Reading time: 2 min

Interview: The future of artificial intelligence is products and services based on lifelong learning

Dr. Pankaj Gupta and Dr. Korbinian Spann talk about the future of Natural Language Processing.

Dr. Pankaj Gupta earned his PhD in Computer Science with a focus on Deep Learning and NLP (Natural Language Processing). He was a senior scientist at Siemens for six years and published more than 15 patents and 20 publications. He founded his own startup, DRIMCo GmbH, in Munich. For, Pankaj works as a “Technology and Science Advisor.”

Korbinian: Why do you think lifelong learning will be a crucial step for artificial intelligence?

Pankaj: Lifelong learning is critical in the age of artificial intelligence. And it’s one of the topics I’m very involved with. Like us humans, this system is constantly gathering new knowledge, using past knowledge for future learning, and minimizing “forgetting” past learning.

Classical machine learning systems are built in isolation, meaning they run on a specific data set for a specific task and therefore do not store and accumulate learned knowledge over time. And because of this forgetting (catastrophic forgetting), these systems cannot model tasks sequentially.

In comparison, lifelong learning systems are designed to retain and reuse interconnected knowledge over a lifetime. An example would be: If I learned to formulate a sentence in English, that doesn’t mean I would forget the letters of the alphabet. I would still remember them, and in fact I use the prior alphabetic knowledge to formulate a sentence. Humans are lifelong learners as a matter of course, but it’s not trivial for machines, so lifelong machine learning systems are critical for general artificial intelligence.

Korbinian: What might a product look like in this respect?

Pankaj: An AI product with lifelong learning capabilities would involve continuous learning, for example, with “human-in-the-loop” or active learning, where humans guide AI systems.

The goal is to provide continuous feedback, and the AI continues to learn with that continuous feedback. And essentially, lifelong learning systems employ transfer learning and “domain adaptation” techniques to provide users with a better AI experience without forgetting what they’ve learned.

At Insaas, when we perform an analysis of customer reviews, the AI lifelong learning product will continuously learn about streams of customer feedback from social media platforms. And this means that over time, the system will learn about streams of customer reviews from different domains and also reuse and retain past learning experiences across domains.

Korbinian: Where do you see the biggest opportunities in natural language processing today?

Pankaj: Since NLP is constantly evolving, I believe there are endless opportunities. Currently, there is lots of basic and cutting-edge research going on at a tremendous pace. In contrast, there is a huge vacuum in practical applications, especially in the industrial area of analysis of texts.

Therefore, I see a great opportunity in using cutting-edge AI research to build high-quality production-ready NLP applications. Especially for different industries, it has some challenges, for example, the lack of training sets, domain shifts, or affordable infrastructure such as GPUs.

BERT models are one example. They are successfully used in the scientific community, but their application in production environments is still a challenge because their inference is still very slow and computationally intensive. And it is challenging to deploy large NLP models in customer infrastructure.

I see another good opportunity in building linked NLP applications for lifelong learning. At DRIMCo, we set that up for any AI functionality because we respect the ownership and privacy of customer data. The goal is to build a decentralized AI system for lifelong learning where the client data stays and the algorithms and predictive models are shared between the clients and the server. The purpose is cross-client transfer, learning from a “phantom” pool of real-world data for a better customer AIX (AI Experience) without actually moving the customer data. provides a software platform for building customer-centric services and products based on data. Our vision is to work with companies to make the transition from standardized mass products to personalized products and services: environmentally friendly, efficiently and automatically produced, meaningful to employees. To aggregate all customer voices in different languages and topics, we are working with with an artificial intelligence that learns “lifelong” (lifelong learning). Over time, this solution will be able to understand human sensibilities and preferences better and better.

Share article
Dr. Korbinian Spann
27. October 2020