Interview about the impact of AI on business and society with Leo Ma (Deloitte AI Institute)
This week's interview is with Dr. Leo Ma, Dean of Deloitte AI Institute
Dr. Pui Wai (Leo) Ma currently serves as the Dean of the Deloitte AI Institute (Hong Kong), where he is responsible for the strategic development of AI applications. He was awarded Doctor of Philosophy from the Department of Electronic and Information Engineering, The Hong Kong Polytechnic University, in 2008. He has authored numerous research papers that have been published in top-tier journals and is a regular speaker and presenter at leading international conferences. His expertise encompasses a wide range of scientific topics, and he employs both conventional and cutting-edge machine learning methods in his research.
Could you explain the difference between Gen AI and AI? Which one would you say is more beneficial to tackle current challenges?
The key difference between Generative AI and Traditional AI lies in their nature: probabilistic vs deterministic. Traditional AI systems are typically deterministic, meaning they produce the same output for a given input. For example, models such as decision trees and support vector machines, follow fixed mathematical rules. As a result, traditional AI is more beneficial for specific, well-defined tasks requiring consistency predictability and accuracy, like medical diagnostics, fraud detection and supply chain optimization.
In contrast, Generative AI is inherently probabilistic and often non-deterministic. It generates new content, such as text, images, or music, by sampling from learned probability distributions. During inference, randomness is introduced through techniques like top-k sampling or temperature scaling, which allows for variability and creativity in its outputs. For instance, a large language model might generate different responses to the same input due to this random sampling. This probabilistic nature enables Generative AI to excel in tasks that require adaptability and creativity. For instance, in education or customer service, Gen AI can personalize interactions, while in drug discovery, it can explore vast possibilities for new compounds. However, it also means that outputs can be unpredictable.
In many cases, combining both approaches can provide the most robust solutions, leveraging the strengths of each to address modern challenges effectively.
What is the biggest opportunity and the biggest threat of AI?
The biggest opportunity of AI lies in its ability to make processes, systems, and workflows significantly more efficient. By automating repetitive tasks, analysing vast amounts of data, and optimizing decision-making, AI has the potential to revolutionize industries and free up human resources for more impactful work. For example, AI can streamline manufacturing through predictive maintenance, optimize logistics to reduce carbon footprints, and enhance healthcare by quickly diagnosing diseases or tailoring treatments to individual patients. Moreover, generative AI enables creativity at scale, producing content, designs, and ideas that can complement human ingenuity. This efficiency and productivity boost can lead to economic growth, improved quality of life, and innovations that solve complex global challenges.
However, one of the biggest threats of AI is the risk of over-reliance, which could lead to a loss of human creativity and critical thinking. As AI becomes more capable, there’s a danger that people may defer too much to machines, sidelining their own judgment and innovative potential. Additionally, if people trust AI systems without fully understanding their limitations or biases, it could lead to poor decision-making or reinforce existing inequalities. This over-reliance risks creating a world where humans become passive consumers of AI outputs, rather than active participants shaping their future.
AI should be treated as a tool to enhance human creativity and decision-making, not replace it. Investing in education, ethical AI development, and fostering a culture of critical engagement with technology will allow us to benefit from AI’s efficiency without losing our unique human capabilities.
What has been the main issue that you have dealt with at Deloitte AI Institute concerning AI / cyber security?
Navigating the rapid emergence of new AI models, tools, and algorithms can be overwhelming. However, this constant influx of innovations also presents a critical opportunity to refine our strategy and differentiate our firm. To succeed in such a fast-paced environment, it is essential to focus on strategic prioritization, value-driven development, and long-term differentiation.
We prioritize our development efforts by aligning them with business impact and client needs. Rather than trying to keep pace with every new tool or algorithm, we focus on areas where AI can provide measurable value for our clients. This involves identifying key industries or sectors where Deloitte has a competitive advantage and tailoring AI solutions to address their most pressing challenges. For instance, industries such as finance, insurance, and government and public sectors are ready for AI-driven innovation.
We concentrate on building our own intellectual property (IP) around AI. While adopting external tools or models can be beneficial, long-term differentiation lies in developing proprietary frameworks, solutions, or methodologies that clients can only access through Deloitte. This involves integrating existing cutting-edge AI tools into tailored solutions or creating unique systems tailored to specific client needs. By combining our expertise in consulting with AI-driven insights, we position the Deloitte AI Institute as a leader in actionable and business-first AI solutions. Additionally, we emphasize sustainability and ethical AI development. As clients become increasingly concerned about the risks of bias, lack of transparency, or over-reliance on AI, embedding ethical practices and explainability into our strategies can help establish trust and credibility in the competitive landscape.
By focusing on value creation through proprietary IP development and ethical considerations in artificial intelligence implementation at the Deloitte AI Institute, we craft an effective strategy that leverages the strengths of Deloitte while standing out in a crowded market.
What are the most important and influential developments within the field of AI?
The AI landscape is undergoing a significant transformation. A key milestone in this journey is the emergence of multimodal models, which seamlessly integrate and process diverse data types, including text, images, audio, and video. Pioneering models like OpenAI's GPT-4 and Google's DeepMind's Gemini are enabling more sophisticated interactions with complex real-world scenarios. This is having a profound impact on areas such as content creation, virtual assistants, and medical diagnostics.
In parallel, the rise of small open-source AI models is democratizing access to AI technology. Initiatives have shown that these compact models can deliver impressive performance while being computationally efficient. By fine-tuning these models, organizations with limited resources can now participate in the AI ecosystem, driving innovation across industries.
Another area where AI is making a significant impact is robotics. The integration of advanced AI algorithms with robotics is revolutionizing sectors such as manufacturing, healthcare, and space exploration. AI-powered robots can adapt to dynamic environments, learn from experience, and make decisions in real-time, greatly enhancing their utility and versatility. Examples of this convergence include autonomous vehicles, robotic surgeries, and disaster response systems.
What is your view on the environmental impact of AI?
The environmental impact of AI, particularly regarding energy consumption and resource use, has become a critical concern as the field continues to grow. Training and deploying large-scale AI models require vast computational resources, often resulting in significant energy consumption. For instance, training state-of-the-art models like GPT or other large language models demands immense electricity and high-performance hardware, leading to substantial carbon emissions. Data centres that power these models contribute to global energy demand, and their reliance on non-renewable energy sources exacerbates environmental challenges. The environmental footprint of AI is further intensified by the production and disposal of specialized hardware, such as GPUs and TPUs, which require rare earth metals and create electronic waste.
Beyond training, the widespread deployment of AI systems in consumer devices, cloud-based services, and autonomous technologies contributes to ongoing energy consumption. While AI applications often bring efficiency gains in industries like transportation and energy management, the net impact can be offset by the additional demand for computational power.
Addressing these challenges requires a multifaceted approach. Researchers and organizations are exploring energy-efficient AI algorithms, model compression techniques, and distributed training systems to reduce computational costs. Furthermore, the adoption of zero-carbon energy sources, such as wind, solar, nuclear, and fusion energy, to power data centres and innovations in hardware design are critical for minimizing resource use. Open-source models and smaller, fine-tuned architectures also demonstrate potential for balancing performance with lower energy requirements. Collaborative efforts across academia, industry, and government are essential to mitigate the environmental effects, enabling AI to advance responsibly while minimizing harm to the planet.
During your presentation you talked about humanoid robots, why do you think people have such a fascination for this?
People are fascinated by humanoid robots because they represent a vision of machines that seamlessly integrate into human environments and routines. One of the primary reasons for this fascination is that humans live in a three-dimensional world designed for human proportions, capabilities, and ergonomics. Everything from door handles to furniture, vehicles, and tools has been built with the human form in mind. A humanoid robot, with its human-like appearance, movement, and dexterity, is inherently well-suited to navigate and operate within these spaces without requiring drastic redesigns of the environment.
Furthermore, the human-like appearance of these robots fosters a sense of familiarity and comfort, making it easier for people to accept and interact with them. This is particularly valuable in caregiving and companionship roles, where emotional connection and trust are crucial. Humanoid robots also symbolize technological progress and the aspiration to create machines that mimic human intelligence and behaviour, fuelling public curiosity and engagement.
How is AI used in daily life?
Artificial intelligence has been integrated into daily life, improving productivity, creativity and convenience. LLM chatbots are used as valuable tools for answering questions, brainstorming ideas, writing reports, and translating text between languages. Interaction with chatbots not only saves time but also inspires new perspectives and simplifies complex tasks.
Entertainment and home automation are significantly impacted by artificial intelligence. The speech recognition system on the TV can issue voice commands through the built-in assistant. Social media platforms use artificial intelligence to curate personalized content and advertising tailored to individual interests. Online search results are optimized to recommend music playlists or travel destinations and interactions become more relevant and personalized.
Additionally, AI-powered financial systems monitor transactions for suspicious activity, improving financial security and well-being. These systems provide a layer of protection and peace of mind, proactively detecting unusual patterns to keep your financial accounts safe. If there is potential fraud, an alert will be sent.