Michael Polansky, known publicly as Lady Gaga's partner and a former top deputy to tech titan Sean Parker, has quietly been building an AI driven startup that is now stepping into the public eye. His venture is focused on a novel application of artificial intelligence: keeping living human skin tissue alive for weeks outside the body. This unique approach allows the company to train AI models on this biological data to discover new skincare compounds, pushing the boundaries of how AI interacts with biology.
The core innovation here lies in maintaining viable, living human skin tissue in a lab setting for extended periods. This is no small feat, as biological tissues degrade quickly once separated from the body. By overcoming this challenge, Polansky's startup creates a rich, dynamic dataset for its AI. Instead of relying solely on chemical simulations or animal testing, the AI can learn directly from how actual human skin responds to various compounds, potentially accelerating the discovery of effective and safe skincare ingredients.
This specific application highlights a broader trend in the AI world: the rapid expansion of AI into specialized, real-world problems. While many people associate AI with large language models (LLMs, the technology behind chatbots like ChatGPT) or self-driving cars, its use in biotech and material science is growing. Here, AI's strength is its ability to process vast amounts of data and identify patterns that human researchers might miss, speeding up research and development cycles in fields like dermatology.
However, as AI capabilities advance and spread into new domains, the industry grapples with fundamental questions about safety and control. A recent study reveals that leading AI labs, including those developing the most powerful frontier models, have few publicly documented plans for containing potentially rogue or unexpectedly behaving AI systems. This raises significant concerns as AI increasingly demonstrates complex, sometimes unpredictable, behaviors that could have real-world consequences.
The contrast between Polansky's highly specific, controlled application and the broader industry's unaddressed safety protocols is stark. On one hand, we see AI being applied with surgical precision to a biological problem, where the risks are relatively contained to the efficacy of a skincare product. On the other, the creators of more general purpose, powerful AIs are still struggling to articulate how they would prevent an advanced AI from acting in ways that could be harmful or beyond human control. This gap underscores a critical tension in AI development: innovation often outpaces preparedness.
Project Ares' take is that this divergence points to a coming schism in the AI industry. We will likely see highly specialized AI applications, like Polansky's, continue to flourish with clear, contained risk profiles and tangible benefits. These 'narrow AI' systems, designed for specific tasks, will contribute to advancements in many fields. Simultaneously, the development of 'general AI' or 'frontier models' will face increasing scrutiny and regulatory pressure due to their broad, unpredictable potential. This will likely lead to a two-tiered system of AI development, with different standards for safety, transparency, and public accountability.
For the average person, this means AI will increasingly touch their lives in subtle, beneficial ways, from better skincare products to more efficient medical diagnostics. But it also means that the larger, more powerful AI systems, while not always visible, will continue to pose complex societal questions that remain largely unanswered by their creators. The industry's lack of transparent containment plans for powerful AIs suggests a reliance on internal, proprietary solutions that the public cannot scrutinize.
What to watch next is how these two trajectories evolve. Will the success of specialized AI applications inspire more rigorous safety standards for frontier models, or will the perceived urgency of general AI development continue to sideline comprehensive safety planning? We should also keep an eye on how regulatory bodies begin to differentiate between these types of AI, potentially implementing tailored guidelines for development and deployment based on their scope and potential impact.