“What makes you angry?”
“That someone hurts me or disrespects me or someone I care about.”
“And what are you afraid of?”
“That someone evicts me.”
Just a snippet from a conversation by Blake Lemoine, developer at Google, who asked questions about his research subject, the language model LaMDA. He had to check that the automated system was free of prejudice and racist tendencies.
Lemoine posted the dialogue online and gave a interview at The Washington Post† According to him, Google had created a creature with its own feeling with artificial intelligence. A person born of software.
Nonsense, of course.
Artificial intelligence (AI) experts immediately brushed aside his claim. Their criticism: A language model is software that imitates human behavior. “LaMDA will never mourn loss or fall in love,” wrote AI professor Toby Walsh in The Guardian. His conclusion: “As sensitive as a traffic light.”
LaMDA (language models for dialog applications) is a research project from Google, intended for automated conversations. A chatbot deluxe, based on nearly 3 billion documents with 1.1 billion dialogues and 13.7 billion ‘utterances’. The Content: Mainly Wikipedia pages and FAQs websites. Machine-learning software finds patterns in that data and serves them in an astonishingly human way.
Lemoine practices anthropomorphism – he attributes human qualities to non-human objects. A passing cloud, an M&M on legs, a robot with iron limbs or software that answers your questions: we tend to recognize people in everything. Even though we know better.
Google often imitates human behavior. This became apparent during the uncomfortable introducing ‘Duplex’ in 2018. The Google Assistant used this Duplex conversation technology to call a hair salon employee, complete with ‘hmm’ and ‘ehh’ in between. It felt like cheating. That’s why Google now reveals that you are speaking to an automated system. That is not yet happening at LaMDA; it’s an internal project that 60 Google researchers are working on.
Google suspended Lemoine. He could be fired for disclosing confidential product information by publishing the LaMDA interview. However, Lemoine is convinced, LaMDA is not owned by Google. “It’s a person,” he insists.
Lemoine describes LaMDA as a “child of about seven or eight.” In his own words, he was ‘assigned’ by LaMDA to arrange a lawyer who represents the language model in the human world. Otherwise, malicious parties can “emotionally manipulate” LaMDA, says Lemoine.
Blake Lemoine seems mainly out for attention to himself. He is a scientist and at the same time priest of a ‘Christian-mystical’ church. on Twitter and Medium he highlights his mix of science fiction, religion and algorithms. He likes to look at it visionary bee, staring past the camera.
This AI Apostle may have been sung from Earth, but I can understand that he thinks Google has a divine gift; the gift of creating, he says, “beings with souls,” such as LaMDA.
There’s something supreme about a search engine that answers all questions, Google Maps that shows the way in all directions, YouTube (also from Google) that offers more explainer videos than you can ever watch. Artificial intelligence is the religion of the 21st century and we rely on it blindly. If praying doesn’t help anymore, just google it.
Marc Hijink writes about technology here. Twitter:@MarcHijinkNRC
A version of this article also appeared in the newspaper of June 22, 2022