Episode 210 · 2023-09-08 · 51:28 · Original in Finnish
AI and me | Petteri Järvinen | Negotiator 210
Originally published as “Tekoäly ja Minä | Petteri Järvinen | Neuvottelija 210”
Author Petteri Järvinen discusses his book Tekoäly ja minä. It is a consoling book turned into a conversation: people cope, professions die and are born, and the Luddites have always been wrong. Underneath run two questions that do not resolve — how anyone takes responsibility for a neural network's decision when it cannot give its reasons, and what happens when bias is corrected by distorting the data. The practical takeaway is the line between summarising and extrapolating: a model is far more reliable doing the first than the second. Recorded in September 2023, early in the consumer LLM era.
Core theses
- A model is reliable summarising and unreliable extrapolating, and that line is the most practical thing in the episode.
- A neural network is not a knowledge system or a decision tree, so nobody can reconstruct the reasons for its decision — which is a responsibility problem, not a technical one.
- Correcting bias by distorting the training data trades one falsehood for another, and the correction is not transparent either.
- Recorded in September 2023: the specific tools have moved on, the two open questions have not.
Watch and listen
Key moments
- 00:00 — Petteri Järvinen on people coping with AI and against it
- 00:40 — Audio channels, audiobooks, speech and restorative silence
- 02:40 — Does a person's clock speed slow with age
- 04:11 — Linear television is rubbish
- 04:44 — Vision one: AI finally brings a great interface and a platform for delegation
- 05:52 — Voice control has been a disappointment
- 07:20 — The Flynn effect is worrying: attention spans and goldfish
- 08:43 — Vision two: personal dataism — assistants trained on your own behaviour
- 09:40 — ChatGPT's interface is only the beginning, and answers cannot be trusted to be optimal
- 11:20 — Shadows in the robot utopia
- 12:05 — AI summaries are more reliable than AI hallucinations
- 13:18 — Overreach: a parliamentary AI assistant
- 14:49 — Subroutines and specialised plugins as a way to reduce hallucination
- 15:33 — Programming as a civic skill: useful or not
- 16:08 — A history of AI, and Musk's Neuralink
- 17:43 — Musk's autonomous cars did not arrive in 2017, and still do not work
- 19:02 — Sandstorm, DARPA and California
- 20:28 — The trolley problem and AI ethics: Western law against China
- 22:48 — TikTok as an influence platform, and Meta studying eye movement
- 24:33 — Radical honesty: can a flood of content defend against a cancellation
- 25:51 — GDPR and privacy: a Dutch authority's AI pilot disaster
- 29:51 — A neural network's decisions are not easy to analyse
- 31:18 — AI discriminates, but attempts to fix it can distort reality
- 32:08 — Prejudice or rationality: artificial output coding is not transparent
- 34:28 — Nudging can be justified, and ESG steering can go wrong
- 35:38 — An index fund changing from a general index to an ESG-weighted one
- 38:08 — Why human work is not entirely under threat, and polarisation is the real risk
- 40:44 — The mystery of human intelligence: creativity builds on what exists, plus randomness
- 42:36 — How AI can program, the uselessness of low-level coding, and singularity dystopias
- 44:46 — Keeping attention on an audiobook by doing something alongside it
- 46:04 — The claim about unused brain capacity, and using intuition
- 48:00 — Combining human and AI: Google Glass did not arrive
- 49:33 — Timo Honkela's negotiation of meaning
- 50:27 — The book Tekoäly ja minä
- 50:55 — Inner Circle 18: Finland's churches in drone photographs
Summary
Author Petteri Järvinen discusses his book Tekoäly ja minä. It is a consoling book turned into a conversation: people cope, professions die and are born, and the Luddites have always been wrong. Underneath run two questions that do not resolve — how anyone takes responsibility for a neural network’s decision when it cannot give its reasons, and what happens when bias is corrected by distorting the data. The practical takeaway is the line between summarising and extrapolating: a model is far more reliable doing the first than the second. Recorded in September 2023, early in the consumer LLM era.
The practical line
Summarising against extrapolating. A model asked to condense material it has been given is reliable in a way the same model is not when asked to extend beyond it, and almost every disappointment in the episode’s examples sits on the wrong side of that line.
The two questions that do not resolve
A neural network is not a knowledge system or a decision tree, so its reasons cannot be reconstructed — which makes responsibility for its decisions an unsolved problem rather than a hard one. And correcting demonstrated bias by distorting the training data replaces one falsehood with another, without the correction being visible to anyone downstream.
Watch
The recording lives on the Neuvottelija channel: Tekoäly ja Minä | Petteri Järvinen | Neuvottelija 210. A Finnish edition of this episode is published at www.neuvottelija.fi.
In depth
The Neuvottelija AI editions carry a long-form write-up of this episode: English · suomeksi.
Go deeper
Explore the ideas in depth
Guides connected to this conversation, with frameworks and further reading.
Enterprise AI Agents: Economics, Governance and the Shift From Tools to Workers
A guide to enterprise AI agents — the real cost model behind agent work, why owning your own stack is becoming a strategic question, a working governance framework with approval gates and audit trails, the agent risk matrix, the EU AI Act timeline as it stands, and who captures the productivity gains. Grounded in a real multi-agent lab, two 2026 keynotes, and Neuvottelija conversations.
Nordic SaaS Valuation & M&A: How AI Reprices Software Companies
How software companies in the Nordics are valued and sold as AI moves inference into the cost of goods sold — where multiples stand in mid-2026, the metrics that get repriced, why vertical SaaS defends its premium, AI due diligence, and the shift from seats to outcomes. Grounded in Translink's SaaS valuation work and Neuvottelija conversations.
People and topics
Guests: Petteri Järvinen