Wallarm CEO Ivan Novikov: “Finding your calling in the era of AI is humanity’s next big leap.”

Wallarm — security company for web threat detection

What doesn’t kill us makes us stronger. This proverb describes how Wallarm’s security system works — by becoming stronger with every attack. Ivan started thinking about implementing AI back in 2009. He saw how hackers mocked companies and their “ridiculous” security precautions that they could easily hack. As a white-hat hacker (one who specializes in testing methods to improve the security of an information system), he wanted to create a product that he could be proud of — one that no one would be able to crack. This is how he came up with the idea of a machine that could learn by itself and become increasingly more resilient.

Past approaches to security were based on analysts’ written signatures (proven methods of defense) against attacks. But such an approach is weak because analysts cannot come up with security solutions fast enough to combat the number of hackers creating new types of attacks. That’s why a self-learning system is a more effective way to fight hackers. But in 2009, there were no available algorithms that could truly support such an approach.

In October 2012, everything changed for Ivan and the rest of the world, when Fei-Fei Li, the head of Stanford’s AI Lab and the founder of the yearly ImageNet computer-vision contest, announced that two students had found a way to double the accuracy of image recognition. This was a turning point that sparked industry leaders to take AI much more seriously. Since then, the neural network system has progressed to a practical level and is being implemented in numerous products…including Ivan’s. Despite this breakthrough, the challenge for any company that wants to take a serious approach to AI is in finding relevant training data.

As Ivan says, in security there are actually a limited amount of web attacks. Ivan’s team began collecting and reproducing these attacks, but still couldn’t create a large enough dataset to begin implementing AI in their product. So they decided to conduct a competition among hackers. They found that the idea of competing with a machine and beating it was really motivating, and hackers willingly took part in the competition to generate samples of attacks — necessary data for the product that is now up to 2TB, according to Ivan. The company also collects data from attacks perpetrated upon existing clients.

While developing the product, it became obvious that the company should be in Silicon Valley. That’s why in 2015 the team moved to the tech mecca and enrolled in Y Combinator. As of today, the company has more than 100 paying clients from the banking, insurance, healthcare, and software industries, dealing with huge amounts of user data.

Ivan highlighted Cylance, AttackIQ, DarkTrace, and Lookout as other security companies with a focus on AI.

The difficulty of building an AI company

Ivan says the greatest challenge is coming up with a sustainable business model: how much money would have to be spent on hiring data scientists and engineers, and on GPUs and cloud services? All those things require careful research, and often costs are hard to predict. The real pain point is in finding data scientists.

“AI is an entirely new industry and there is a high probability that a candidate (even if he is experienced in the field) has never been faced with the tasks that need to be solved by my product. So it’s always an option to hire someone and educate them,’’ says Ivan.

Another challenge that the company faced — Wallarm’s AI is different. “We do not rely on a neural network approach in our machine learning. Instead, like in a time series, we look at streams of data,’’ he says. “This means that our main technical issue is not storing and processing huge amounts of historical data; it’s creating a new AI paradigm that does incremental learning and is trained only on new data. We managed to do just that and it helped cut costs in the cloud. We can look at threats occurring in real time, analyze them, and make predictions. The incremental learning algorithm was the main challenge for our business, aside from employees asking for raises,’’ smiles Ivan.

Getting customers to deploy and realize the full potential of the product within their environment has also proven to be a challenge. Left to their own devices, customers use only a fraction of the features Wallarm provides. “That’s why we often provide onboarding and technical support to help clients learn how to fully use our product,’’ he concludes.

Visionary: What’s our future with AI?

“True AI doesn’t exist yet — that’s the most fascinating part of it,” admits Ivan. “But we are working on building the future with AI. And I truly believe that it can develop a consciousness. If AI can become self-aware (maybe we will learn how in 100 years…or our grandchildren will), it means that humanity might also potentially ‘function’ based on specific algorithms. So, it means that our present world might be a simulation.’’

‘’Nobody knows if AI can have feelings and emotions. Honestly, even now, we don’t know how neural networks work. But we know all the components of them: there are available frameworks like TensorFlow, or pre-trained neural networks, training datasets, and GPUs. But the point is when we put all those components together and receive astonishing results (much better compared to analog algorithms), we have no clue how such an outcome was achieved as we didn’t specifically program it. It just happens. AI logically classifies users’ personal data and recognizes objects in images, sounds and voices, etc. It’s truly magical.”

“I think in 100 or even 50 years people will have free transportation thanks to AI,’’ continues Ivan. ‘’The government or companies might even pay people to motivate them to use public transportation through an ad campaign. I came up with this prediction because the costs for transportation are inordinately high between energy and overhead charges. Self-driving cars don’t require additional expenditures. So there is a high probability of this happening.”

“Now, we know that AI is superficial. But our children and grandchildren will not distinguish AI from something alive. If they see a future robotic dog, they might never guess that it was created in a lab, just as we don’t give thought to how our food is grown and reaches the grocery store. Now millennials are wholly invested in computers, smartphones, and social media as authentic and integral parts of their lives, and can’t imagine life without these things. Our grandchildren will share the same attitude about AI, including personal assistants.’’

“Already today, robots are trading on the stock exchange. The stock exchange greatly defines and influences our lives. So I would argue that robots have already taken over for us. Everything will be automated in the future. And this raises a very interesting issue about choice of life path. When it’s a matter of survival (earning money for a living), people often don’t have much choice. They take a boring routine job just to earn some money to live. Imagine what life would be like if we were not bound to routine, menial jobs. Then the problem of forming an identity and finding one’s calling will become crucial. It might sound easy to just choose a profession one loves to both satisfy one’s existential and survival needs. But the truth is that finding one’s ‘true’ calling might be the real challenge once you remove the necessity of survival.”

AI profile:

Number of data scientists and engineers specifically working on AI — 3

Size of the training dataset — 2 TB

Cost for the cloud — $10K/month

Number of clients — more than 100 paying clients

Raised $2.3M seed funding

Founded in 2013