How intelligent do AI companions actually need to be?

In this episode, we wanted to explore what sort of intelligence we are actually looking for in our AI tools and companions. In the 1960s, Joseph Weizenbaum developed ELIZA: a rule-based chatbot therapist that kept users talking about themselves by matching keywords from user input to its scripted responses. ELIZA’s creator was surprised and shocked to find that people attributed intelligence, understanding, and other human attributes to this fairly simple pattern-matching program.

Now, nearly 60 years later, we’re actually being sold computer software with the promise of intelligence and reasoning. And while technologists again debate whether large language models are actually intelligent or just stochastic parrots, the AI companions they helped build are having very real impacts on people’s lives, even contributing to suicide. 

Our complicated human lives are so much more than the intelligence measured in standardized tests. And popular sci-fi depictions of AI companions in TV shows like Star Trek or movies like Her also reveal that our aspirations for machine intelligence often go beyond reasoning. Do we actually need AI or AGI that is superintelligent, or are we just craving a somebody or a some-bot that will make us feel heard and seen – something that the ELIZA chatbot achieved with a far simpler algorithm? And what kind of intelligence are the AI creators and investors actually looking for?

Our discussion is also an invitation to rethink the language we use for (artificial) intelligence and to reflect on the dark history and shadows of the field.

If you’d like to wonder and wander with us, join Tethix firekeepers Alja Isaković and Mathew Mytka in this meandering exploration inspired by the latest Pathfinders Newmoonsletter. In addition to the resources explored in the Newmoonsletter, we recommend exploring:

• ELIZA chatbot

• Stochastic Parrots paper by Bender et al.

• Let’s forget the term AI. Let’s call them Systematic Approaches to Learning Algorithms and Machine Inferences (SALAMI)

• The TESCREAL Bundle 

• Livingry by Buckminster Fuller

• Phronesis

• Tethix Mirrors

Related episodes of the Pathfinders Podcast:

• Why aren’t more people using AI conversational interfaces for conversational learning?

---

You can learn more about the Tethix pathfinding adventure at: https://tethix.co/pathfinders/ and support this podcast on Substack: https://tethix.substack.com/subscribe