The Turing Test

“A computer would deserve to be called intelligent if it could deceive a human into believing that it was human.” — Alan Turing

Alan Turing, widely acknowledged as the father of computer science, proposed in a 1950 paper that the most effective way to address the question “Can machines think?” was to reframe the inquiry. He contended that the term “think” lacked a formal definition, rendering the original question futile.

Turing introduced “the imitation game,” now referred to as the Turing test. This game involves three participants: an interrogator (agent C) and two agents (A and B). The interrogator poses questions such as, “Please write me a sonnet on the subject of the Forth Bridge” or “Add 34957 to 70764.” Agent A’s objective is to mislead the interrogator into believing it is agent B, creating a scenario where if agent A is a man and agent B is a woman, the man must convince the interrogator he is the woman.

In this context, Turing suggested replacing the original query with, “What occurs when a machine takes on the role of A in this game?” By substituting agent A with a universal digital computer, we could assess its cognitive capabilities. If the computer succeeded in tricking the interrogator into thinking it was human, it would merit the label of intelligence.

The Future of AI

“The future depends on some graduate student who is deeply suspicious of everything I have said.” — Geoffrey Hinton

Known as the “Godfather” of AI, Geoffrey Hinton initially trained as a cognitive psychologist. His unwavering commitment to making artificial neural networks (ANN) functional persisted even when few others believed in their potential. At that time, many experts regarded connectionist AI as a futile endeavor, despite the brain being composed of biological neural networks. Hinton maintained that to make artificial intelligence viable, “we have to perform computations in a manner akin to the human brain.”

Yet, deep learning diverges from human brain function. Although ANNs draw their name from neuroscience, they fundamentally differ from biological neural networks. An artificial neuron is a highly simplified version of a biological neuron, typically seen as just a basic calculator. However, it has been demonstrated that a single neuron can execute remarkably complex functions, such as object recognition.

Hinton's skepticism remains valid: computer vision and convolutional neural networks (CNNs) do not mirror our visual system. Deep learning algorithms require vast datasets, while humans can learn from minimal data due to inherent brain structures. Additionally, training cutting-edge AI demands significant computational power, while the human brain operates on a mere 20 watts. Hinton has laid the groundwork for deep learning and recognizes that future AI development will take a different trajectory.

Conscious AI

“No one has a clue how to build a conscious machine, at all.” — Stuart Russell

Numerous books have been authored on consciousness, yet its essence remains elusive. As Lawrence Krauss noted in a conversation with Noam Chomsky, “there’s an inverse relationship between what’s known in a field and the number of books that are written about it.”

Our understanding of consciousness is minimal. It is generally accepted that brain activity underpins mental processes and our subjective experience. However, the mechanisms behind this emergence remain a mystery. The transition from bioelectrical signals to the vast realm of thoughts, emotions, and sensations is still unknown. Without a grasp of the underlying science, it is naive to assume we can develop the corresponding technology.

Nevertheless, concerns persist about the potential creation of conscious AI. If such a breakthrough were to occur, its implications would surpass those of developing unconscious general intelligence. The distinction between these two concepts is significant: AGI refers to human-level intelligence that may not be conscious, while Strong AI, a term introduced by John Searle, describes human-level AI that is conscious. If we were to achieve Strong AI, a fundamental reevaluation of society would be necessary.

A Meta-Invention

“Anything that could give rise to smarter-than-human intelligence — in the form of Artificial Intelligence, brain-computer interfaces, or neuroscience-based human intelligence enhancement — wins hands down beyond contest as doing the most to change the world. Nothing else is even in the same league.” — Eliezer Yudkowsky

As cultural beings, we have always been inventing. Technology has propelled civilization forward long before recorded history. Innovations such as writing, agriculture, cities, the printing press, electricity, the internet, and social media have all significantly transformed the world.

However, the concept of smarter-than-human intelligence, as Yudkowsky defines it, represents a unique category of invention; it is a meta-invention. Traditional technologies serve specific purposes and are designed for defined tasks. By creating general artificial intelligence, we would effectively have invented an inventor—one that excels in virtually all domains.

To illustrate this, consider the computer, designed as a universal machine capable of performing the functions of various devices. In a sense, it serves as a meta-device. A more contemporary example would be the language model GPT-3, which possesses meta-learning capabilities, enabling it to learn how to learn rather than simply execute a predefined task. AGI would embody meta-invention in its most expansive form, prompting a fascinating yet daunting question: If we have managed to create so many inventions, what might a super-intelligent inventor bring into existence?

The Dangers of AI

“The development of full artificial intelligence could spell the end of the human race. […] It would take off on its own, and re-design itself at an ever-increasing rate. Humans, who are limited by slow biological evolution, couldn’t compete, and would be superseded.” — Stephen Hawking

Stephen Hawking, the eminent physicist, cautioned us about the risks associated with advanced AI. He believed that an AI capable of surpassing human intelligence could break free from our control. Even if we design AI to align with our values, a minor deviation could lead to catastrophic outcomes. Some argue that if we can create an all-powerful AI, the key to ensuring its benevolence lies in instilling desires that benefit humanity while rejecting harmful impulses. However, achieving this alignment poses significant challenges in practice.

Optimists maintain that once we construct the AI, it will adhere to the boundaries we have carefully established. Yet, Hawking's concerns about re-design are valid: humans, shaped by evolution, random mutations, and natural selection over millennia, have only recently begun to alter our genetic foundations. If we can manipulate our DNA, we will have transcended the limits set by evolution, our primary “designer.” What leads us to believe that machines, possessing intelligence comparable to ours, could not eventually do the same?

Given that machines already outstrip us in various areas—memory, precision, and computational ability—they would likely rapidly surpass us in other respects by modifying themselves. In such a scenario, any conflict that arises would likely result in our being “superseded.”

Our Last Invention

“Machine intelligence is the last invention that humanity will ever need to make.” — Nick Bostrom

Nick Bostrom, a philosopher at Oxford University and author of Superintelligence: Paths, Dangers, Strategies, contends that superintelligence—artificial intelligences that vastly exceed AGI levels—will emerge, and we must prepare for this eventuality.

Bostrom posits that once AGI is achieved, artificial superintelligence (ASI) will inevitably follow. Such intelligence will not only replicate human capabilities but exceed them to such an extent that we would appear to be like ants attempting to comprehend human civilization. Consequently, it will possess the capacity to create everything that can be created (aligning with Yudkowsky's perspective).

Bostrom envisions two contrasting scenarios for the future: either ASI will be aligned with humanity's interests, proving to be a positive force, or it will not be, rendering machine intelligence not just the last invention humanity will ever need to make, but the final invention we will ever produce.

The Singularity

“Within a few decades, machine intelligence will surpass human intelligence, leading to The Singularity — technological change so rapid and profound it represents a rupture in the fabric of human history.” — Ray Kurzweil

Ray Kurzweil, a prominent futurist, introduced the concept of “The Law of Accelerating Returns,” which states that evolutionary systems, including technology, evolve at an exponential pace. He asserts that this principle implies that the Singularity is imminent, predicting it will occur before the close of the 21st century, resulting in “a rupture in the fabric of human history.”

Humans will ultimately merge with machines, potentially achieving immortality as software-based entities, while artificial intelligence will become exponentially more powerful than humanity combined, extending its influence throughout the universe. Kurzweil anticipates that this transformation will commence within the next 25 years, setting the date for the Singularity at 2045.

However, his predictions have faced substantial criticism. Physicist Paul Davis argues that while growth can be exponential at times, it is ultimately constrained by resource limitations. Physicist Theodore Modis posits that “nothing in nature follows a pure exponential,” suggesting that growth typically follows a logistic curve. Although the two functions may initially appear similar, the logistic curve ultimately flattens out, reflecting real-world outcomes. Regardless, we will witness whether Kurzweil's forecasts come to fruition.

_Travel to the future with me_ _for more content on AI, philosophy, and cognitive sciences! If you have questions, feel free to ask in the comments or connect on LinkedIn or Twitter! :)_