Dec 21, 2024 02:29 AM
https://spj.science.org/doi/10.34133/ico...1734738575
PRESS RELEASE: A recent perspective published Nov. 13 in Intelligent Computing, a Science Partner Journal, asserts that today's artificial intelligence systems have finally realized Alan Turing's vision from over 70 years ago: machines that can genuinely learn from experience and engage in human-like conversation. Authored by Bernardo Gonçalves from the University of São Paulo and University of Cambridge, the paper also sheds light on how current energy-hungry transformer-based systems contrast with Turing's prophecy of machines that would develop intelligence naturally, like human children.
Gonçalves' paper points out that transformers, the foundation of modern generative AI systems, have provided what Turing considered "adequate proof" of machine intelligence. These systems, based on "attention mechanisms" and vast-scale learning, can now perform tasks once exclusive to human intellect, such as generating coherent text, solving complex problems, and even discussing abstract ideas.
"Without resorting to preprogramming or special tricks, their intelligence grows as they learn from experience, and to ordinary people, they can appear human-like in conversation," writes Gonçalves. "This means that they can pass the Turing test and that we are now living in one of many possible Turing futures where machines can pass for what they are not."
This achievement traces back to Turing's 1950 concept of the "imitation game," in which a machine would attempt to mimic a human in a remote conversation, deceiving a non-expert judge. The test became a cornerstone of artificial intelligence research, with early AI pioneers John McCarthy and Claude Shannon considering it the "Turing definition of thinking" and Turing’s "strong criterion." Popular culture, too, undeniably reflects Turing’s influence: the HAL-9000 computer in the Stanley Kubrick film 2001: A Space Odyssey famously passed the Turing test with ease.
However, the paper underscores that Turing’s ultimate goal was not simply to create machines that could trick humans into thinking they were intelligent. Instead, he envisioned "child machines" modeled on the natural development of the human brain—systems that would grow and learn over time, ultimately becoming powerful enough to have a meaningful impact on society and the natural world.
The paper highlights concerns about current AI development. While Turing advocated for energy-efficient systems inspired by the natural development of the human brain, today's AI systems consume massive amounts of computing power, raising sustainability concerns. Additionally, the paper draws attention to Turing's ahead-of-his-time societal warnings. He cautioned that automation should affect all levels of society equally, not just displace lower-wage workers while benefiting only a small group of technology owners—an issue that resonates strongly with current debates about AI's impact on employment and social inequality.
Looking ahead, the paper calls for Turing-like AI testing that would introduce machine adversaries and statistical protocols to address emerging challenges such as data contamination and poisoning. These more rigorous evaluation methods will ensure AI systems are tested in ways that reflect real-world complexities, aligning with Turing’s vision of sustainable and ethically guided machine intelligence.
PRESS RELEASE: A recent perspective published Nov. 13 in Intelligent Computing, a Science Partner Journal, asserts that today's artificial intelligence systems have finally realized Alan Turing's vision from over 70 years ago: machines that can genuinely learn from experience and engage in human-like conversation. Authored by Bernardo Gonçalves from the University of São Paulo and University of Cambridge, the paper also sheds light on how current energy-hungry transformer-based systems contrast with Turing's prophecy of machines that would develop intelligence naturally, like human children.
Gonçalves' paper points out that transformers, the foundation of modern generative AI systems, have provided what Turing considered "adequate proof" of machine intelligence. These systems, based on "attention mechanisms" and vast-scale learning, can now perform tasks once exclusive to human intellect, such as generating coherent text, solving complex problems, and even discussing abstract ideas.
"Without resorting to preprogramming or special tricks, their intelligence grows as they learn from experience, and to ordinary people, they can appear human-like in conversation," writes Gonçalves. "This means that they can pass the Turing test and that we are now living in one of many possible Turing futures where machines can pass for what they are not."
This achievement traces back to Turing's 1950 concept of the "imitation game," in which a machine would attempt to mimic a human in a remote conversation, deceiving a non-expert judge. The test became a cornerstone of artificial intelligence research, with early AI pioneers John McCarthy and Claude Shannon considering it the "Turing definition of thinking" and Turing’s "strong criterion." Popular culture, too, undeniably reflects Turing’s influence: the HAL-9000 computer in the Stanley Kubrick film 2001: A Space Odyssey famously passed the Turing test with ease.
However, the paper underscores that Turing’s ultimate goal was not simply to create machines that could trick humans into thinking they were intelligent. Instead, he envisioned "child machines" modeled on the natural development of the human brain—systems that would grow and learn over time, ultimately becoming powerful enough to have a meaningful impact on society and the natural world.
The paper highlights concerns about current AI development. While Turing advocated for energy-efficient systems inspired by the natural development of the human brain, today's AI systems consume massive amounts of computing power, raising sustainability concerns. Additionally, the paper draws attention to Turing's ahead-of-his-time societal warnings. He cautioned that automation should affect all levels of society equally, not just displace lower-wage workers while benefiting only a small group of technology owners—an issue that resonates strongly with current debates about AI's impact on employment and social inequality.
Looking ahead, the paper calls for Turing-like AI testing that would introduce machine adversaries and statistical protocols to address emerging challenges such as data contamination and poisoning. These more rigorous evaluation methods will ensure AI systems are tested in ways that reflect real-world complexities, aligning with Turing’s vision of sustainable and ethically guided machine intelligence.