‘Google model translated Bengali without being trained for it’: Sundar Pichai’s old clip on AI behaviour goes viral

A short clip from a 2023 interview featuring Sundar Pichai has resurfaced online, reigniting debate about how modern artificial intelligence systems develop unexpected capabilities. In the clip, the Google chief executive describes a language model that was able to translate Bengali accurately after minimal prompting, despite not being explicitly trained for that task. Pichai also acknowledged that parts of such systems function as a “black box”, meaning their internal decision-making is not always fully understood, even by the engineers who build them. The resurfacing has sparked fresh discussion around AI transparency, control, and public trust.

What Sundar Pichai said in the original interview

The comments were made during a 2023 appearance on 60 Minutes, where Pichai and other Google executives discussed the rapid progress of large language models. He cited the Bengali example to illustrate how AI systems can display abilities that engineers did not directly programme or anticipate.

The model was not learning Bengali from scratch in real time. Instead, it had already been trained on vast multilingual datasets during its initial development. What surprised researchers was how quickly the translation ability appeared once the right prompts were given. This phenomenon, where new skills seem to emerge suddenly as models scale, is commonly referred to as “emergent behaviour” in AI research.

This an ‘emergent ability’

In large language models, certain capabilities do not improve gradually. Instead, they can appear abruptly once a model reaches a particular size or complexity. Translation, reasoning, and few-shot learning are among the skills known to surface this way. While researchers can measure when these abilities appear, explaining precisely why they emerge at that moment remains an active area of study.





The clip’s resurgence comes amid heightened public concern over AI safety, regulation, and corporate responsibility. As AI tools become more visible in everyday life, older statements from technology leaders are being revisited through a more sceptical lens. For some viewers, the clip confirms fears about systems that outpace human understanding. For others, it reflects a normal and well-documented feature of complex machine-learning systems.

Divided reactions online

Social media responses to the resurfaced clip have been sharply divided. Some users interpret Pichai’s remarks as evidence that AI is becoming uncontrollable or autonomous. Others argue the comments have been taken out of context, accusing critics of exaggerating a technical concept to fuel alarm or undermine trust in AI development.

The renewed attention puts a familiar dilemma back in the spotlight. AI systems are becoming more capable at a pace few industries have seen before, but the way they reach their conclusions often remains unclear. That gap between performance and understanding is growing harder to ignore as companies like Google roll out more powerful models to the public.