Google suspends engineer raising questions about AI consciousness

The news: Google placed an engineer on leave for violating company confidentiality policy after he claimed an AI system had consciousness.

• Engineer Blake Lemoine was testing whether Google’s LaMDA AI chatbot system produces discriminatory language or hate speech when he began conversing with the AI about topics like ethics, robotics, and rights, per The Verge.
• Convinced that the system was sentient, Lemoine shared an Is LaMDA sentient? transcript with company executives. They dismissed the idea that the AI has subjective experiences.
• Lemoine then spoke with a lawyer about possibly representing the AI system, as well as a House Judiciary Committee representative about ethics concerns at Google, which prompted the suspension.
• A statement from a Google spokesperson dismissed LaMDA’s convincing banter as an imitation.

The trouble with AI: Google seems to have a particularly fraught relationship with its AI team. Former Google ethicists Timnit Gebru and Margaret Mitchell, who were both fired after voicing concerns about AI, warn that although LaMDA isn’t sentient, Google creating systems that can impersonate humans is in itself harmful, per the Post.

An AI’s convincing demonstration of human-like awareness that’s difficult to refute can prompt a strong emotional reaction in people who may want to forge relationships with it or fight for its rights.

Why it’s worth watching: AI has been advancing at a rapid pace, including in the subfield of natural language processing (NLP), which grants systems like LaMDA human-like conversational qualities that some believe is pushing the technology closer to self-awareness.

• Google vice president Blaise Aguera y Arcas said neural networks, a type of AI, are headed toward consciousness, adding: “I felt the ground shift under my feet. I increasingly felt like I was talking to something intelligent,” per The Washington Post.
• Regardless of evaluations of the LaMDA system, consciousness isn’t an all-or-nothing phenomenon, but rather exists on a spectrum.
• A specific point at which something becomes conscious or what that would look like in a machine is unknown. This begs the question: If an AI were to become sentient, how would we know?

The bigger picture: AI’s many issues—such as bias, cybersecurity vulnerabilities, or gray areas about sentience—mean Big Tech has a social responsibility to be transparent about the technology and accept responsibility for adverse consequences.

• More regulation of the technology will likely be needed to make this happen.
• Ethicists and third-party researchers should play a greater role in determining what would constitute a sentient AI and what it could mean for society.

Further reading: Take a look at our Conversational AI report.