At Rise of AI 2017 in Berlin, Ronnie Vuine argues that today's AI does not yet exist as popularly imagined and is, in his words, a fantasy, and explores why that fantasy is nonetheless what makes AI exciting. He says the field has gotten very good at pattern recognition and, in his own company's work, at reinforcement learning that links pattern recognition to acting systems, but that no one has any idea how to build the actual thinking that would take context, culture and long-term deliberation into account, and that this gap cannot be closed by scaling neural networks with more data. He traces this excitement to science fiction, telling the story of 1920s writer and rocket pioneer Max Valier, whose fictional hero anticipates the atomic bomb and who later dies testing real rocket engines, a lineage he connects to Wernher von Braun and the postwar American space program. He then reads a passage from William Gibson's Neuromancer to show how literature intensifies technological fantasies, arguing Gibson did not predict the future but helped create it, and closes with a quote from Mark Fisher on fiction and reality merging.
Key takeaways
- Vuine argues that today's AI is not the thinking AI of popular imagination; current systems do pattern recognition and reinforcement learning but nothing that plans or reasons with context.
- He says scaling neural networks with more data will not close this gap, because pattern recognizers are not thinkers.
- He connects AI's excitement to a class of technologies, including rocketry and VR, that promise to free people from a specific human limitation.
- He uses Max Valier's life, from writing a 1919 rocket novel to dying in a 1930 rocket engine test, to trace a direct line from science fiction to the American space program.
- He argues William Gibson's Neuromancer did not predict AI's future but helped intensify and create the fantasy that drives people toward it.
Detailed notes
Why AI does not yet exist
Vuine opens by noting that despite the conference's name, there are no AIs in the room: AI, in the strong sense, has not been figured out. He says current systems are very good at pattern recognition (classifying pixels as "giraffe") and, in the reinforcement learning his own company works on, at connecting pattern recognition to systems that act and improve from feedback. What is missing is the "little black box" that would actually think: taking context and cultural background into account, making long-winded plans, setting a strategy aside for years and later returning to it. Nobody today knows what to put in that box, and he stresses that this cannot be solved by scaling, by throwing more data at neural networks, because they are pattern recognizers, not thinkers.
AI as science-fiction fantasy
Vuine argues that this gap between the AI that exists and the AI people imagine is not a weakness but precisely what makes AI exciting, because it has lived as a fantasy in science fiction for a long time. He frames a class of technologies, including AI, space travel and VR, as distinct because each promises to free people from a specific limitation of the human condition, such as being bound to the ground or to a single body, and science fiction gives these possibilities a compelling narrative pull that he calls "the globe."
Max Valier: from rocket fiction to rocket death
He tells the story of Max Valier, a science-fiction writer born in Boston in 1895 who published a science-fiction novel in 1919 (the title is unclear in the recording), whose hero builds flying technologies and, angered by rejection, destroys the earth with a device resembling an atomic bomb. Ten years later, in Germany, the same Valier is building real rocket engine prototypes and testing whether they can generate enough lift for orbit; he dies in 1930 in a test explosion, becoming, in Vuine's telling, the first person killed by spaceflight technology before it existed. Vuine draws a line from Valier's rocket-mail pitch to a young Wernher von Braun, who later built the V-2 ("Aggregat 4") for the Nazis and whose team was subsequently brought to the US to help start the American space program, a direct lineage from 1920s fiction to modern rocketry.
William Gibson's Neuromancer and "the globe"
To show how literature intensifies a technology's excitement beyond its physical constraints, Vuine reads a passage from Gibson's Neuromancer describing a character experiencing another person's full sensory input through a device. He notes the novel is highly sexualized and written, in his reading, for adolescent boys, but argues this is exactly Gibson's method: connecting real technological possibilities to what excites people to intensify the experience of AI and VR in fiction. He points out Gibson foresaw advanced AI but not the cell phone, concluding Gibson was not a prophet predicting the future but someone who helped create it by making certain technological fantasies impossible to let go of.
Fiction as sorcery
Vuine closes by describing what people in the room, and AI practitioners generally, are doing as a form of sorcery: using signs and symbols, science-fiction fantasies, capital and hard work to try to make an imagined future real. He ends with a quote from the late theorist Mark Fisher, from "Flatline Constructs," on fiction forming a "resonant contact" with the world rather than merely paralleling it.
Quotes
We won't scale our way out of this ignorance.
Ronnie Vuine
He didn't foresee the future, he created it.
Ronnie Vuine
Questions answered in this talk
Why does Ronnie Vuine say AI does not yet exist?
According to Ronnie Vuine at Rise of AI 2017, current systems can do pattern recognition and reinforcement learning but no one knows how to build the component that would actually think, taking context and culture into account, and this cannot be fixed by scaling neural networks with more data.
What does Max Valier's story have to do with AI hype?
According to Ronnie Vuine at Rise of AI 2017, Max Valier wrote a 1919 science-fiction novel featuring a bomb-like superweapon, then became a real rocket engineer and died testing rocket engines in 1930, illustrating how science-fiction fantasy can pull people toward genuinely dangerous, difficult technological pursuits, in a line that runs to the American space program.
How does William Gibson's Neuromancer relate to excitement about AI, according to Ronnie Vuine?
According to Ronnie Vuine at Rise of AI 2017, Gibson used real technological possibilities and intense, sexualized narrative to make AI and virtual reality feel exciting, and in doing so helped create rather than merely predict the fantasies that now drive people to work on these technologies.