Turns out, we are huge fans of Simon Stålenhag for his highly imaginative and gripping visual and narrative universe. Most recent addition to our collection is “Things from the flood“, an illustrated novel from 2016, set to take place after the closing of the loop, an advanced research facility playing the lead in the preceding book “Tales from the loop“.
One particular page struck me as highly relevant to what we’re trying to achieve with our work at Jetpack Cognition Lab. It is reproduced below as a testimony, a mini-episode titled “The Russian Teddy”.
The episode contains a few elements that are real and worth a closer look. There is the notion of trashy AI-ware which leads to an overall expectation of a simulation of a personality, rather than a true character of a robot. There is the aspect of asymmetry in regulations across different regions and markets, leading to an inflow of dark imports from less regulated domains to more regulated ones. Market physics, yes. Thirdly, there is the desire of the human kid to go to the end and find out whether the robot’s behavior is really only simulated, or if there is something underneath overt behavior, which comes closer to the singular existential experience we, and all other biological life claim to have, coming out regularly when threatened with death by external agenthood.
To us, as the roboticists and general life-embracing creatives we are, this is precisely antithetic, a negative example, a scenario that we would like to avoid, and that we think we know how to avoid. It is the reason why we insist that robots need to be strongly grounded in the most basic perceptions upwards for everything they do, and be functionally honest. That means, that robots shouldn’t pretend to have spoken language competence, for example, when they don’t have a lot of many more basic audio-motor skills, that humans and animals do have. The basic skills in this example would be awareness of sound sources, their locations, the fundamental characteristics of a sound source, like something dangling in the wind, a machine whirring, or an animal or human doing some activity, up to distinguishing between a non-word utterance and spoken language proper. This is just to name a few. All of our own perceptions, especially those that enter the conscious mind, are always based on literally thousands of subordinate cues. This is not a bug, this is a feature. It is what makes our perceptions so incredibly robust, for all practical purposes (fapp).
Our hypothesis here is, that machines built in such a way, and only such machines, will get close enough to an appropriate behavior because there will be plenty of micro-cues and preceding evidence for the development of a situation, allowing it to understand and change its behavior long before any irreversible escalation.
The “Russian Teddies” were cuddly toys equipped with simple AI chips and a voice module. They were supposed to be able to talk to you, and they were supposed to at least appear to have a personality. In Sweden, AI chips were banned for commercial use, and most AI electronics were smuggled in from Russia, which apparently had a different view of artificial intelligence and artificial life.Simon Stalenhag, Things from the flood, translated with deepl.com from German edition, page 22
Die »russischen Teddys« waren mit simplen AI-Chips und Sprachmodul ausgestattete Kuscheltiere. Man sollte sich mit ihnen unterhalten können, und sie sollten zumindest den Anschein erwecken, eine Persönlichkeit zu haben. In Schweden waren AI-Chips für den kommerziellen Gebrauch verboten, und die meiste AI-Elektronik wurde aus Russland eingeschleust, wo man offenbar eine andere Meinung zu künstlicher Intelligenz und künstlichem Leben vertrat.Simon Stalenhag, Things from the flood, German edition, page 22