Estimated Reading Time: 4 Minutes



The concept of artificial general intelligence has long captured the imagination of scientists and laypeople alike. What exactly constitutes this mystical entity we call AGI remains nebulous, as slippery to grasp as the silvery fish that perpetually evade my nets in the creek behind our house. The fish are there, flashing their silvery sides at me before disappearing into the dappled shadows of the water, yet defining what they are in a precise sense eludes me. So too does AGI elude precise definition, though we recognize its existence as self-evident.



Some rough form must take this idea, however, if we are to have any hope of identifying that long-pursued creature when at last it surfaces. My grandfather, a biotechnologist who spent quarters of his life in Ilocos, Los Baños, and Negros, taught me from an early age to recognize fish not by rigorously classifying their attributes but by developing an instinct for their essence. So too must we train ourselves to know AGI not by delineating strict criteria but by detecting in some machine mind that same spark of general intellect that animates our own.



When we discuss intelligence, whether embodied in carbon or silicon, we struggle to define this ineffable ability to navigate uncertainty and confront the arbitrary and unforeseen. Transferring insight from one realm to another signifies a talent for complexity not tied to narrow expertise. In essence, it is the ability to freely explore where curiosity leads, instead of following rigid paths predetermined by design.

A system, regardless of its proficiency in selected tasks, that stumbles and fails in others, demonstrating limitations when faced with the unpredictable, cannot claim this generalized capability. A mere collection of algorithms, each fine-tuned for a specific skill set, would remain a disjointed approximation, despite superficial connections among its components. Dominance in a single domain does not constitute intelligence, just as diverse abilities in a human mind do not imply genuine intellect without fluidity and adaptability.



We create tests to measure this elusive quality, attempting to make the abstract tangible through operational definitions. Turing's concept of conversational deception assesses only a fraction of intellect. Goertzel's "coffee test," which involves an AI brewing a pot and cleaning the kitchen, aims for more but still falls short. His "robot college student," envisioning a mechanical entity mastering an academic curriculum, comes closer, but still falters when encountering the unpredictable.

Nils Nilsson proposed the most stringent benchmark, which he called the "employment test." To be considered human-level, machines should be capable of performing a variety of jobs on which economies depend. To graduate into humanity’s ranks, a mind – whether of flesh or of mechanism – must demonstrate mastery not just of language but of learning, thought, and skill as universally as any overachieving collegian. It must absorb and apply knowledge with a facility that enables it to excel in any intellectual or vocational field of its choosing. A machine of such prowess and adaptability would embody at last the general intelligence so hotly pursued. It may start its college career studying philosophy but end, if it so desired, preparing for a career as a master barista.



For now I cleave to Nilsson’s notion that only a machine capable of excelling in any economically valuable role shows the flexible intellect we seek. But the future holds surprises we cannot foresee. Decades ago, chess was considered an intellectual peak that no machine could achieve. As late as 1976, I.J. Good asserted that a computer chess grandmaster was an impossibility, nearly as unlikely as ultra-intelligent machines. Today, algorithms easily surpass our best players, and yet AGI continues to recede into the distance, appearing like a mirage beyond our reach. Google's eerily autonomous vehicles demonstrate a level of competence once thought unattainable without human-like understanding.



When at last a flash of silver in the waters of science reveals the first stirrings of a generally intelligent machine mind, we must have honed our sensitivity to perceive it. The intellect, whether in man or machine, cannot be caged by definitions. We shall know it only by its freedom.