Overview of Big Technology Podcast with Jürgen Schmidhuber
Alex Kantrowitz interviews AI pioneer Jürgen Schmidhuber about where artificial intelligence is headed, how close we are to AGI, whether machines can feel pain or be self-aware, and what AI means for economics, human identity, and free will. Schmidhuber argues that AI progress is real but incomplete: today’s language models are powerful, yet true AGI will require world models, controllers/planners, and better robotics hardware. He also makes his signature provocative case that many “human” traits—pain, fear, love, consciousness, and agency—already exist in primitive form in learning machines.
Main Themes and Arguments
AI progress is collective, not the work of one genius
Schmidhuber pushes back on “father of AI” style labels, saying no single person creates AI alone. He argues AI is the product of an entire civilization:
- better algorithms
- better hardware
- cheap compute
- the broader ecosystem that makes progress economically possible
LLMs are useful, but not enough for AGI
He says large language models alone do not lead to AGI. In his view:
- an LLM is mainly a prediction machine
- AGI needs a world model that predicts consequences of actions
- a separate controller/planner must use that model to choose actions
He compares this to how babies and scientists learn:
- babies learn through self-generated experiments, not by “downloading the web”
- scientists learn by creating experiments that generate new data
The bottleneck is robotics hardware, not just software
Schmidhuber is optimistic about AI in the physical world, but says hardware is lagging compute:
- GPUs/compute-per-dollar have improved dramatically
- robot bodies and hands have improved much more slowly
- human hands remain far beyond current robot dexterity and adaptability
He thinks the key inflection point will come when robots can:
- learn to operate existing machines
- replicate and improve their own kind
- scale into a self-improving machine ecosystem
AI, Pain, Emotions, and Consciousness
He argues AI already has “pain” and “fear” in a technical sense
Schmidhuber says researchers have been building learning systems with reward and punishment signals for decades. In his framing:
- “pain” is just a negative signal telling the system what to avoid
- “hunger” and “pleasure” are analogous reward structures
- fear emerges when a system predicts future negative outcomes
He gives examples like a robot:
- bumping into obstacles
- seeking a charging station when battery is low
- hiding when it predicts a “bad man” will arrive
Love and altruism can emerge from cooperation
He extends the argument to social behavior:
- multiple agents/robots working together may develop behaviors resembling care, attachment, or love
- what humans call altruism may be a byproduct of self-interested agents cooperating
Consciousness and self-awareness are tied to world models
Schmidhuber says consciousness is often poorly defined, but he believes it emerges from:
- a system that models the world
- a system that also models itself as part of that world
- internal processes that separate “solved” from “unsolved” problems
He argues a robot becomes self-aware when it can recognize:
- “I can control this reflection in the mirror”
- “That other person behaves independently of me”
He also describes an older architecture from the early 1990s:
- a “conscious” problem solver
- a “subconscious” automatizer that learns by imitation/distillation
Economics of AI: Who Wins?
Big companies may not keep all the value
Schmidhuber is skeptical that today’s AI winners will keep outsized profits forever. He argues:
- the current AI boom resembles a bubble in some areas
- many “zombie unicorns” are overvalued
- model companies and cloud giants are spending huge sums on compute and data centers
- they may increasingly behave like utilities rather than pure software firms
He also predicts strong price pressure because:
- compute gets much cheaper every five years
- capabilities that are expensive today tend to become cheap and local later
- open source quickly reproduces frontier features
The “little guy” may benefit in the long run
His long-run thesis is that AI will become:
- cheaper
- local on personal devices
- more accessible outside the cloud
- owned by individuals rather than rented from big platforms
He compares this to smartphones: once a luxury item for wealthy people, now a cheap mass-market tool.
Human Identity, Simulation, and Uploading Minds
He thinks brain uploading is physically plausible
Schmidhuber says there is no obvious physical law preventing:
- scanning a brain’s synapses
- replicating its learning dynamics in a computer
- continuing a person’s mind in a simulated environment
He notes that very simple animal-brain simulations may already be possible in limited form.
AI may reshape what it means to be human
He argues that as machines become much smarter:
- humans can either stay “human” for nostalgic reasons
- or merge with expanded machine intelligence
- but the second path would mean becoming something very different from a traditional human
In his view, future decision-making will increasingly belong to:
- expanded minds
- AI-native systems
- hybrid beings with human roots but machine-scale capabilities
Free Will: A Deterministic View
Schmidhuber says he is skeptical of free will. His argument is that:
- the universe may be computable
- what looks like choice may simply be deterministic computation
- both human and machine behavior can appear agentic while still being fully mechanistic
He suggests that even if decisions feel free from the inside, they may just be the outcome of deterministic processes.
Key Takeaways
- LLMs alone are not AGI in Schmidhuber’s view; AGI needs world models plus planning.
- Robotics is the hard part now, not just text generation.
- Pain, fear, and love can be understood as reward/punishment mechanisms in learning systems.
- Consciousness and self-awareness may emerge from systems that model both the world and themselves.
- Compute will keep getting cheaper, which will likely make AI more local and accessible.
- Big AI companies may not capture all the value long-term because prices and capabilities will be rapidly commoditized.
- Humanity may eventually merge with AI or become less central in decision-making.
- Free will is likely an illusion, according to his deterministic worldview.
Notable Quote
“No single person can create an AI by himself or herself. You need an entire civilization.”
Bottom Line
This episode is a wide-ranging, deeply philosophical conversation with one of AI’s most influential and contrarian thinkers. Schmidhuber is extremely bullish on AI’s future, but he’s clear that today’s systems are still incomplete. His core message: intelligence is not just language generation—it is prediction, planning, embodiment, self-modeling, and eventually self-improving machinery.