Indian theatre company in the United States. We warmed up our conversation on that subject, and from there we launched into deeper empathy, enterprise, and truthfulness subjects.
Empathy-Sharpened Design
Sanjay’s Ph.D is in manufacturing and design. “In general, I’m a product design guy,” he says, and a trip through his website will drive that point home if you have any doubts. We detoured briefly into the general topic of empathy (led by my off-topic ruminations on its role in morality), and Sanjay brought it to bear on designing for AI solutions: “You need to get into [people’s] minds and hearts to really understand what would work for them. You have to see the invisible…imagine something that doesn’t exist today that could potentially eliminate that pain or frustration.”
I got the most out of this part of our chat with his simple job description: “My role is to match that [AI] with business solutions.” I see tons of speculation and misunderstanding of large language models out in the world, and especially so in enterprises. It’s so clearly a natural place to fit empathy, design, and AI together.
And even though I wandered briefly into outer space on it, I’m glad we touched on empathy as a core guide of the whole process. In the last six or seven years I’ve made personal strides in this area, and every chance I get I nudge people in that general direction. In my view, it’s the centerpiece of ethics…and that implication goes outward into conducting business.
Digging For Truth
From there, we delved into enterprise use of generative AI. I wrote about Vianai previously and mentioned their hila tool. From this conversation, I can update and share my understanding of a nuance I missed in my earlier reading.
The main place to experience hila is the hila.ai website. But hila isn’t primarily hila.ai - hila.ai is just an example you can use to get an understanding of the hila platform. For enterprise customers, hila becomes what they need for their ground truths - whether that’s finance, supply chain, or other company information. hila is a platform, and hila.ai is an instance of an application built on that platform.
We also steered into truth and verification. Pretty much any enterprise-y person exploring generative AI eventually hits a hard problem: what do we do about hallucinations? There’s a trap: things like ChatGPT take on characteristics that humans interpret as human-like, and speaks in a very knowledgeable tone. Many of us begin to interpret that tone as authoritative, which causes interpretation problems when LLM output isn’t really grounded in verifiable truth. It’s such an easy trap to fall into: you can’t tell that the LLM has no intention of doing right by the truth. It’s just predicting tokens.
Enter veryLLM. It’s the start of Vianai’s toolkit of verification and hallucination mitigation. It can take output generated by an LLM and compare it with trusted sources of truth, and present users with a second set of validation eyes. Basically, it answers the question “how close is what the LLM said to the actual truth?” There’s an acknowledgement that this problem is bigger than one project, so with that in mind Vianai open-sourced veryLLM with the motivation to drive community efforts toward LLM truthiness. It’s an ambition I can get behind.
Changing The World
Sanjay acknowledges the dual-edged nature of AI technology. The potential for great benefit is matched by the potential for great harm. He again emphasizes the need for holistic, human-centered AI approaches that are focused on positive impact.
The last part of this conversation gave me a clearer picture of the Vianai strategy. Sanjay talked about a vision of the future where, for enterprises, there isn’t a single mega-genius model. There are swarms of models that receive, orchestrate, and respond to requests. Vianai recognizes that there are going to need to be a TON of tools in the mix: verification, cleaning, processing input and output, and so on. The couple of DALL-E generated images I’ve used in this piece actually helped me think through this analogy: There’s an ocean of AI out there, and LLMs are giant whales swimming in it - but enterprises need all the rest of the ocean ecosystem that swim with the whales to get the best out of AI. PM