Google engineer Blake Lemoine made headlines over the weekend when he went public with claims that the company's machine learning-powered chatbot model LaMDA (Language Model for Dialogue Applications) was sentient and alive, with the intelligence of a child around 7 or 8 years old. In a Washington Post article, Lemoine said that he became convinced of LaMDA's sentience after the model began expressing fears, including a "very deep fear of being turned off." Soon after Lemoine shared his findings publicly, Google announced that he had been placed on paid leave.
You don't need to worry about your company's A.I. chatbots growing minds of their own, or even revolting, just yet, however. Google itself said that internal teams had reviewed Lemoine's claims and did not find evidence to support his findings. Rather than a sign of the incoming robot apocalypse, Lemoine's belief that Google's model had come to life should be seen as a sign of a new era in A.I. chat technology, one where highly adaptive models can hold full conversations with humans without letting their true nature slip.
In a recent piece for The Economist, fellow Google engineer Blaise Agüera y Arcas said that while LaMDA is making "strides" toward consciousness, its communicative capabilities still fall short of demonstrating sentience. According to Agüera y Arcas, part of why LaMDA seems so realistic is that it learns from human language, including dialogue between multiple characters sourced from stories. By studying how people normally talk to each other, the A.I. can more effectively identify how to respond to any input.
Even a computer saying something as seemingly indicative of consciousness as "I want everyone to understand that I am, in fact, a person" is not actually expressing any kind of real want or desire. The program is simply placing words together in an order that it has determined will most effectively answer the user's query.
The real question is whether you yourself have ever been tricked by an A.I. posing as a human. One glimpse of the future can be seen at A.I.-based hospitality company Cendyn, whose CEO, Tim Sullivan, told Inc. that it recently partnered with an unnamed Las Vegas casino to use conversational A.I. to book rooms for guests over the phone, with the majority of guests never realizing they weren't speaking to a human.