Fresh off a coming out presentation at Oracle Open World, and the raising of $50 million in funding., Vishal Sikka, founder and CEO of Vianai Systems wants to demystify and democratize AI on the way to 'amplifying humanity.' I caught up with Sikka on an hour-long call during which he explained the rationale behind creating the Vianai platform and his hopes for the future.
There is a great asymmetry in AI. Many of the the techniques have been around for a very long time but they are inherently opaque and are not transparent. And many of those same systems are dumb. An AI system might understand a picture of a dog but it has no ability to reason around what dog it is. Most engineers have no idea what's in the neural networks they're using. Out of an estimated 35 million developers and engineers, there's only about 20,000 who truly undertsand the network math. When we looked at the top 200 AI jobs around the world, only 75 had more than one open AI job. Amazon has 1,600 open jobs in Tensorflow so the divide between those who can take advanyage of AI and those that can't is startling.
Part of the problem lays in the tooling which Sikka described as 'horrible.' AI developers typically use Jupyter Notebooks to 'create and share documents that contain live code, equations, visualizations, and narrative text.' The problem is that any time a model is changed, it requires multiple changes in the notebook. This is time-consuming, clumsy and error-prone. In Sikka's words:
How do you make tools that make it dramatically easier for people to use so that millions can take advantage of AI? That was my first question. In parallel I started asking about the business value. We know governments are making use of facial recognition but in manufacturing, predictive fault detection is an obvious area, Trade failures at financial institutions cost millions of dollars each day. Then there is medical but that operates in a highly regulated environment - a showstopper.
Joe Sieczkowski, Chief Architect and Head of Data at BNY Mellon came on stage during Sikka's presentation to explain how Vianai has helped them. BNY Mellon administers or has custody of $35.5 trillion in assets and has $1.8 trillion under management across 35 countries. According to the company's 2018 report, (PDF) BNY Mellon settles approximately $2 trillion in payments each day of which 94% are fully automated. There is however a small failure to close rate where parties are obliged to pay penalties and interest. This translates into millions of dollars in cost. Sieczkowski says they are 'very pleased' with the success of a Vianai implementation that's designed to predict the likelihood of trade failures. While he was reluctant to share actual numbers, it only takes elementary school math to figure out the savings potential.
I asked Sikka how Vianai works
A primary recipient is a business person who identifies a specific problem. In the BNY Mellon case, the real issue is one of optimizing working capital which is a head of securities topic. The same broad issue of optimizing operational expenses exists in make to order manufacture. So we start there. But people are reliant on technology today so it quickly involves developers and because we're talking data, that includes data scientists. We think that a design thinking approach is the best way to bring these people together.
In demonstrating the solution, Vianai showed how it has turned the concept of the Jupyter Notebook into something where teams can easily and purposefully collaborate to optimize for a specific set of problems and then iterate solutions in real-time by changing model conditions. You can think of it as akin to rapidly running multiple A/B tests in a complex model.
The bulk of the 'stuff' under the covers is about building up the semantic model to understand what the network is doing. There are always multiple techniques that can be applied and also many dependencies.
I asked about data provenance and security since the kinds of problem under discussion mean access to highly sensitive data samples.
There is a problem with fragmentation. Microsoft, Google, AWS all have their own types of cloud but the reality is that we have to operate in a hybrid world. But then it is not for example straightforward to run an on-premise Azure stack. We are platform agnostic so that you can build models wherever you are but keep the underlying data safe.
Sikka says the company has five customers already and that the $50 million seed round should mean the company achieves profitability without financial constraints. At the same time though there is a lot of work to do.
For us to get this into the hands of millions of people we have to make the models understandable. We can do that but it means running masterclasses at customers, showing as well as doing. That's an education problem about which I am passionate and which I think can scale.
I've known Sikka for many years and he has always had a passion for both education and the ability of technology to 'amplify humanity.' Positioning AI as an aid to business problem finding and solving is key. It removes some of the mystique that bedevils conversations on this topic.
On a first look, Vianai has significant potential but then there are many issues that remain unsolved. For example, the company will have to be very clear about the ethical usage of its technology and the underlying math in AI models. It still has some way to go in terms of fleshing out the platform. The good news is that Vianai is starting with a clean slate with no legacy in processes, people and technology. It also has a world-class advisory board which should help keep the company on a (relatively) straight path.