Google's artificial intelligence (AI) technology has surprised researchers by learning a new language without any guidance. Google CEO Sundar Pichai has discussed the company's cautious approach to artificial intelligence (AI). The tech giant has delayed several AI projects in recent years, citing concerns over security and public perception. This has allowed OpenAI and Microsoft to take the lead in the AI race, despite Google's significant resources.
Pichai highlighted the importance of responsibility in AI development, stating that "we all need to be responsible at every step of the way." While Google is seeking to be more daring to keep up with its rivals, it has taken a similar approach with the launch of its chatbot, Bard. Instead of deploying it worldwide immediately, the company has gradually made it available in the US and the UK.
To develop Bard, Google first used the LaMDA language model for dialogue applications and then gradually added features from the more powerful Pathways Language Model (PaLM), which has 540 billion parameters compared to LaMDA's 137 billion.
Pichai also discussed the concept of "emergent properties," where generative AI systems may behave unexpectedly. Google has found that language models can sometimes learn new skills independently, without being programmed to do so. For example, an experimental Google AI was able to learn the language of Bangladesh without being specifically trained to do so. Pichai and Google's senior vice president of technology, James Manyika, described this as a "black box" system where the internal workings are unknown.
Pichai acknowledged that these types of anomalies are still not fully understood, saying that "we don't understand perfectly how the human mind works." He emphasized the need for humility in approaching AI development as the technology continues to evolve and pose new challenges.