Vishal Sikka is the Founder and CEO of Vianai, former CTO of SAP AG, and former CEO of Infosys. He currently also serves on Oracle’s board of directors, the supervisory board of the BMW Group and as an advisor to the Stanford Institute of Human-Centered AI.
The Vianai platform combines open-source elements, Vianai-proprietary techniques and optimizations, and human-centered design to bring AI to the enterprise at scale, across diverse landscapes. With the platform, large organizations can build, optimize, deploy and manage sophisticated ML models on existing infrastructure and improve the operations and performance of ML models across the enterprise,
What initially attracted you to machine learning?
I became interested in AI as a teenager, when I read Marvin Minsky’s musings on our minds as societies of simple agents, and learned about Joe Weizenbaum’s Eliza (a very early chatbot) and John McCarthy’s critique of it. Later, I had the honor of having McCarthy chair my AI qualifying examination committee at Stanford. McCarthy and Minsky were the two fathers of the field of Artificial Intelligence, and both had deep insights into the powers as well as limitations of it, and I was lucky enough to study with both of them.
We can still see today that AI has great potential, and at the same time has significant limitations. The same challenges we were grappling with 30 years ago are still apparent today, in particular when we look at AI in the enterprise. I was inspired by the work as a student, to see if the value of AI could somehow be unlocked, and I have continued to be passionate about it.
You’ve previously written some instrumental papers, which paper do you believe was the most instrumental in evolving your views on AI?
As a student I must have read several thousand papers. McCarthy’s prescient papers on an “Advice Taker,” on some key philosophical problems of AI, Marvin’s papers on the mind as a society, on bringing together the connectionist (neural network based) and symbolic approaches to AI, Judea Pearl’s papers on probabilistic reasoning and causal intelligence, and papers by David Marr (on vision), Pat Winston (on learning descriptions of objects from examples), Waldinger’s work on program synthesis, and many others shaped my views. More recently, I have been reading works by Hinton, Lecun, the attention folks, as well as the works of Cynthia Rudin, Fernanda Viegas, and others.
You’ve stated that the developer experience of building an AI system is fragmented and broken, what are some of the current issues behind building an AI system?
AI systems today can really only be explained by a relatively small number of people — statistics vary, but it seems there may really only be about 20-30,000 in the world that understand the true methods of how AI systems run. This is vastly smaller than the 52,000 or so people we estimate are MLOps professionals, or the 1 million we estimate are data scientists. Many of them could not tell you why the system is doing what it is, why it makes the recommendations it does or what could possibly run amuck, or how the underlying techniques work.
Put this against the backdrop of a vastly complex landscape. There are over 300 MLOps vendors that Gartner is tracking at any given time. Each of these have a specialized offering. The large cloud vendors on the other hand have their own flavor of everything, and often seek to lock companies into their ecosystems and their infrastructure.
Then, the compute itself is often too expensive for companies to truly build and train some of the most advanced models available. Those are left to a few companies that have the talent and the resources required to manage an AI system’s demands.
The lack of understanding, the complexity of the tooling and the cost of the compute combine to create a disjointed and challenging landscape for any company seeking to be AI proficient. At Vianai, we’re building methods to make AI easier to use and easier to understand and observe, while vastly reducing the resources and costs associated with getting the best performance.
Could you share the genesis story behind Vianai?
I had spent many years working to bring new, disruptive innovations to enterprises. My teams and I built several products that reached tens of thousands of enterprises and were considered breakthroughs. I also led two fundamental transformations in my two journeys prior to starting Vianai and participated in transformations at hundreds of enterprises. Adding to this was my many years of studying AI and focusing on how to make AI better, more relevant, and in the service of humanity.
In a somewhat unusual way – these things came together. I was on vacation with my family in Southeast Asia [in late 2018]. We were shopping in a small market, and the vendor had beautiful, hand-crafted jewelry. It was made with traditional techniques and local stones, and it was stunning, but, of course, no one outside of this small town had heard of them. And I had this question come to my mind, “What if this vendor could use AI? What would that look like? How would the systems have to operate?” At that moment it hit me that every business in the world was going to be transformed with AI, and that this transformation could not be looked at with the lenses of yesterday, but needed products and ideas that had to start from a blank slate.
About a month later, I founded Vianai with a mission to bring true, human-centered AI to businesses worldwide. This means providing products and services, applications and technologies, tools that enable business users, data scientists, ML engineers and even vendors in remote parts of the world to truly reap the benefits of AI.
Since then, we’ve created applications to help businesses get started on AI, a platform to help ML practitioners manage and monitor their AI models, and optimization techniques to enable more companies to access AI.
Through everything, we have found that the significant potential of bringing the power of human understanding, judgment, and collaboration together with data and the best AI techniques remains untapped. Based on our work with leading enterprise companies, I saw that the same techniques that would help the small vendor would help the biggest enterprises in the world.
Vianai is all about human centered AI, could you define what this is and why it’s important?
Human-centered AI is AI that seeks to amplify human work and improve human judgment. Machine learning is far too often thought of as a replacement for human labor. But AI is complementary to humans — it offers scale and repeatability and precision that humans cannot replicate. But AI cannot replicate human judgment, human experiences, or our understanding of context.
There are obvious examples of this, of AI mistaking a turtle for a rifle for example, but far more often we place too much trust in AI when it hasn’t proven itself to be trustworthy yet. An infamous story comes from a decade ago, when one firm’s AI was allowed to trade without human intervention. The algorithm lost $440 million in less than an hour.
For a more recent example, cutting-edge language models remain relatively easy to confuse or bias. Text-to-image generators are potentially powerful, but require very specific commands from a human user to get their full potential.
Human-centered AI, then, is a kind of focus in the design of our products. We bring the power of human understanding – like judgment and collaboration – together with the best data and AI techniques, to create intelligent systems that can greatly improve business outcomes and processes.
Could you explain the need for a feedback loop behind humans and AI?
There’s a whole branch of AI called “human in the loop” that relies on the feedback mechanisms of humans to naturally improve the AI’s performance. This is natural, and makes sense for any system.
AI systems can improve over time, through retraining, which incorporates whatever actions the user took. This is, of course, a part of our applications as well. Let me give an example.
Before Covid, we were working with a large, financial services firm on demand forecasting. Because of how we designed the system, when Covid came and broke so many other models, ours adjusted to the changes rapidly and never had to be rebuilt. This is the second and most important aspect of human-centered AI, designing the systems from the start to incorporate the complexities of modern life.
This creates trust and a system that grows with the organization and user.
What makes Vianai a next generation AI platform?
While there is a lot of discussion around risk, regulation, and the promise of AI, few have sought what we find to be the solution — the concept of human-centered AI.
Our platform is then ready for the problems that will come as AI becomes more real in the enterprise. It is to tackle issues around trust, bias, and transparency. It enables companies to scale AI with monitoring and optimization. And it enables non-technical users to harness AI through our applications.
What are some of the challenges behind building a platform that dramatically streamlines the experience for enterprise AI?
The biggest challenges we see in companies incorporating AI are talent, tools, and technology. First, talent tends to be concentrated in a few places, especially in larger tech companies. This makes it very hard for outside team members to participate in the oversight, governance, and shaping of the AI program and can create even more bias as only a limited number of team members are working on the operations.
Technology and tools can also be a challenge in streamlining AI. Right now, both technology and tools are limited. Chips to run AI are scarce and very expensive, and tools are locked into certain vendors which reduces the freedom to improve cost while extending value. No matter where a company may be in its enterprise AI journey, these challenges can make implementing useful and ethical AI challenging as it creates a disconnected, fragmented strategy and removes the tools necessary to execute the proper functions. Organizations need to be able to support all areas of AI from implementation to maintenance, and have the team support and offer input to make it a success.
For true success, I have found that platform capabilities need to be completely open, modular, flexible, and not dependent on costly hardware and software upgrades. And with a human-centered approach, humans are still able to bring the knowledge, context, experiences, and creativity to solving problems – this is then amplified by the AI platform, not replaced.
Is there anything else that you would like to share about Vianai?
In many ways, we are living in the times of AI. There is a lot of hype and discussion around AI, which on the whole is a good thing. We are seeing a lot of advancements and a wider adoption than in the past in areas such as Generative AI and other areas. However, we should also work to recognize the limitations of AI – the realities of AI technology today as well as the realities of the scarcity of expertise in AI, and the lack of trust in AI especially in enterprises. If we can frame AI as an amplifier of our lives, society, our work, our potential, and have the necessary oversight of AI to ensure this, then I do believe we will finally see it come to life in meaningful and transformational ways.
Thank you for the great interview, readers who wish to learn more should visit Vianai.