Stevan Harnad on Symbol Grounding and the Morality of Suffering

In this two-part interview Stevan Harnad¹ examines his foundational concept of the Symbol Grounding Problem² and its direct relevance to modern artificial intelligence. Across the conversation, Harnad unpacks why current AI paradigms fall short of true understanding and transitions into the deeper ethical implications of sentience.

Part 1: The Core Problem and the Illusion of Meaning

The Chinese Dictionary Metaphor: Harnad explains the essence of the Symbol Grounding Problem by comparing it to learning Chinese using only a Chinese-to-Chinese dictionary. No matter how many definitions you read, you are trapped in an infinite loop of meaningless symbols unless you have a way to connect those symbols to the real world.

The Limits of Large Language Models (LLMs): You discuss how systems like GPT-3, despite their fluency, are ultimately just massive, ungrounded symbol manipulators. They do not possess intrinsic meaning; instead, their “meaning” is parasitic. They steal semantics from the massive corpus of human text they were trained on, and the meaning only actually exists in the mind of the human reading the output.

The Robotic Turing Test (T3): Harnad argues that a purely text-based Turing Test (T2) is entirely insufficient for producing a mind. For symbols to be truly grounded, an AI must be embodied. It must pass the T3 – a robotic Turing test requiring sensorimotor capacities to perceive, interact with, and navigate the physical environment just as a human does.

Part 2: Categorical Perception and the “Hard Problem”

Categorical Perception: Harnad delves into how cognitive agents build meaning by carving the world into distinct categories. True understanding requires an agent to physically interact with the world, discriminate between different stimuli (like learning to visually distinguish a predator from prey), and attach symbolic labels to those grounded sensorimotor experiences.

Doing vs. Feeling: The conversation bridges the gap between mechanical performance and consciousness. Harnad distinguishes between a machine simply doing things that look intelligent and a being actually feeling anything. He maintains that true meaning and understanding are inextricably linked to sentience.

The Ethical Imperative (Animal Sentience): Because Harnad ties the mystery of meaning so closely to sentience (the capacity to feel and suffer), he brings the discussion to his strong advocacy for animal rights. He challenges the listener to worry less about the hypothetical suffering of ungrounded AI code and focus instead on the undeniable, ongoing suffering of highly sentient non-human animals in the physical world.

The Morality of Suffering

Stevan Harnad identifies as a “negative utilitarian,” arguing that the only true moral imperative is the reduction of suffering. They reject classical utilitarian calculations that try to balance suffering against happiness, arguing that pain and pleasure are entirely separate, incommensurable continuums (using the sensation of hot and cold as an analogy).

Animal Rights vs. Human Exceptionalism

Aligning with philosopher Jeremy Bentham, the speaker asserts that a being’s capacity to suffer – not its rationality or intelligence – is what dictates moral consideration. They strongly condemn eating animals, describing it as inflicting “needless suffering” merely for human gustatory pleasure, and they reject any justification based on human superiority.

The “Other Minds” Problem and Sentience

The interview touches on the philosophical skepticism of knowing if others actually have minds. The speaker argues this skepticism is dangerous when applied to non-human animals. Because humans evolved as mammals that need to “mind-read” their young and peers to survive, we are biologically well-equipped to recognise sentience in other creatures.

• Bivalves vs. Plants: The speaker argues bivalves (like oysters) likely feel pain because they possess nervous systems and can move to escape noxious stimuli. Conversely, they argue it is highly improbable that plants feel pain, as they lack nervous systems and mobility, meaning pain would serve no evolutionary purpose.

Personal Journey to Veganism

Stevan shares that they became a vegetarian at age 17 after a doctor assured them it wouldn’t harm their health. After nearly 50 years as a vegetarian, they attended an animal law symposium that exposed the inherent cruelty of the dairy industry, prompting an immediate switch to veganism. They express deep regret for not making the change sooner and emphasise that the transition was incredibly easy, noting that their palate completely adapted to make plant-based foods taste far more vibrant.

1. Stevan Harnad is a cognitive scientist and open-access advocate known for formulating the symbol grounding problem in artificial intelligence. ↩︎
2. The symbol grounding problem is the challenge of connecting abstract symbols, words, or computer code to real-world physical meanings rather than defining them using other meaningless symbols. ↩︎