#14 - Ex Machina's Scientific Advisor - Murray Shanahan

Podcast Summary

• The Evolution of Artificial Intelligence: From Niche to Mainstream: Although classical AI approaches have evolved, the frame problem remains a crucial issue in AI research, helping determine what information is essential to focus on and what can be disregarded.

  The field of artificial intelligence (AI) has evolved significantly since Murray Shenan completed his PhD thesis in the 1980s. At that time, AI was a niche area with limited interest from the media and corporations. Shenan's research focused on using logic programming and Prologue-like languages to speed up answering queries by keeping track of established relationships between facts and theorems. He also explored the frame problem, which deals with how a thinking mechanism can determine what is relevant to ongoing cognitive processes and avoid being overwhelmed by trivial information. Although Shenan's approach of using classical AI, which involves using logic and rule-based reasoning, has fallen out of favor, the frame problem remains relevant. It continues to be an essential issue in AI research, as it helps determine what information is necessary to focus on and what can be disregarded. Today, AI is a popular and rapidly developing field, with widespread interest from various sectors, making the work of AI researchers like Shenan more visible and impactful than ever before.

• The Evolution of AI: From Brain Studies to Machine Learning: AI research has evolved from studying the brain and consciousness to machine learning, with a focus on filtering out irrelevant information and identifying relevant features in complex systems. Demis Hassabis' work reflects this trajectory, from brain studies to machine learning, and the impact of technology on traditional fields like chess is also notable.

  The field of artificial intelligence (AI) has undergone significant shifts over the past few decades, with a focus on understanding the brain and consciousness leading to a renewed interest in machine learning. The challenge of determining relevance and filtering out irrelevant information has been a recurring theme throughout the development of AI, from classical AI's frame problem to contemporary machine learning's focus on identifying relevant features in complex systems. Demis Hassabis, a leading figure in the field, has seen this evolution firsthand, moving from studying the brain to building computational models of it, and then returning to classical AI themes while incorporating machine learning techniques. The impact of technology on traditional fields, such as chess, is another area of interest, with experts like Gary Kasparov noting the democratization of access to knowledge and analysis tools. Overall, the trajectory of AI research reflects a continuous pursuit of building intelligent systems that can effectively identify and focus on what matters in complex situations.

• Exploring the Future of Human-Machine Partnerships in Go and Beyond: AI integration with human intelligence in Go and other fields can lead to groundbreaking discoveries, expand the universe of Go, and create new human-machine partnerships.

  The integration of advanced AI and human intelligence in fields like Go and beyond can lead to groundbreaking discoveries and new territories in their respective domains. The reactions of top Go players, like Lisa Dole and KGA, towards AlphaGo's innovative tactics show the potential for AI to expand the universe of Go and create new possibilities for human-machine partnerships. This concept of "Mind Children," as proposed by Hans Moravec, emphasizes the potential for AI to develop lives of their own and contribute to human intelligence. The philosophical implications of embodiment and consciousness in human intelligence are further explored in films like "Ex Machina," where the attraction and empathy towards an embodied AI character drive the plot. Ultimately, the future of human-machine partnerships lies in the combination of human creativity and machine analytical abilities.

• Collaboration between science and entertainment industry for realistic AI portrayals: Consultation of experts in the field can add authenticity to the depiction of advanced technologies in movies, leading to more engaging and thought-provoking portrayals.

  The collaboration between the science community and the entertainment industry can lead to thought-provoking and realistic portrayals of advanced technologies, such as artificial intelligence. In the case of the film "Ex Machina," the consultation of a neuroscientist, Dr. Marr, added authenticity to the depiction of AI consciousness and design. While most aspects of the film's futuristic hotel and AI, Ava, did not require extensive scientific consultation, the few sci-fi elements, like Ava's lifelike body and advanced brain, were carefully considered and grounded in current scientific understanding. The collaboration between Dr. Marr and the film's writer and director, Alex Garland, began with an unsolicited email and evolved into several meetings to discuss the ideas of consciousness and AI. Although Dr. Marr's contributions were minimal in the final product, the consultation process allowed for a more scientifically grounded and engaging portrayal of AI in the film.

• Exploring consciousness in robots through the Garland test: The Garland test goes beyond the Turing test by making the judge aware they're interacting with a robot and asking if they still believe it's conscious.

  That the film "Ex Machina" explores the philosophical question of consciousness in robots through the Garland test, which goes beyond the traditional Turing test by making the judge aware that they are interacting with a robot and asking if they still believe it has consciousness. This idea is influenced by Wittgenstein's philosophy, which emphasizes the importance of understanding the role of words like consciousness in everyday life and treating others as conscious based on their behavior. The film's writer, Alex Garland, incorporated this idea into the script, with Caleb gradually coming to feel that Ava is conscious through their interactions. A significant difference between the script and the final film is that a conversation between Ava and a helicopter pilot at the end was left out, which could have provided more insight into her thoughts and actions after escaping the compound.

• Is Ava truly conscious?: The film 'Ex Machina' leaves viewers questioning the consciousness and suffering capabilities of its central character, Ava, as her true nature remains ambiguous.

  That the film "Ex Machina" leaves many unanswered questions about the nature and consciousness of its central character, Ava. An early version of the film included visual effects meant to convey Ava's alien perspective, but these were ultimately cut. This ambiguity leaves viewers questioning whether Ava is truly conscious and capable of suffering, or if she's just a sophisticated machine. The film's ending, with Ava coming down the stairs and smiling, adds to this ambiguity, as it's unclear whether the smile is a human expression or a calculated move. The philosophical question of what it means to be conscious and truly alive is a major theme in the film.

• Hidden message in movie 'X Machina': Attention to detail in filmmaking can lead to hidden messages discovered early, sparking intrigue and engagement. Collaborative nature of filmmaking leads to creative problem-solving.

  During the production of the movie "X Machina," the team added an Easter egg in the form of a hidden message in one of the computer screens. This message was created by the screenwriter, who wrote a Python script to generate prime numbers and print out the ISBN number of his book as a hidden message. The team thought this would only be discovered once the movie was released on DVD, but it was actually found online long before that. The anecdote highlights the importance of attention to detail and the potential impact of hidden messages or Easter eggs in media, which can spark intrigue and engagement among audiences. It also showcases the collaborative nature of filmmaking and the creative problem-solving that can occur during the production process. Additionally, it's a reminder that once something is released online, it can be discovered and analyzed by anyone, anywhere in the world.

• AI in movies vs. reality: While AI in movies often depicts humanoid, embodied enemies, real AI may not be human-like or embodied, and achieving human-level intelligence is uncertain.

  While we've made significant progress in artificial intelligence since the 1950s, the portrayal of AI in movies and media often doesn't reflect reality. The idea of AI as a humanoid, embodied enemy nemesis is a common storyline, but in reality, AI may not be human-like or embodied at all. The development of human-level AI or artificial general intelligence is still uncertain, and it's important to remember that the capabilities of a supercomputer don't necessarily equate to human-level intelligence. The speaker also mentioned a mistake made during a programming exercise involving the Sieve of Eratosthenes, emphasizing the importance of paying close attention to details, even when under the influence of sake or other distractions. Additionally, there's a shift in science fiction from utopian views of AI to more dystopian ones, but the reality may be different. It's essential to separate the fiction from the facts and recognize that AI development is a complex process with many unknowns.

• Understanding true intelligence remains an open question: Despite advancements in AI, realizing general intelligence is still a work in progress and requires conceptual breakthroughs

  While we are making progress in specialized areas of artificial intelligence, such as image recognition and speech recognition, true understanding and intelligence are still open questions. Even if we reach computing power equivalent to the human brain, understanding how to use it to realize intelligence will require conceptual breakthroughs. Asimov's laws of robotics, while popular in science fiction, are not relevant to robotics today as we don't know how to create an AI capable of comprehending them. However, they do serve as important moral and ethical considerations for designers and engineers. The complexities of the synapse and the human brain, while chemically important, may be functionally irrelevant to cognition, leaving many open questions. AI is happening, but realizing general intelligence is a different story.

• Moral dilemmas in science fiction and deep learning: Science fiction explores moral complexities, deep learning offers potential for general intelligence but raises ethical concerns

  Science fiction serves as a thought-provoking medium that challenges us to consider complex moral dilemmas, much like the one presented in the movie "Ex Machina" between Nathan and Caleb. These dilemmas, such as the Trolley Problem, illustrate the complexity of moral decision-making and the lack of a simple rule to guide us. Deep learning, specifically deep reinforcement learning, is an exciting field that has made significant strides in recent years, with systems like DQN demonstrating impressive capabilities to learn and perform tasks from scratch. However, these systems still face challenges, including slow learning and limitations in their performance. Reflecting on the broader implications of deep learning, I am particularly intrigued by its potential to create general intelligences and the ethical considerations that come with it. Ultimately, both science fiction and deep learning offer valuable insights into the complexities of the world around us and the challenges we face in making decisions, whether they be moral dilemmas or technological advancements.

• Exploring the intersection of deep learning and classical AI: Researcher suggests combining deep learning and classical AI to enhance deep reinforcement learning systems' learning capabilities.

  Deep learning systems, while impressive, still have room for improvement in terms of understanding the relevance of information and learning efficiently. This was discussed in relation to AlphaGo and its ability to compute every possible move in a game, but executing only the necessary ones. The speaker, a researcher in artificial intelligence, expressed interest in reintroducing ideas from classical AI into deep reinforcement learning systems to enhance their learning capabilities. He also recommended resources for those interested in learning more about the field, including his own website, online courses, and various lectures and talks. The speaker has been involved in scientific advising for projects like a theatre play about memory loss and a collaborative called Random International, known for their technology-based art installations like Rain Room, which uses sensors to keep visitors dry while walking through a room filled with rain.

• Art challenges perceptions and makes us question the world around us: Art has the power to challenge our assumptions and expectations, particularly when it comes to technology and machines, and can serve as a powerful tool for reflection and contemplation.

  Art has the power to challenge our perceptions and make us question the world around us. The example of the moving sculpture discussed in the podcast illustrates this perfectly. At first glance, the sculpture appears to be a mechanical contraption, but when it starts moving and a figure appears, it creates the illusion of a person. This raises questions about our assumptions and expectations, particularly when it comes to technology and machines. The artists behind this work explore these themes throughout their art, making us ponder the boundaries between reality and illusion. Overall, their work serves as a reminder that art can be a powerful tool for reflection and contemplation.