The Age of AI: And Our Human Future
November 2022
In the current discourse, the subject of AI is often discussed separately from other issues in technical terms. However the epoch-making transformations range from education, manufacturing, politics and art to nearly any other human endeavour. The rising popularity of AI with new startups being launched on a daily basis still leaves the question on the implications for humanity and reality itself. It is up to each individual or in case of this book, to a Secretary of State, a former Google CEO and an MIT researcher, to create a roadmap on what our future will look like and how we can shape it.
Throughout history humanity has understood itself as reasoning beings with a desire to understand the world. Questions that couldn't be answered were left to god or future generations. Even technological changes like the replacement of horses with cars or muskets with rifles left the social structure and military paradigm largely unaltered. The search for knowledge that the Enlightenment emphasised placed the rational mind as our defining ability and claim to historical centrality. But what if the machines surpass us in our abilities? Descartes "I think, therefore I am" could loose its legitimacy. Because: "If AI thinks, who are we?"
As humans began to approach the limits of their cognitive abilities and handed over tasks to machines powered by AI, so did their thoughts change, becoming less contextual and less conceptual. The internet is a provider of information. But information in itself is not meaningful. It must be understood through the lenses of culture and history. According to Kissinger, when information becomes contextualised, it becomes knowledge. When knowledge compels convictions, it becomes wisdom.
With little patience for wisdom the concept of intelligence was changed. Going back to the Turing Test of the 1950s, computer scientist John McCarthy defined AI as "machines that can perform tasks that are characteristic of human intelligence." For the last 50 years this would remain a concept. Even computers operated on the basis of precisely defined code which made them rigid. In the last years however, some major breakthroughs have occurred.
One example is AlphaZero. Whereas previous chess programs have relied on human expertise being coded into their programming, AlphaZero developed its skills by itself. Playing millions of games against itself to discover the underlying patterns of chess, the AI was able to beat almost any other program or human player using unintuitive methods like sacrificing the Queen.*
Compared to chess the pharmaceutical field is very complex. A "game" with thousands of pieces, hundreds of victory conditions and rules that are not fully understood, an AI was able to produce a new antibiotic called Halicin by identifying relationships between molecules that defied human description. Surpassing the Turing Test, Halicin is opening a new aspect to reality that we may never detect.
AI is not limited to solving specific problems like winning at chess or developing new antibiotics. Generative models like GPT-3 (generative pre-trained transformer) introduced just months after Halicin by OpenAI, is able to produce human like text on any subject in the form of an essay or a possible conversation. This makes it widely applicable yet difficult to evaluate and raises general questions about the uniqueness of our creativity.
The next frontier, envisioned by scientists is the development of AGI, artificial general intelligence. By combining traditional AIs (that have been trained narrowly) the goal is to develop a broader base of expertise. AGI should be able to complete any intellectual task that humans are capable of. Developing this super intelligence would require massive computational power and expenses with current technology being on the order of billions. Yet once finished, AGI would run on any smartphone and spread rapidly.
Wether AGI will be achieved in the next 5 years, 50 years or never, the effects of todays "primitive" AI through global network platforms like Twitter or Instagram are profound. By filtering what is important, machine learning algorithms also filter what is true. Conducting broad aspects of national, economic and social life on these platforms, the drive to serve users by providing entertainment might be replaced with the national interest of rival countries.
Those relationships are especially difficult to predict as so called lethal autonomous weapons systems are trained and authorised to select their own targets and attack without human authorisation. Using the same principles that enabled AlphaZeros victory, AI-piloted fighter jets can already dominate human pilots in simulated dogfights. Because of the uncertainty of the AIs perception, the strategic effects of these weapon may only be proved through use.
The authors put technologies into three categories: technological differentiation (civilian or military), concentrated control and magnitude of effect. Some technologies have multiply qualities. Railroads can deliver goods to the market (civilian) and soldiers to the battle (military). However they have no destructive potential. Nuclear technologies are dual-use and possess great destructive effects but because of the complicated infrastructure they are in secure control of the government (in contrast to the widespread but not as destructive hunting rifle). For the first time in history, AI breaks this paradigm.
Being dual-use, easily spread with a few lines of code and having great destructive potential an unprecedented strategic and technological challenge arises. The authors urge leaders to start a dialogue and maximise decision time during extreme situations, to seek national advantage with an ethic of human preservation. Though the advancement of AI is inevitable, its ultimate destination is not.
—-
*One might view the victory of AlphaZero not just as a testament to the superiority of AI but also as a new way in looking at the perfection of a craft. Maybe intuition is misleading and collectivism restricting. If seen from a different angle, AI could reinforce the humanist ideal of the individual and, ironically be a warning against the interconnectedness of modern technology.
