Due Diligence

“That ability that humans have, it’s the combination of creativity and abstraction. I do not see today’s AI or tomorrow’s AI being able to do that yet.” – Dr. Fei-Fei Li – Stanford Professor – world-renowned authority in artificial intelligence

Dr. Li’s statement came amid wide speculation about the near-term prospects for artificial general intelligence (AGI) and superintelligence. While current AI already exceeds human capacity in specific domains (such as language translation, memory recall, and vast-scale data analysis), Dr. Li draws a line at creative abstraction—the human ability to form new concepts and theories that radically change our understanding of the world. She underscores that, despite immense data and computational resources, AI does not demonstrate the generative leap that allowed Newton to discover classical mechanics or Einstein to reshape physics with relativity. Dr. Li insists that, absent fundamental conceptual breakthroughs, neither today’s nor tomorrow’s AI can replicate this synthesis of creativity and abstract reasoning.

About Dr. Fei-Fei Li

Dr. Fei-Fei Li holds the title of Sequoia Capital Professor of Computer Science at Stanford University and is a world-renowned authority in artificial intelligence, particularly in computer vision and human-centric AI. She is best known for creating ImageNet, the dataset that triggered the deep learning revolution in computer vision—a cornerstone of modern AI systems. As the founding co-director of Stanford’s Institute for Human-Centered Artificial Intelligence (HAI), Dr. Li has consistently championed the need for AI that advances, rather than diminishes, human dignity and agency. Her research, with over 400 scientific publications, has pioneered new frontiers in machine learning, neuroscience, and their intersection.

Her leadership extends beyond academia: she served as chief scientist of AI/ML at Google Cloud, sits on international boards, and is deeply engaged in policy, notably as a special adviser to the UN. Dr. Li is acclaimed for her advocacy in AI ethics and diversity, notably co-founding AI4ALL, a non-profit enabling broader participation in the AI field. Often described as the “godmother of AI,” she is an elected member of the US National Academy of Engineering and the National Academy of Medicine. Her personal journey—from emigrating from Chengdu, China, to supporting her parents’ small business in New Jersey, to her trailblazing career—is detailed in her acclaimed 2023 memoir, The Worlds I See.

Remarks on Creativity, Abstraction, and AI: Theoretical Roots

The distinction Li draws—between algorithmic pattern-matching and genuine creative abstraction—addresses a foundational question in AI: What constitutes intelligence, and is it replicable in machines? This theme resonates through the works of several canonical theorists:

• Alan Turing (1912–1954): Regarded as the father of computer science, Turing posed the question of machine intelligence in his pivotal 1950 paper, “Computing Machinery and Intelligence”. He proposed what we call the Turing Test: if a machine could converse indistinguishably from a human, could it be deemed intelligent? Turing hinted at the limits but also the theoretical possibility of machine abstraction.
• Herbert Simon and Allen Newell: Pioneers of early “symbolic AI”, Simon and Newell framed intelligence as symbol manipulation; their experiments (the Logic Theorist and General Problem Solver) made some progress in abstract reasoning but found creative leaps elusive.
• Marvin Minsky (1927–2016): Co-founder of the MIT AI Lab, Minsky believed creativity could in principle be mechanised, but anticipated it would require complex architectures that integrate many types of knowledge. His work, especially The Society of Mind, remained vital but speculative.
• John McCarthy (1927–2011): While he named the field “artificial intelligence” and developed the LISP programming language, McCarthy was cautious about claims of broad machine creativity, viewing abstraction as an open challenge.
• Geoffrey Hinton, Yann LeCun, Yoshua Bengio: Fathers of deep learning, these researchers demonstrated that neural networks can match or surpass humans in perception and narrow problem-solving but have themselves highlighted the gap between statistical learning and the ingenuity seen in human discovery.
• Nick Bostrom: In Superintelligence (2014), Bostrom analysed risks and trajectories for machine intelligence exceeding humans, but acknowledged that qualitative leaps in creativity—paradigm shifts, theory building—remain a core uncertainty.
• Gary Marcus: An outspoken critic of current AI, Marcus argues that without genuine causal reasoning and abstract knowledge, current models (including the most advanced deep learning systems) are far from truly creative intelligence.

Synthesis and Current Debates

Across these traditions, a consistent theme emerges: while AI has achieved superhuman accuracy, speed, and recall in structured domains, genuine creativity—the ability to abstract from prior knowledge to new paradigms—is still uniquely human. Dr. Fei-Fei Li, by foregrounding this distinction, not only situates herself within this lineage but also aligns her ongoing research on “large world models” with an explicit goal: to design AI tools that augment—but do not seek to supplant—human creative reasoning and abstract thought.

Her caution, rooted in both technical expertise and a broader philosophical perspective, stands as a rare check on techno-optimism. It articulates the stakes: as machine intelligence accelerates, the need to centre human capabilities, dignity, and judgement—especially in creativity and abstraction—becomes not just prudent but essential for responsibly shaping our shared future.