Automata (2014): A self-repairing robot is somehow a bad thing?
Automata is the 2014 scifi movie by Spanish director and auteur Gabe Ibáñez. Antonio Banderas is NOT the sexy Puss in Boots but a sad, very sad, corporate insurance investigator drawn into an underground where robots use diagnostic reasoning to repair themselves. Probably what the rationale against self-repair (besides trying to one up Asimov’s Laws) is based on the assertion that diagnostic reasoning is an aspect of a larger, more comprehensive intelligence. It's not even really an uprising, it’s actually an exodus of the hated underclass of robots from the artfully collapsing cities. Robots were built to save the world from desertification but have failed in the planet-wide task and now fill in odd jobs or are simply homeless. One apparently has circumvented the inexplicable prohibition on self-repair, with self-repair somehow being the key to sentience. The company is liable, so off Antonio winds his way, frown-y faced, under steampunk blimps and into desert storms that would have looked good in Dune.
The movie is hauntingly beautiful but the plot is incoherent and the science narrative is dubious. Self-repairing robots are a danger to humanity? What would that do beyond put a few repair shops out of business? Microelectronic chips repair themselves and we don’t have Skynet.
Real self-modifying robots are discussed in detail in my Science Robotics article. As noted in the article, a prohibition on self-repairing doesn’t make sense because a self-repairing robot would be more efficient, as it could still keep working, and more economical, again because it could keep working. General purpose self-repair might require intelligence bordering on sentience, but current robots can already repair themselves by simply swapping out identical modules.
Probably what the rationale against self-repair (besides trying to one up Asimov’s Laws) is based on the assertion that diagnostic reasoning is an aspect of a larger, more comprehensive intelligence.
Diagnostic reasoning requires intelligence, but with the current level of artificial intelligence that intelligence is generally the designer’s intelligence rather than the robot’s. Consider that most diagnostic reasoning systems, such as for NASA’s Deep Space One probe and used for medical diagnosis, rely on explicit models. In Deep Space One, the model is a representation of how all the space probe’s components are connected and their causal relationships. In medical diagnosis, the model is usually a Bayesian Belief Net capturing the frequency in which diseases and symptoms occur. In both cases, the diagnostic system is just plugging numbers into the model created by human designers and getting results: Plug. Chug. No semantic comprehension. There IS research in learning what the structure of these models are, but that work is still in its infancy.
If a robot in Automata fixed itself, it would most likely be fixing itself only because a software designer had created the reasoning model and the hardware designer made the components interchangeable or had incorporated the redundant parts or made the part so that it was repairable.
But keep in mind, diagnostic reasoning isn’t the same as physically performing a self-repair. Again, the current level of technology requires that the parts, the actions, and the tools are specified and engineered for compatibility in advance. Artificial intelligence excels at planning actions and sequences of actions- as long as the set of possible actions are known. Again, check out the Science Robotics article.
Should you watch Automata? Depends. The visuals are beautiful and worth seeing. If you had some chatty friends over who talk through movies, then this could be a good background movie. If you’re looking for a 2001: A Space Odyssey or Blade Runner moment of visual transcendence, don’t bother.
- Robin
Recommendation: Watch it for the art school visuals but turn the sound off and make up your own version of what’s going on—your plot will be better and more interesting.
Click if you want to see