Summary
Highlights
The BBC World Service's 'The Engineers' program, hosted by Caroline Steel, broadcasts from Imperial College London to discuss the technical revolution of artificial intelligence. The episode introduces three leading engineers in the field who will share their work and its implications for humans. Examples of AI's current capabilities, such as defeating Go masters, writing movies, and discovering new antibiotics, are highlighted.
The panel includes Paolo Pirjanian, CEO of Embodied, who builds emotionally intelligent robots for child development; David Silver, Principal Research Scientist at Google DeepMind, known for leading the team that developed AlphaGo and working on artificial general intelligence; and Regina Barzilay, Distinguished Professor for AI and Health at MIT, who made breakthroughs in early breast cancer detection and antibiotic discovery.
Regina Barzilay shares her personal motivation for shifting her AI work to oncology after being diagnosed with breast cancer in 2014. She observed a significant lack of AI or even basic information technology in her treatment process. Her experience led her to question why MIT's advanced technology wasn't reaching hospitals, eventually collaborating with Dr. Connie Lehman to apply AI for early cancer detection.
David Silver explains how his background in the games industry led him to reinforcement learning. He found the AI used in games disappointing and sought to build 'real AI.' Reinforcement learning, inspired by how humans and animals learn through trial and error and feedback, allows systems to learn what works and what doesn't by assigning positive or negative numerical 'rewards' to actions.
Paolo Pirjanian, driven by experiences of alienation as a young person in foreign countries, created a real 'imaginary friend' in the form of a robot companion. This robot aims to address the growing need for companionship and therapy, particularly for children on the autism spectrum. The robot helps children practice social skills like eye contact and turn-taking, fostering deep relationships and aiding integration into society, without replacing human contact.
Regina Barzilay elaborates on what AI can do in cancer understanding that humans cannot. She highlights the challenge of uncertainty in diagnosis and treatment. Her own late diagnosis, despite earlier mammograms containing signs of cancer, demonstrated AI's ability to process complex data and remove the 'guessing' from medical decisions, detecting anomalies invisible to the human eye.
David Silver explains why Go was a much harder challenge for AI than chess, despite its seemingly simple rules. Go's complexity lies in its layers of intricate patterns and the human intuition and creativity required to play it, rather than just tactical lookahead. AlphaGo's success required major breakthroughs to mimic this intuitive sense of how outcomes would unfold over many moves.
Regina Barzilay details her team's breakthrough in discovering a potentially new antibiotic using AI. This was a challenging area for pharmaceutical companies due to economic factors. By feeding the machine thousands of molecules and their effects on bacteria, AI identified a molecule with a unique mechanism of action that effectively kills drug-resistant bacteria like E. Coli and MRSA, which humans might not have created.
David Silver defines Artificial General Intelligence (AGI) as an AI capable of approaching any number of problems with intelligence, similar to diverse human skills. He emphasizes that AGI aims to solve multiple problems, not just a single one. He believes AGI is still a spectrum over many years and requires numerous breakthroughs to reach human-level intelligence.
Regina Barzilay discusses why AI is not more widely adopted in medicine despite its maturity for many tasks. Key issues include evolving regulations across different countries and the billing structures in healthcare. She explains that in the US, doctors using AI might actually lose money, as billing is often tied to time spent with patients, disincentivizing faster, more effective AI-driven treatments.
Paolo Pirjanian expresses optimism for social emotional AI systems to help humanity. He foresees robots supporting children with autism and also providing companionship and assistive care for the elderly to combat social isolation and promote independent living with dignity within the next decade. These robots could engage in activities like cooking and walking.
David Silver discusses Google's Gemini project, an AI aiming to handle tasks from tax returns to writing novels. He envisions AI not taking over human culture but serving as an incredibly powerful tool for human authors, accelerating creative processes. He references music authoring systems that drastically speed up songwriting, hoping for a future where AI and humans collaborate to create more amazing works.
The panel addresses a question about the British government's approach to AI regulation, focusing on safety by understanding consequences before development. David Silver agrees on the need for regulation but suggests a nuanced approach, as AI applications vary greatly (e.g., medicine vs. chatbots). Regina Barzilay offers a counter-perspective, arguing that overly cautious regulation causes suffering by delaying beneficial AI applications for incurable diseases.
Paolo Pirjanian jokingly, then seriously, discusses AI's impact on jobs. He acknowledges the risks of unregulated AI due to malicious intent but also highlights AI's strategic importance. He argues that slowing down AI development through regulation could put a nation at a disadvantage in a global 'arms race,' making international agreement essential but historically difficult.
A member of the audience asks about AI's potential to improve human performance in sports. David Silver confirms that significant research is underway, citing Google DeepMind's collaboration with Liverpool Football Club to enhance tactical strategies. He notes that sports, being highly refined, are open to new discoveries and approaches offered by AI.
A young audience member raises concerns that humans might depend too much on AI and learn less. David Silver envisions AGI as a personal teacher and assistant, helping humans learn more effectively. Regina Barzilay shares her experience using AI for writing, which removes the pressure of small mistakes and allows her to focus on ideas, advocating for a 'symbiosis' where AI enhances, rather than replaces, human skills.
Paolo Pirjanian concludes by imagining the impact if historical figures like Isaac Newton or Albert Einstein had access to today's AI tools, implying it would make them even more prolific. He predicts that within five years, individuals won't need to code but can simply instruct AI to accomplish complex tasks that currently take years, highlighting AI's potential to dramatically boost human productivity.