Interview with Michał Iwanowski, VP of Engineering, about AI development in 2021.

How do you rate 2020’s AI development?

AI is no longer a buzzword. Increasingly, from the R&D phase, it moves towards proven solutions, and nowhere more so than in computer vision and natural language processing.

The computer vision projects deepsense.ai delivered in 2020 prove that numerous industries can benefit from advanced image analysis technology. We have implemented successful projects in modern manufacturing that incorporated machine learning throughout the production process. These included not only intelligent quality control systems, but also safety monitoring and workplace security solutions. Our 2021 projects will develop this potential based on AIoT systems.

NLP-based solutions are likewise becoming more widely applicable. Our portfolio has been enriched by projects ranging from virtual meeting assistants to solutions capable of summarizing large chunks of text and evaluating the quality of student submissions in selected areas of study. With the rise of enormous language models such as GPT-3, we expect to see this area of AI –  including natural language understanding (NLU) – grow more robustly in the coming quarters.

Where can we look for indicators of AI trends for 2021?

Top global research and consulting companies have indicated that 2021 is to be intensive and promising for AI. The emerging trends include the democratization of AI and industrialization of AI platforms. More and more businesses are seeing the real value in analyzing data and are convinced that using AI based on platforms that enable scalability and safety is a wise step forward.

The Covid-19 pandemic appears to have accelerated customers’ openness to AI. Despite the crisis, clients have continued to invest in the technology. They understand that in order to build competitive advantage, AI must become a core element of their business models. With more than 70 data science and machine learning experts on board, we are ready to turn clients’ vision into practical AI-focused solutions.

Do you notice any changes in your customers’ approach to AI solutions?

We work mainly with clients who already have some knowledge of and experience with AI projects. This allows us to act creatively on new AI solutions for various businesses.  When a client retains us for a single project, we frequently uncover several new areas where we can further develop and implement AI.

We highly value clients who trust us enough to jointly take on challenges related to innovative AI applications. One such client is Volkswagen, with whom over the past 18 months we have moved neural network policy from a car simulator to reality. After multiple training sessions in the simulated environment, our algorithm overcame the  sim-to-real gap in the real world. This milestone has opened up the possibility of using distributed computing to train AI models that are easily transferable to physical vehicles and can be tested in real world-environments.

We have also observed increased interest in our R&D projects. Many clients want to get involved in our research work because it leads them to explore new opportunities for their businesses. After a successful reinforcement learning project conducted together with Google Brain, the University of Warsaw and the University of Illinois at Urbana-Champaign, we continue our work in this area. The research conclusions bring tangible benefits to our clients by setting solid algorithmic foundations for tackling business problems that don’t present an obvious objective function, but rather require an AI agent to learn from its mistakes. This paradigm shift opens up brand new areas of application for AI.

Last but not least, we see many clients that are ready to build AI capabilities in-house. With our comprehensive training programs, they get best-in-class know-how that bears long-term results for real-life data challenges.

How do you sum up deepsense.ai’s plans for 2021 ?

Apart from pursuing new projects and challenges in computer vision, natural language processing and predictive analytics, there are two crucial tasks ahead of us in the upcoming year.

The first is to maintain an open-minded and creative approach to the changing reality of remote work and the migration of more and more services to the virtual environment – with a focus on how these transitions can be converted into business opportunities using AI and automation.

The second challenge is to bridge the gap between cutting-edge advancements on the frontier of AI and Machine Learning and reliable, well-tested business solutions. As a company that operates at the crossroads of business and research, we believe it’s on us to take a pioneering role in this challenge.