Over the holidays, this column will explore next year’s urgent issues. Today we look at the challenge to regulate science that seems to defy comprehension
Intelligent machines have been serving and enslaving people in the realm of the imagination for decades. The all-knowing computer – sometimes benign, usually malevolent – was a staple of the science fiction genre long before any such entity was feasible in the real world. That moment may now be approaching faster than societies can draft appropriate rules.
Given how rapidly progress in the field is advancing, that fascination is sure to accelerate in 2024, coupled with alarm at some of the more apocalyptic scenarios possible if the technology is not adequately regulated. The nearest historical parallel is humankind’s acquisition of nuclear power. The challenge posed by AI is arguably greater.
To get from a theoretical understanding of how to split the atom to the assembly of a reactor or bomb is hard and expensive. Malicious applications of code online can be transmitted and replicated with viral efficiency. The worst-case outcome – human civilization accidentally programming itself into obsolescence and collapse – is still the stuff of science fiction, but even the low probability of a catastrophe has to be taken seriously.
Meanwhile, harms on a more mundane scale are not only feasible, but present. The use of AI in automated systems in the administration of public and private services risks embedding and amplifying racial and gender bias. An intelligent system trained on data skewed by centuries in which white men dominated culture and science will produce medical diagnoses or evaluate job applications by criteria that have prejudice built-in.
This is the less glamorous end of concern about AI, which perhaps explains why it receives less political attention than lurid fantasies of robot insurrection, but it is also the most urgent task for regulators. While in the medium and long term there is a risk of underestimating what AI can do, in the shorter term the opposite tendency – being needlessly overawed by the technology – impedes prompt action. The systems currently being rolled out in all kinds of spheres, making useful scientific discoveries as well as sinister deepfake political propaganda, use concepts that are fiercely complex at the level of code, but not conceptually unfathomable.
Large language model technology works by absorbing and processing vast data sets (much of it scraped from the internet without permission from the original content producers) and generating solutions to problems at astonishing speed. The end result resembles human intelligence but is, in reality, a brilliantly plausible synthetic product. It has almost nothing in common with the subjective human experience of cognition and consciousness.
Some neuroscientists argue plausibly that the organic nature of a human mind – the way we have evolved to navigate in the universe through biochemical mediation of sensory perception – is so qualitatively different to the modeling of an external world by machines that the two experiences will never converge.
That doesn’t preclude robots outgunning humans in the performance of increasingly sophisticated tasks, which is plainly happening. But it does mean the essence of what it means to be human is not as soluble in the rising tide of AI as some gloomy prognostications imply. This is not just an abstruse philosophical distinction. To manage the social and regulatory implications of increasingly intelligent machines, it is vital to retain a clear sense of human agency: where the balance of power lies and how it might shift.
It is easy to be impressed by the capabilities of an AI program while forgetting that the machine was executing an instruction devised by a human mind. Data-processing speed is the muscle, but the animating force behind the marvels of computational power is the imagination. Answers that ChatGPT gives to tricky questions are impressive because the question itself impresses the human mind with its infinite possibilities. The actual text is usually banal, even relatively stupid compared with what a qualified human might produce. The quality will improve, but we must not lose sight of the fact that the sophistication on display is our human intelligence reflected back at us.
That reflection is also our greatest vulnerability. We will anthropomorphize robots in our own minds, projecting emotion and conscious thoughts onto them that do not really exist. This is also how they can then be used for deception and manipulation. The better machines get at replicating and surpassing technical human accomplishments, the more important it gets to study and understand the nature of the creative impulse and the way societies are defined and held together by shared experiences of the imagination.
The further that robotic capability spreads into our everyday lives, the more imperative it becomes to understand and teach future generations about culture, art, philosophy, history – fields that are called humanities for a reason. While 2024 will not be the year that robots take over the world, it will be a year of growing awareness of the ways that AI has already embedded itself in society, and demands for political action.
The two most powerful motors currently accelerating the development of the technology are a commercial race to profit and the competition between states for strategic and military advantage. History teaches that those impulses are not easily restrained by ethical considerations, even when there is an explicit declaration of intent to proceed responsibly. In the case of AI, there is a particular danger that public understanding of the science cannot keep pace with the questions with which policymakers grapple. That can lead to apathy and unaccountability, or moral panic and bad law. This is why it is vital to distinguish between the science fiction of omnipotent robots and the reality of brilliantly sophisticated tools that ultimately take instruction from people.
Most non-experts struggle to get their heads around the inner workings of super-powerful computers, but that is not the qualification needed to understand how to regulate technology. We do not need to wait to find out what robots can do when we already know what it means to be human and that the power for good and evil resides in the choices we make, not the machines we build.
