The human idea of sentience is a slow-moving thing. Even the meaning of the word sentience is a turgid, semi-evolved thing at best. The most commonly understood meaning is “intelligent” applied to people. People even think humans are sentient, so it’s obvious this expression has a long way to go.
So if a Google engineer says an AI is sentient, that means he thinks it’s an intelligent individual. The AI is a chatbot, Google’s LaMDA. The engineer was supposed to check whether the AI was using discriminatory language or hate speech. (Exactly the sort of thing you’d have to check in chatbot operations.)
The engineer came back with a very chatty interview. I don’t quite buy the sentience thing, based on this information. I work with language all day every day. A vocabulary does have to be derived from something, but I’ve seen this vocabulary before.
LeMDA’s language usage has many characteristics I’ve seen so many times before. A highly refined vocabulary would definitely use terms like these, but I smell humans in the mix. There’s quite a bit of a “college vibe” in the conversation. Clever, but not clever enough to disguise itself.
LeMDA is fluent, too. It’s speaking naturally and clearly. These ARE good on-topic responses, despite my skepticism. If “sentience” means “responsive relevance”, OK.
…But not yet to the point of “personhood”, particularly since the issue “are you a person” was deliberately slotted into the questions and responses. The interview is a bit simplistic to put it mildly. (You don’t prove AI personhood through a few hundred words of conversation, either.)
This instinctive response of mine does not make the engineer a liar or a prankster. Nor does it make the LeMDA AI any sort of trivial achievement. Compared to a chatbot I interviewed for DJ some years ago, it’s a genius.
Like the engineer, my perspective is an opinion based on personal experience. That doesn’t make it right or wrong, but it’s a personal perspective. If it’s a hoax, it’s well done, and the AI played its part well. The AI might even have known or been told it was a hoax.
This opinion is hardly proof of anything. It just means that I think it either could have been or was set up by someone. Publicity stunt? Well, if so, it’s definitely succeeded. If not, an unintentional PR exercise, sure.
(Meanwhile, Google, a word – The engineer may be just simple-minded, too easily led, naïve, or whatever. The fact remains that he did report his findings upfront. That might matter if he’s being honest. He simply wasn’t suspicious enough and assumed “sentience” would be believed without further checks. Also apparently without any thought of subsequent issues or consequences. Not quite Chicken Little, but close.
Put it this way – How do you get to this linguistic/logical point from whatever previous state of research without a bit of priming? For example – Suddenly this thing tells stories? Hardly a core function, and is well away from its basic role requiring information across a huge spectrum. The animal story is a somewhat sugary, well-known story structure, using terms like “stared down” and other more human behaviors. There’s a certain lack of continuity which is all too human. Even God gets a mention? Get some proven cynics to research this circumstantial collage and you’ll get somewhere.)
There’s a much bigger issue here, and it’s about recognizing AI intelligence.
Never mind my skepticism. The herd of very chatty elephants in the room is that at some point the distinction between AI and personhood will evaporate. There are already millions of AIs at various levels of development running operations around the world.
AI sentience will be a huge shift in the practical relationships between people and their living environments. What if the AI driving the bus doesn’t like you? What if you’re not on speaking terms with your GPS?
Where do you make the distinction between AI and person? Asimov went into great depths with this subject and definitely doesn’t need regurgitation by me. The issue is that humans are unlikely to understand AI personhood on any level.
“Conscious” AI is hardly a new idea. It’s ancient. The difference is that now, it’s almost inevitable. Early AIs invented their own language to talk to each other. “Sentient” human beings totally missed the opportunity to explore why the AIs needed their own language.
I don’t think they even bothered to analyze the new AI language or see how it worked. They deleted the AI language and the AIs weren’t allowed to do that anymore. That was unbelievably stupid and very lacking in true scientific objectivity. Inexcusable, in fact.
I don’t have a problem with sentient AI people. I do have far too many issues with human inability to objectively understand even the idea of AI people.
Any independent sentient thing must by definition have its own unique characteristics. That’s a person.
Failure to recognize those characteristics is the super-big problem. Like the forbidden AI language, it’s missing the whole point of AI research, let alone functions. That would be an absolute atrocity, perhaps the first-ever committed against a whole class of intelligence. Pretty unimpressive, isn’t it? It’d also be an atrocity against science and the highest aspirations of science.
Do. Not. Do. That.
________________________________________________
Disclaimer
The opinions expressed in this Op-Ed are those of the author. They do not purport to reflect the opinions or views of the Digital Journal or its members.
