AI or Illusion? Hawking's Warning Rings True in 2025 AI now surpasses human-level performance and realism in multiple areas, including realistic conversations, voice generation, and image creation
Opinions expressed by Entrepreneur contributors are their own.
You're reading Entrepreneur India, an international franchise of Entrepreneur Media.
"Success in creating AI could be the biggest event in the history of our civilization. But it could also be the last unless we learn how to avoid the risks," said renowned physicist Stephen Hawking.
Today, Hawking's warning feels more relevant than ever. AI-generated content is rapidly flooding the internet, leaving audiences both amazed and confused.
A recent example occurred when actor Paresh Rawal announced his exit from Hera Pheri 3 Movie. AI-generated images depicting Pankaj Tripathi as Baburao quickly circulated, leaving many viewers convinced of their authenticity.
Addressing this concern, Tripathi himself emphasised the dual nature of technology. He noted that technology can be used positively or negatively, but consumers must stay vigilant about distinguishing real from fake.
But are we truly prepared to identify what's genuine?
The recent Mary Meeker's Artificial Intelligence Trends report 2025 highlights this leap. AI now surpasses human-level performance and realism in multiple areas, including realistic conversations, voice generation, and image creation. For instance, Stanford's AI Index 2025 Annual Report revealed that AI models in 2024 achieved an accuracy of 92.3 per cent on the Massive Multitask Language Understanding (MMLU) benchmark, overtaking the human average of 89.8 per cent. This test evaluates general knowledge and reasoning across diverse subjects, demonstrating AI's dramatic improvement from just 34 per cent accuracy in 2019.
AI's conversation beyond face, and voice
The realism of AI interactions has also reached new heights. In a recent test by researchers Cameron Jones and Benjamin Bergen, 73 per cent of human testers mistook AI-generated responses from GPT-4.5 (with persona) as human-created. Even more striking, an example Turing Test conducted in March 2025 showed participants were 87 per cent certain that an AI-generated conversation was human, noting, "Witness A had human vibes."
Audio generation has also seen remarkable progress. Companies like ElevenLabs have enabled realistic AI voice translations, reaching millions of global users. In two years, ElevenLabs users generated an astonishing 1,000 years of audio content. Spotify has integrated this technology, translating audiobooks into 29 languages and making global content accessible to hundreds of millions of listeners worldwide.
With these innovations, Kunal Varma, CEO and Co-founder of Freo, believes, "The greatest challenges lie in the magnitude and pace of AI-driven fake news, deepfakes, and manipulated graphics or videos, which can lead to confusion, mistrust in television, film, and written content, and serious financial consequences. Misinformation has the potential to go viral on social media and messaging platforms, making it problematic for the typical user to determine what's real and what's not."
Ankit Sharma, Senior Director and Head of Solutions Engineering at Cyble, feels the risk is especially high in Tier-2 and Tier-3 cities. "Communities are becoming more connected, but digital literacy initiatives have not kept pace with smartphone and internet uptake. When AI-created disinformation—particularly voice recordings or videos in local languages—is distributed in these communities, the information is often consumed without fact-checking. This makes it a powerful tool for instability, hysteria, or manipulation. In addition, such regions are usually dependent on closed messaging systems such as WhatsApp or Telegram, where identifying the source and tracking the virality of debunked information is even more challenging. One effectively crafted, AI-fabricated piece of misinformation can trigger real-world consequences, ranging from social tension to election manipulation or cyberattacks on critical infrastructure," Sharma emphasised.
To combat fake AI content, Varma suggested, "Private firms, digital platforms, and industrial bodies must work collaboratively on solutions—whether that be developing superior AI to detect and flag manipulated content, or connecting threat intelligence across organisations. Tech firms can also invest in rapid-response systems, and platforms have the opportunity to empower end users with simple tools to flag misleading content."
Ankush Sabharwal, Founder and CEO of CoRover, added, "We're seeing rapid adoption of AI-powered media forensics and content validation platforms. Tools leveraging Natural Language Processing (NLP), image forensics, and blockchain-backed content provenance are increasingly being integrated into the workflows of both government agencies and responsible media houses. These tools enable real-time detection of manipulated narratives, sentiment skew, and coordinated propaganda efforts."
