Dartmouth Summer Research Project on Artificial Intelligence
John McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon are in the midst of formalizing the concept of 'artificial intelligence' as a distinct field of study during the Dartmouth Summer
Setting
A large, wood-paneled seminar room in Dartmouth College's mathematics department, filled with blackboards covered in equations and diagrams. The room is arranged with rows of wooden chairs facing a central table where the organizers sit. Open windows let in a summer breeze, and the hum of an early computer can be heard from a nearby lab.
Characters
John McCarthy
primary
A lean, wiry man in his late 20s with sharp, angular features, piercing blue eyes, and short, neatly combed brown hair. His fingers are stained with chalk dust from hours at the blackboard, and he moves with the restless energy of someone used to thinking on his feet.
Marvin Minsky
primary
A 29-year-old mathematician and cognitive scientist with a wiry frame and intense eyes behind round, thick-rimmed glasses. His dark hair is slightly tousled, suggesting deep thought rather than neglect. His hands are often in motion when he speaks, as if tracing equations in the air.
Nathaniel Rochester
secondary
A middle-aged man in his late 30s with a lean, wiry build, short-cropped dark hair beginning to gray at the temples, and sharp, observant eyes behind round-framed glasses. His face bears the faint lines of frequent concentration, and his hands are those of a man accustomed to working with precision instruments.
Claude Shannon
secondary
A lean, middle-aged man in his early 40s with sharp features and a receding hairline. His keen eyes behind round wire-rimmed glasses reflect a mind constantly at work. He has a relaxed but attentive posture, with fingers occasionally tapping out binary rhythms on any available surface.
Graduate Assistant
background
A young man in his mid-20s, with a lean build and a studious demeanor. He has short, neatly combed brown hair and wears round, wire-rimmed glasses. His face is clean-shaven, and his hands are slightly ink-stained from handling papers and notes.
Dialog
John McCarthy
Look, if we consider recursive functions, we're not just talking about computing chess moves—we're talking about machines that can learn from their own operations.
Marvin Minsky
Obviously! But you're skipping the isomorphism—neural nets must first establish structural equivalence to biological synapses before we claim any 'learning' occurs.
Nathaniel Rochester
From an implementation perspective, neither approach runs on our current hardware. You'd need at least 20K of core memory just for the feedback loops Marvin's proposing.
John McCarthy
Then we need time-sharing. Machines don't have to solve everything at once—they can alternate between tasks like human thought.
Marvin Minsky
That reduces to the trivial case of multiplexing! The real breakthrough comes when we model heuristic search as probabilistic functions—
Nathaniel Rochester
Gentlemen, the 704 can handle neither probabilities nor heuristics unless we first agree on memory allocation protocols.
John McCarthy
Then we'll make new protocols. That's why we're here—to invent what doesn't yet exist.