TLDR: Despite rapid advancements, artificial intelligence continues to grapple with the fundamental challenge of acquiring human-like common sense. Experts argue that current AI, primarily based on deep learning, functions as a pattern matcher lacking the architectural innovations seen in biological intelligence for world modeling, simulation, and understanding intentions. This inherent limitation prevents AI from performing seemingly simple tasks that humans accomplish effortlessly, highlighting a significant gap between current capabilities and true general intelligence.
Artificial intelligence, while demonstrating remarkable prowess in specialized domains, remains profoundly ‘dumb’ when it comes to common sense – a capability humans wield effortlessly. This is the central thesis explored by artificial intelligence entrepreneur and neuroscience researcher Max Bennett in his book, ‘A Brief History of Intelligence: Why The Evolution Of The Brain Holds The Key To The Future Of AI’. Published on October 25, 2025, the article in the Times of India highlights Bennett’s argument that understanding the evolution of biological intelligence is crucial for guiding the path toward more advanced AI.
Bennett identifies five major breakthroughs in the evolution of intelligence that are critical, each representing a fundamental capability: steering (600+ million years ago), reinforcement learning (500+ million years ago), simulating (100+ million years ago), mentalising (30+ million years ago), and language (100,000+ years ago). Current AI systems, largely built on deep learning networks, excel at pattern recognition and optimization, a feat akin to advanced reinforcement learning. However, they struggle significantly with ‘simulating’ – the capacity to create internal models of the world and mentally run them to imagine outcomes before acting. This bottleneck is further exacerbated by their inability to ‘mentalise,’ or understand others’ intentions, a cornerstone of social intelligence.
‘Current AI is only deep learning networks that are essentially pattern matchers without separate systems for world modelling and simulation of futures,’ states the Times of India article. The transition from reinforcement learning to simulation in biological evolution was not merely about adding more neurons but involved qualitatively different brain structures, such as the neocortex for rich sensory models and the hippocampus for cognitive maps and episodic simulations. Scaling up current AI approaches, Bennett suggests, may not be sufficient; architectural innovations are needed, mirroring the 100+ million years of evolutionary tinkering that humans are attempting to replicate within decades.
Beyond the lack of common sense, other significant limitations of AI in 2025 include a ‘lack of explainability,’ where most models cannot articulate their decision-making processes, hindering trust in critical sectors. AI also struggles with ‘genuine creativity,’ often producing derivative content based on existing human works rather than exhibiting true inspiration. Furthermore, AI lacks ’emotional intelligence and empathy,’ making it ineffective in roles requiring a human touch, such as therapy or caregiving. These systems are also ‘vulnerable to adversarial attacks’ and prone to ‘bias in training data and algorithmic discrimination,’ leading to irrational choices and reinforcing existing societal prejudices. A recent study revealed that AI systems like GPT-3.5 and GPT-4 can exhibit human-like reasoning flaws, including biases and irrational choices, in nearly half the cases tested.
Also Read:
- Navigating the AI Landscape: Why Businesses Must Diversify Beyond Single Models and Avoid Over-Reliance
- AI’s Transformative Impact: The Top Trends of 2025 Reshaping Daily Life
In essence, while AI excels in speed and data processing for narrow tasks, it fundamentally lacks the contextual understanding, abstract thinking, and ethical reasoning that define human intelligence. This means AI will likely continue to complement, rather than fully replace, human roles, necessitating ongoing human oversight and a focus on responsible AI development.
