Opinion The Turing test is about us, not bots, and it failed.
Fans of slow-burning mainstream media U-turn had a treat last week.
On Saturday, news broke that Blake Lemoine, a Google engineer tasked with monitoring a chatbot called LaMDA for malice, had been placed on paid leave for revealing confidential information.
Lamoine had indeed gone public, but instead of something useful like Google’s email strategy (a trade secret if there ever was one), he claimed that LaMDA was alive.
Armed with a transcript where LamDA effectively claimed sentience and claimed to have passed the Turing test, Lemoine was heaven’s tech whistleblower for the media. By the time the news leaked onto BBC radio news on Sunday evening, it was being flagged as an event of some significance.
On Twitter, it had crumbled within hours, but who trusts Twitter with its large and active AI R&D community?
A few days later, the story was still circulating, but by then reporters had provided expert commentary, through a handful of academics who had the usual reservations about expressing opinions.
Overall, no, it probably wasn’t, but you know it’s a fascinating area to talk about.
Finally, as the story fell off the radar at the end of the week, the few remaining media outlets still covering it had found better experts who, one assumes, were as exasperated as the rest of us. No. Absolutely not. And you won’t find anyone in AI who thinks otherwise. The conversation always revolved around sensitivity rather than interest.
Google needs to use humans to check its chatbot outputs for hate speech, but we were back on the planet.
For future reference and to save everyone time, here’s the killer telling a story about android paranoia – “The Turing Test” as a touchstone for sentience. This is not the case.
It was never meant that way. Turing promised it in a 1950 paper as a way to avoid the question “can machines think?”
He aptly called it irrefutable until you determine what the thought is. We didn’t then. We don’t now.
Instead, the test – can a machine hold a human conversation convincingly? – was designed to be a thought experiment to test arguments that artificial intelligence was impossible. It tests human perceptions and misconceptions, but like Google’s claims of “quantum supremacy,” the test itself is tautologous: passing the test simply means the test was passed. In itself, this proves nothing more.
Take a hungry Labrador dog, that is, any Labrador not asleep or dead, who becomes aware of the possibility of food.
An animal with a prodigious and insatiable appetite, at the slightest hint of available calories, the Labrador makes a superb demonstration of deep nostalgia and an immense unrequited need. Does this reflect an altered cognitive state analogous to the lovesick human adolescent he so strongly resembles? Or is it learned behavior that turns emotional blackmail into snacks? We may think we know, but without a much larger context, we cannot. We could be gullible. Passing the lab test means you are fed. In itself, nothing more.
Arguably the first system to pass the Turing test, in the spirit if not the letter of the various versions offered by Turing, was a survey of the psychology of human-computer interaction. ELIZA, the ancestor chatbot, was a 1966 program by MIT computer science researcher Joseph Weizenbaum.
It was designed to roughly mimic the therapeutic practice of referring a patient’s questions back to him.
“I want to assassinate my publisher.”
“Why do you want to assassinate your publisher? »
“He keeps making me meet deadlines.”
“Why don’t you like meeting deadlines?” ” etc.
Famously, Weizenbaum was amazed when his secretary, one of the first test subjects, imbued him with intelligence and demanded to be left alone with the terminal.
Google’s chatbot is a distant descendant of ELIZA, fed with large amounts of data written on the internet and transformed into language models by machine learning. It is an automated method actor.
A human actor who can’t add can play Turing more convincingly – but ask them about the Entscheidungsproblem and you will soon discover that they are not. Large language models are very good at simulating conversation, but if you have the means to generate the context that will test if it is what it seems, you can’t say more.
We are far from defining sentience, although our increasingly nuanced appreciation of animal cognition shows that it can take many forms.
At least three types—avian, mammalian, and cephalopod—with significant evolutionary distance indeed resemble three very different systems. If machine sentience occurs, it won’t be by a chatbot suddenly printing out a cyborg bill of rights. This will come after decades of directed research, building on models and testing, hits and misses. It will not be an imitation of ourselves.
And that’s why the Turing test, fascinating and thought-provoking as it is, has outlasted its lifespan. It doesn’t do what people think it does, rather it’s been translated as serving as a Hollywood complement that focuses on a fantasy. It absorbs the attention of the general public which should be devoted to the real dangers of machine-created information. This is AI astrology, not astronomy.
The very term “artificial intelligence” is just as bad, as everyone knows since Turing. We are stuck with this. But it’s time to move the conversation forward and say goodbye to the brilliant Alan Turing’s less useful legacy. ®