Uncharted Waters: From Da Gama to DeepMind, the Enduring Quest to Push Boundaries
Uncharted Waters: From Da Gama to DeepMind, the Enduring Quest to Push Boundaries
Traversing unmapped seas to reach new horizons, yesterday's trailblazing explorers share a kindred spirit with today's boundary-pushing AI innovators.
Throughout history, periods of rapid progress have rewritten the rules overnight. As Portuguese fleets and British inventions reshaped sea travel, today’s AI upstarts are poised to disrupt their predecessors. The relentless march of innovation waits for no one.
It's 1497 in Lisbon, Portugal, and King Manuel I's royal court is filled with an air of excitement and anticipation. Three years prior, Manuel had inherited a kingdom obsessed with discovering a sea route to India and claiming its fabled riches. For decades, Portugal had been pushing down the coast of Africa, establishing trading posts along the way. With Bartolomeu Dias' recent rounding of the Cape of Good Hope, the prize now seemed within grasp.
Yet doubts lingered. Skeptics feared crossing the vast Indian Ocean could not be done safely. But Manuel was undeterred, convinced India could be reached by sea. He placed his trust in the accomplished navigator Vasco da Gama, who came from a noble family known for exploration. To Manuel, da Gama was the perfect man to finally complete Portugal's long-held mission.
On July 8, 1497, da Gama set sail from Lisbon with four vessels and 170 men. The expedition faced great peril, losing over half its crew along the two-year journey. But da Gama returned triumphant, having secured trading routes and treaties in India.
The Shelf Life of Secrecy is Limited
Eager to exploit this hard-won advantage, Manuel clamped down on the maps. All geographical knowledge from da Gama's trips became a closely guarded state secret. Upon returning to Portugal, any sailor who had maps was required to submit them to the Casa da Índia, or House of Trade. Sharing details of the route to India was made punishable by death.
Yet this secrecy could only delay, not prevent, competing powers from making their own voyages. Sure enough, the value of Portugal's route to India expired in the following years. One by one, other European nations set sail to stake their own claims in the east.
Manuel's attempt to tightly control valuable knowledge reflects a recurring theme throughout history. Discovery brings temporary advantages, but the march of progress waits for no one.
Centuries after Manuel's pursuit of a route to India, a new quest emerged. Rather than spices or silk, its treasure was advanced artificial intelligence.
OpenAI Sets Sail
In 2015, an organization called OpenAI embarked on this uncharted journey. Like Portugal's voyages of discovery, OpenAI’s mission sparked excitement and skepticism. Could AI truly be advanced for the benefit of all, or did its power portend danger?
To calm doubts, OpenAI preached transparency, freely dispensing maps in the form of open-source code.
The waters post-2019 proved perilous, rife with potent models and the siren song of large funding rounds. To traverse safely, OpenAI restructured to include a for-profit company. Their hard-won maps and discoveries became guarded secrets, offered only to trusted allies.
Many can access OpenAI’s latest marvels today, but the training data and model details underpinning innovations like DALL-E or ChatGPT remain a secret.
The ripe fruits of triumph entice endless seekers to pluck from the same tree. OpenAI's successful models have spawned a legion of attempts to chart a course to advanced intelligence, training massive language models using reinforcement learning from human feedback. Much like Vasco da Gama's celebrated return from India set off a frenzy centuries before.
Today's Leader, Tomorrow's Follower
Portugal's early advantage soon sparked an arms race among seafaring empires. In 1595, the first Dutch voyage to Asia was launched under Cornelis de Houtman, seeking to bypass Portugal's control of the spice trade. The British, French, and Danes followed, establishing rival East India companies in the 1600s.
Yet first mover status proved fleeting. While pioneering the route, Portugal failed to maintain supremacy. Their lead dissolved against the ingenuity of British scientists.
As the 1700s dawned, the British navy began dominating the seas. Through brilliant innovation, they mastered navigation as never before. The sextant, octant, and marine chronometer - all British inventions - profoundly expanded their capabilities. By quantifying longitude and latitude, these tools enabled unprecedented precision.
Combining scientific insight with naval skill, Britain commanded the oceans by the century's end.
In With the New Breed, Out With the Tweaks
If history repeats itself, the same patterns will play out in the quest for advanced AI. OpenAI has had success building what appears to be the first sign of an AI model with broad applicability and hints of intelligence through their use of transformer architectures, massive training data, and reinforcement learning from human feedback.
The current projects are taking the same route. Some, like Bloomberg, are building domain-specific models like BloombergGPT. Others are building general models with unique capabilities, as Anthropic aims to do with its Claude model and focus on responsible AI. The open-source community is also developing its own versions of proprietary models, adding optimizations and enhancements along the way.
But as in the past, the next major leap will likely come not from optimization alone but from fresh innovation. This could provide the 10x improvement that redefines the state of the art. Many are pursuing this, but one interesting effort to watch is Project Gemini from Google.
Google's current Bard chatbot is their rapid attempt to counter ChatGPT. But in Gemini, Google has combined its DeepMind deep learning technology with large language models. DeepMind is the company behind AlphaGo, which beat world champion Lee Se-dol at the game Go in 2016. Deep learning also enabled DeepMind's AlphaFold system to predict protein structures much faster, sometimes reducing timescales from months to hours.
In a recent Wired interview, DeepMind CEO Demis Hassabis boldly claimed Gemini would "dwarf" ChatGPT. Such lofty promises are common when exciting new technologies debut, though the actual impact is difficult to predict this early.
One thing history shows is that leadership in any domain is fleeting. Just as Britain eventually surpassed early explorer Portugal on the seas, the AI landscape will continue to evolve. Other innovators, whether DeepMind or new challengers, will have opportunities to spur the next great leap forward.
The rapid pace of progress ensures no one retains dominance for long. There will be future moments, like Britain's naval innovations, that profoundly reshape the playing field.