Episode Description
Summary
In this episode, Anastassia and Rae are continuing to discuss portrayals of AI in movies and fiction. This time, they discuss “Her,” a film that has achieved cult status among technologists, in which Scarlett Johansson performed the voice of Samantha, an Operating System that the main character, Theodore, falls for. Anastassia offers explanations of how language models work and how they differ from humans, as they can’t reason, build causal relationships, and ‘think’ in ‘what if’ scenarios. Rae discusses how schools and universities recognise AIs in essays and asks whether humanity utilises AIs for what they do best, rather than trying to fit these technologies into everything humans are capable of doing. Topics such as human loneliness and AI responses, using AI as a metaphor to describe human problems, the Necessity of going through challenges to learn and appreciate relationships, and the nuances of context are explored. Anastassia reflects on whether modern AIs can be freed from biases and what alternative technology architectures might offer.
Key takeaways
- “Her” has a cult status to many technologists building AI products and services.
- The movie offers another way to reflect on AIs, as here a human (Theodore) is a professional writer falling for an AI, which is only represented by a voice.
- Samantha does not exist today. In the movie, she is capable of learning. Today’s AIs don’t master causality and reasoning; they are frozen in time.
- LLMs don’t learn from counterfactuals/ in ‘what if’ scenarios.
- Samantha’s character offers insights into the distinction between humanity and a performative act.
- Marshall McLuhan was discussing how travel and rapid communication were shrinking the world. AIs might do the same.
- Siri was one of the first AIs allowing autistic children to receive information no one else wanted or could provide.
- AIs are a reflection tool to tell us about ourselves.
- Domain expertise is paramount for building AIs today. Universal AIs are currently notoriously difficult to implement.
- We must recognise human expertise to determine how and where to utilise AI.
- Regulators must find ways to incentivise investments in fundamental research to change the current architecture, rather than insisting on something that can’t be mediated due to the underlying mathematics (e.g., removal of biases).
Chapters
- 2:32 What is “Her” about?
- 8:23 The Movie “Her” isn’t a dark portrayal of AI.
- 11:32 There is no reasoning and understanding in today’s AIs.
- 17:31 We mistake Samantha for a human ‘just’ because of her voice
- 21:30 Human loneliness and complexity of emotions vs. cutting corners because an OS is always ‘on’
- 23:50 Marshall McLuhan and the ‘shrinking world’ hypotheses
- 24:40 Teaching AI and ethics through a metaphor
- 27:02 A new concept of consumerism when it comes to an ever-available AI
- 28:01 Siri as a communication companion
- 30:22 AI as a reflection tool to teach humans about themselves
- 33:13 Use of language in the movie “Her” and in current AIs
- 34:40 How do educators recognise plagiarism, and the role of context
- 37:42 Necessity to check sources and links when doing an LLMs-based search (Perplexity)
- 38:36 Domain expertise is essential in building AIs well
- 40:40 AI can look for patterns, but it can’t read for context
- 42:11 The difficulties of roboticizing a hand
- 43:49 To understand the maturity and implementability of a technology, we must look into the semiconductors’ roadmaps and research the IP portfolios of companies
- 45:25 We must invest in alternative architectures to optimise AIs
- 48:10 Universities aren’t the primary source of research today, Big Tech is
- 49:46 Are there ways around biases?