Remember when the Turing test was the gold standard for detecting machine intelligence?
It was a simpler time, just nine months ago. The test was straightforward: human judges would hold text conversations with two hidden entities, one human and one computer. If the computer fooled at least 30% of the judges, it was deemed capable of thought.
For 70 years, this seemed an insurmountable hurdle. Then, large language models (LLMs) like GPT and Bard appeared, and the Turing test started looking a bit dated. Sure, GPT-4 might pass a Turing test by impersonating a human, but so what? LLMs lack long-term memory, the ability to form relationships, and many other human capabilities. They're not quite ready to be our friends, employees, or elected officials.
But the Turing test was never just a pass/fail benchmark. Its creator, Alan Turing, believed that the difference between genuine intelligence and a convincing imitation was only as wide as our prejudice. He argued that when a computer provokes real human responses – engaging our intellect, amazement, gratitude, empathy, and even fear – it's more than mere mimicry.
So, perhaps we need a new test: the Actual Alan Turing Test. Imagine bringing Alan Turing, the father of modern computing, into a lab with an open MacBook. After explaining the concept of a "Turing machine" and giving him access to GPT's source code, we let him loose. Turing would likely be astounded by the capabilities of modern AI, despite recognising its limitations.
The Actual Alan Turing Test is not a test of AI, but of us humans. Are we passing – or failing?
When ChatGPT burst onto the scene in November 2022, it inspired both amazement and unease. Its potential for societal disruption was hotly debated. For some, it represented an unnerving acceleration of the timeline for human-like AI. For others, it looked like a gateway to cheating and job-stealing.
But as we've become more familiar with LLMs, we've started to appreciate their unique form of intelligence. They may lack many human capabilities, but they also demonstrate impressive analytical and creative skills. And as they continue to evolve, we may need to rethink our definitions of intelligence and personhood.
Relating to an intelligent machine may be one of humanity's greatest empathic challenges. But our history gives us hope. We have a track record of recognising and respecting the humanity in others, even when they seem very different from us. As we navigate the age of AI, we should strive to keep that spirit of empathy and wonder alive.
Made with TRUST_AI - see the Charter: https://www.modelprop.co.uk/trust-ai
Commentaires