Grew up in Woodstock. Lived in NYC, Mexico, Tokyo, France, now SF. Wrote for Wired. Work? Chooch AI
Artificial intelligence has been a brainless idiot for more than a half-century, but humanity has kept trying to replicate intelligence with the fervor of “Young Frankenstein”.
Now, technological rigor has delivered AI from the murky ferment of hype and science fiction. AI has become ubiquitous, but it is not close to human.
Alan Turing, credited with inventing AI, challenged humanity to create an entity so artificially intelligent that a human could not distinguish whether they were talking with a ”robot” or another human, but passing the “Turing Test” isn’t our goal.
The reality is, Amazon’s amazing customer service bot pretty much knows why you are chatting with it even before you type, that’s the kind of knowledge we want from our robots.
AI takes a completely different approach from humans to tracking packages and playing chess. Yes, IBM created Deep Blue and beat humanity at chess, but Deep Blue isn’t going to pass the Turing test.
Deep Blue doesn’t mimic humans, but instead searches every possible move into the future of the game, while the last human chess grandmasters rely on pattern recognition, their brains absorbed through repetitive play.
To AI, tracking packages and playing chess are primarily the same, mathematics. Artificial intelligence models are generated by algorithms.
An AI model of kung fu is just math, too, but what is kung fu for you and me? In the film The Matrix, the protagonist Neo settled back in the VR chair on the good ship Nebuchadnezzar to learn kung fu.
He plugged his head into the network and his limbs twitching as he experienced the martial art. When the install was over, he opened his eyes and said, “I know Kung Fu.”
Neo’s kung fu moves on Morpheus were convincing, but what did Keanu really know? And what is knowledge? That question has been pondered by philosophers for thousands of years.
The Rationalists would argue that you can know things without experience, while Empiricists see experience as the only true source of knowledge. As Plato put it, “Knowledge is not given by the senses but acquired through them as reason organizes and makes sense out of that which is perceived.”
One could say the same of AI, Socrates.
AI can acquire kung fu moves. Train the AI with kung fu perceptions obtained by kung fu sensors and the AI acquires kung fu knowledge by organizing those perceptions in a machine learning framework.
Plato would be proud.
Once the AI knows kung fu, the model can then be installed in Neo’s brain or any “edge device”. When a kung fu move is detected, a kung fu response is triggered.
Here’s a more realistic and reductionist example: training an AI to identify and count cells through a microscope. Researchers have been engaged in this monotonous activity for over one hundred years, but now AI can count cells 100x faster and more accurately. AI does seem to fit the Platonic Test for knowledge both in training with skills, AI “organizes and makes sense out of that which is perceived.”
Neo “perceived” his kung fu moves in his VR cram session. Mr. Smith, his AI
antagonist, knows kung fu, too. How Mr. Smith gained his wisdom is unknown, but Mr. Smith is part of the Matrix. T
herefore, the Matrix must also know kung fu, perhaps having learned it from old Bruce Lee, Jackie Chan and Jet Li movies, along with martial arts instructional videos. Since the Matrix can serve up those ice cubes of kung fu knowhow, Neo and Mr. Smith may have learned kung fu from the same server, importing the same AI models.
It’s a fair assumption since they are so well matched. Only the Oracle would know the answer, but as we witnessed in the movie, Mr. Smith is superior in one significant sense.
Mr. Smith can be replicated, and his clones also know kung fu. Similarly, AI models that count cells can be installed on millions of microscopes. Tesla’s onboard processors are clones, and every time pedestrian detection is updated in a Tesla, the AI gets better.
Soon, every car in the world will “know” to avoid pedestrians. It may feel odd to think that cars can know anything, but we can update cars to improve their sense of road signs, landmarks, and whether it’s about to rain. We can update refrigerators to know new snacks and give robots an edge in picking the ripest strawberries.
Chess-playing or cell counting, kung fu fighting or bots chatting about packages, it’s all about training AI models with mad skills. This is how we are iteratively giving machines knowledge, like kung fu.
Conversely, humans are learning iteratively from machines. More than 200 million Nintendo Gameboys have been sold and each of them inject the Tetris effect into our biological neural networks.
If “knowledge is not given by the senses but acquired” then Tetris is acquired from machines just as much as flight simulators and phone apps upload artificial intelligence models to your brain.
Microscopes and cars learn differently from you and Neo, but in each case, knowledge is acquired. Humans can train machines to understand the word “mama” and they can train babies, too.
When we compare human and AI knowledge acquisition, however different the process is, the lines become blurred around the final result, which is making sense of perceptions.
AI can know kung fu, pedestrian avoidance, or what cells look like.
Sure, Morpheus explains to Neo in the Matrix that AI basically takes over the world and turns humans into batteries, but that’s not happening next week or anytime in the foreseeable future. AI is able to perceive, organize those perceptions, and know things, and that alone is a giant leap forward for artificial intelligence.
We can now label any object in a frame of video, run the videotape, and quickly generate accurate AI models of anything visual (Disclosure: I Work at chooch.ai) in any spectrum of light, from microscopic to geospatial.
We can also label kung fu moves, resulting in AI that can then make sense of perceptions like this move is “black tiger steals heart” and that move is “white snake spits venom.”
Combine that with mechatronics, and it’s not hard to imagine a Boston Dynamics robot with kung fu moves, and eventually, a Mr. Smith.
Neo eventually defeats Mr. Smith by using his imagination to bend reality and that is the strength of human beings. We humans don’t turn perceptions into models.
You might be able to spot a microscopic imperfection in a painted metal surface, but you can’t do it endlessly, a million times per minute, while that surface is running at high speed through an industrial process. An AI can, but what an AI cannot do is shape the future with imagination.
AI can spot scratches better, faster, endlessly. An AI can be copied and replicated to infinity. Thousands of cameras can stream their video to mathematical models that know what a nanometer-long scratch looks like or how an intruder acts when coming through a window of a house. Those cameras can be watching for scratches in factories or for burglars all over the world every millisecond of the day.
Humans know what a burglar or a scratch looks like, and although humans can’t export and clone AI models, they can create symphonies and solve problems. We hear about AI being creative, but non-linear humans have flashes of insight that spawn new realities, AI being one of those paradigm shifts.
Yes, AI will keep learning and humans will keep imagining the future, what we are going to collectively build. Cypher betrays his shipmates in the Matrix because he regrets taking the red pill and yearns to return to blissful ignorance, but humanity is not a weak, dithering lot, imprisoned by reality. The concept of the Matrix itself is a leap of faith, not an illusion generated by math.
We don’t want movies from AI. We just want to know where our packages are and how many cells are on a slide, so we don’t have to count them. AI already can do these things faster and more accurately than we can, and various AI platforms are acquiring more knowledge every day, from visual to language to online behavior. AI can identify more celebrities faster and more accurately than any human, and pretty soon, AI is going to know pretty much everything.
It’s a no brainer.