Q&A with David Bader, NJIT Magazine, Spring 2025

Q&A with David Bader, Distinguished Professor, Data Science

AI and Society

A conversation with Evan Koblentz and David Bader

Q: A lot of people who are information workers are afraid that AI will make their careers obsolete. Technological progress can’t be stopped, so how should people adapt?

A: In the face of technological progress, particularly with the rapid advancement of AI, it’s understandable that information workers may feel apprehensive about the future of their careers. However, rather than viewing AI as a harbinger of obsolescence, it’s crucial to see it as a catalyst for evolution and innovation in our work practices. The key to adapting is in embracing these technologies, learning to work alongside them, and leveraging their capabilities to enhance our own skill sets and productivity. The first step in this adaptation process is to cultivate a mindset of lifelong learning. As AI and other technologies continue to evolve, so too must our skills and knowledge. This means staying informed about new technologies, seeking out educational opportunities, and being open to acquiring new competencies that complement the capabilities of AI.

For instance, developing skills in data literacy, AI ethics and understanding the principles of machine learning can make workers more versatile and valuable in an AI-integrated workplace. Additionally, it’s important to focus on the uniquely human skills that AI cannot replicate, such as creativity, emotional intelligence and critical thinking. By honing these abilities, workers can ensure they remain irreplaceable components of the workforce, capable of tasks that require a human touch, from complex decision-making to empathetic interactions with customers or clients.

Q: Other than creative prompt-making, what should non-programmers learn now about AI?

A: For non-programmers looking to delve deeper into AI, understanding the ethical implications and societal impacts of AI is paramount. It’s important to be aware of how AI decisions are made, the potential biases in AI systems, and the ethical considerations of AI use. Additionally, developing data literacy is crucial, as it enables individuals to evaluate AI outputs and understand the importance of data quality and biases. A basic grasp of AI and machine learning concepts, even without programming skills, can demystify AI technologies and reveal their potential applications. Staying informed about AI advancements across various sectors can also inspire innovative ideas and foster interdisciplinary collaborations. By focusing on these areas, non-programmers can contribute meaningfully to the AI conversation and its future direction.

Q: There’s a popular sci-fi plot where the computers get so smart that people lose control. The new class of user-friendly AI is certainly making people excited but also nervous. Should we be afraid?

A: The emergence of user-friendly AI technologies has indeed brought this conversation into the mainstream, highlighting the balance we must strike between harnessing the benefits of AI and addressing valid concerns about its implications. It’s critical to recognize that the AI technologies we’re creating today are built with numerous safeguards, are subject to ethical guidelines, and operate within evolving regulatory environments. These measures are designed to ensure AI systems augment human abilities and decision-making, rather than supplanting or undermining human control.

While it’s natural to harbor concerns about the rapid progression of AI, allowing fear to dominate the discourse would be a disservice to the potential benefits these technologies can offer. Instead, this moment calls for proactive engagement with AI, an investment in understanding its inner workings, limitations and the ethical dilemmas it presents. By advocating for responsible AI development, emphasizing education and promoting transparency, we can foster an environment where AI serves as a tool for societal advancement. This approach ensures that we remain at the helm of AI’s trajectory, steering it towards outcomes that uplift humanity rather than scenarios that fuel dystopian fears.

Q: You and your peers at the Institute for Data Science (IDS) are known for researching the building blocks and tools that help make AI infrastructure possible. Specifically, what would you say are IDS’ most important contributions so far?

A: The Institute for Data Science at NJIT has made groundbreaking contributions to graph analytics and high-performance computing through the development of Arachne, a sophisticated and open-source framework for processing massivescale graphs. At its foundation, Arachne implements a hybrid edge list and adjacency structure that revolutionizes how large-scale graphs are processed. This architectural innovation represents a significant leap forward in handling the complexities of large-scale network analysis.

Beyond its technical architecture Arachne has demonstrated remarkable versatility in real-world applications. In cybersecurity, it enables rapid detection of emerging threat patterns through its pattern matching capabilities. The framework’s ability to track community structure and identify anomalies has proven valuable for social network analysis, while its high-performance processing has enhanced financial fraud detection systems. These advances in graph processing highlight IDS’ broader contributions to high-performance computing.

The significance of these contributions extends beyond their immediate applications. As artificial intelligence systems increasingly rely on graph-based representations for processing complex relationships, the optimizations and algorithms developed at IDS, particularly through Arachne, have become fundamental to making these systems more practical and scalable. Their work continues to bridge the gap between theoretical computer science and practical applications, enabling the next generation of AI infrastructure through innovative approaches to graph processing and high-performance computing.

Q: Some of your own research focuses on democratizing supercomputing power. Can that help lead to another approach for AI equal access?

A: The concept of democratizing supercomputing offers intriguing possibilities for expanding AI access. When we consider how democratized supercomputing could influence AI development, several key pathways emerge. Fundamentally, the process of making high-performance computing more accessible to diverse researchers and institutions, rather than concentrating it among elite organizations, could reshape how AI capabilities develop and spread.

As supercomputing becomes more widely available, smaller organizations and independent researchers gain the ability to train and run AI models without massive capital investments in dedicated hardware. This democratization creates opportunities for innovation from previously excluded participants in the AI development landscape.

However, the path to democratized AI through supercomputing faces several significant challenges. Computing power, while crucial, represents just one element in the complex ecosystem of AI development. Equal consideration must be given to data access, technical expertise and algorithmic innovation. The environmental impact of distributed supercomputing systems requires careful assessment, particularly regarding energy consumption. Additionally, any distributed computing approach to AI development must address robust security and privacy protections.

This intersection of democratized supercomputing and AI access highlights broader questions about how we can make artificial intelligence technology more equitable and accessible while maintaining necessary safeguards and standards.

Q: How does supercomputer democratization impact the overall work of the Institute for Data Science?

A: The rise of user-friendly artificial intelligence systems like ChatGPT marks a pivotal moment in our pursuit to democratize data science and supercomputing. For my work, this evolution serves as both a tool and a testament to the power of making complex computational capabilities accessible to a broader audience. It enriches the palette of methodologies and technologies at our disposal, enabling us to tackle more ambitious projects with greater efficiency and creativity. By integrating these AI systems into our research and educational programs, we’re not just enhancing our ability to process and analyze data, we’re also empowering students and researchers with the means to innovate and explore new horizons in data science without being hindered by the technical complexities that once acted as barriers. For the Institute for Data Science, the impact of such AI systems is transformative. They serve as a bridge between advanced computational technologies and a diverse range of disciplinary domains, facilitating interdisciplinary research and collaboration.

Q: A new model from China, called DeepSeek, seems to be as good as Western models but has far lower costs and technology requirements. How did they do it, and what can Western companies learn from this?

A: DeepSeek’s emergence represents a significant challenge to established thinking about AI development. While Western companies have typically pursued AI advancement through massive computational resources and extensive funding, DeepSeek has demonstrated that remarkable results can be achieved through more efficient methods and careful engineering. The company’s approach centers on targeted reinforcement learning focused specifically on reasoning tasks, rather than the broader supervised learning methods common in Western models. They’ve developed an innovative rulebased reward system that actually outperforms traditional neural reward models, while using significantly fewer resources. Perhaps most impressively, they’ve managed to compress advanced capabilities into relatively small models, achieving with 1.5 billion parameters what others do with far larger models.

The financial implications are striking. DeepSeek developed their R1 model for less than $6 million, a fraction of the hundreds of millions typically spent by Western competitors. They’ve translated this cost efficiency into their pricing model, offering services at $0.55 for input and $2.19 for output per million tokens, substantially undercutting market rates while maintaining comparable quality.

This success suggests that the future of AI development might lie more in clever engineering and efficient methodology than in raw computational power. It challenges the assumption that advanced AI development requires massive resources and suggests that innovative approaches to training and model architecture might be more important than sheer scale. The success of DeepSeek also suggests that competitive advantage in AI might come from unexpected directions, and that the barriers to entry for significant AI advancement might be lower than previously thought. It’s a reminder that technological breakthroughs often come not from doing things bigger, but from doing them smarter.

- Evan Koblentz

https://magazine.njit.edu/