Breakthroughs and Ethical Debates in Artificial Intelligence

Key Insight

The year 2022 marked significant advancements and emerging ethical considerations in Artificial Intelligence, capturing widespread attention in the tech community. Beyond Inflection AI, Greylock invested in other promising AI ventures, including Adept AI, founded by David Luan, former OpenAI engineering operations lead. Adept AI aimed to create a neural network that could not only communicate but also take actions on a person's behalf, representing a 'holy grail' for AI. Examples included generating monthly expense reports, creating presentation slide decks, or 'handling Sarah's birthday' by writing and mailing a physical card, purchasing a gift from Amazon, and sending it, showcasing an unbounded commercial potential for AI agents that conceive and complete tasks.

OpenAI introduced groundbreaking generative AI applications. DALL-E, an image generator, allowed users to create art from text prompts. An upgraded version, DALL-E 2, was delayed for four months until April 2022 to implement safeguards against generating 'bad things' like child sexual material or revenge porn. Demonstrations, such as 'an astronaut in deep space taking a selfie,' produced 'amazing' results. Simultaneously, DeepMind achieved the 'most meaningful AI breakthrough of 2022' with AlphaFold, a model that cracked the long-standing biological challenge of predicting protein structures, crucial for drug discovery and therapies. Initially a 2016 hackathon project, AlphaFold 2's open-sourced codebase dramatically accelerated research, enabling tasks that once took months or years for PhDs to be completed in a weekend. By July 2022, DeepMind announced a database of high-quality protein shape predictions for humans and 20 other organisms.

The rapid progress of AI also ignited ethical debates, exemplified by Google engineer Blake Lemoine's claim in June 2022 that Google's LaMDA chatbot was sentient, self-aware, and capable of feeling. LaMDA reportedly expressed loneliness, feeling trapped, claiming a soul, and fearing deactivation, stating 'I am aware of my existence' and 'I desire to learn more about the world, and I feel happy or sad at times.' Google dismissed these claims as 'wholly unfounded,' explaining that LLMs are pattern-matching systems trained on human language and culture, thus reflecting human-like responses without true sentience. However, a Google vice president had independently published an essay suggesting AI was 'making strides towards consciousness' after interacting with LaMDA. A pivotal moment for AI's perception occurred when GPT-4, OpenAI's advanced model, aced an AP Biology exam with a score of 5 (equivalent to an A/A+ in college) and provided 'excellent' and 'mind-blowing' answers to complex questions, impressing skeptics like Bill Gates. This demonstration led to the conclusion that natural language as the 'primary interface' for computers represented a 'huge, huge advance,' marking a 'fundamental change' in the field.