I, For One, Welcome Our New Self-Driving Overlords

8 Replies to “I, For One, Welcome Our New Self-Driving Overlords”

1. Interesting. As usual all is not as it seems.

2. I guess having captcha identify a mailbox as a parking meter doesn’t really count.

3. I was re-watching a Jordan Peterson lecture the other day and he made an interesting point about intelligence and artificial intelligence in general. He suggested that without a “body” or a way to perceive the world that intelligence couldn’t really exist in any useful way. If intelligence is going to do anything it needs to be able to have a concept of what’s around it. That concept doesn’t have to be real or the same as what you and I perceive, but it can’t operate in a vacuum.

   1. Artificial Intelligence sounds clever if you don’t think about what the words mean. “Artificial” is not real, so artificial intelligence is not real intelligence. AI doesn’t know anything and it can’t think.

4. The article is about research papers in computer science, which is not where the state of the art in AI is, any more than research papers in computer science is where the state of the art in wide-scale distributed systems orchestration is.

5. Monty Python”For there’s bugger all here on earth”,Intelligent Life.
   The Old Testament,short version,God creates man,turns to the Light Bringer and says “Behold my work”
   Poor Henchman could not stop laughing and got bounced right out of headquarters.
   The mockery of mankind as designed and built by committee is well played.
   But hey,Artificial Intelligence has come a long way,once the fuel of unicorn farts and pixie dust is fully sorted,why its “Forward”,to our glorious future.

6. You make a good point. Part of the problem is Moore’s Law.

   We can’t model every neuron in the human brain and all its connections and interactions with high temporal fidelity. There’s no way to model ten thousand connections on a neuron, so we don’t. There’s no way to model the neuron output at a high enough sampling frequency, so we don’t.

   Instead, we rely on a mathematical model developed back in the late 1960s and early 1970s that uses a whole bunch of simplifications. A biological neuron sometimes fires, either a pulse or a train of pulses of varying intensity, or it isn’t firing. Instead of computing an instantaneous value we’ve modeled the *average* output of a neuron over a sampling period, so some value between 0 and 1 or between -1 and 1 represents the activity of a mathematical neuron as an average value. And instead of tens of thousands of neural inputs, we’ve limited the number of inputs to a neuron to be some subset of the previous layer of the neural network. And of course dividing the arrangement of the neurons into layers, inputs feeding the first layer, that feeding the next layer and so forth to the output layer, then feeding the difference between the expected output and the actual output back through the network to adjust its internal parameters.

   And even that big simplification was hugely expensive computationally, at the time. All the hype about things like “deep learning” algorithms are just these same neural networks, just with way more layers than we could do in the 70s or 80s because computation got cheaper.

   And this is what I mean about Moore’s Law. An equivalent in aviation might be some advance in materials science to create lightweight material like carbon composites. Imagine if people in the late 1800s had access to such lightweight materials; lighter materials combined with the new gasoline engines might have made it possible for an ornithopter to get off the ground and fly, flapping its wings and barely staying aloft. That ornithopter would have mimicked flight successfully, but that is only through sheer brute force and happenstance, not because it’s taking advantage of the principles of aerodynamics like the Wright brothers did.

   The neural network model in use since Minsky and Papert in 1968 was sort of a first attempt to use computers as a “wind tunnel of the mind”, and it has its uses, but it is hugely computationally expensive, equivalent to examining each feather of a bird in flight and missing the underlying principles of aerodynamics. It gets its answers through brute force and a lot of human guidance and thousands or millions of training examples. That’s good enough to beat the human world champion at Chess or Go, and we’re watching the late stage training sessions with self-driving automobiles.

   So yeah, neural networks are useful, but they’re sort of a half-way step towards drilling down to those underlying principles of conscious-dynamics. And they’re just a start. If you imagine that wrinkly cerebrum in a human brain all flattened out so that the wrinkles were gone, it would look about the size of a tablecloth; one weave on that tablecloth would be equivalent to a column of a few thousand neurons, and these columns are repeated all over that sheet. And one of these deep learning networks like Deep Blue or the AI in a self-driving car are the equivalent of ONE column of neurons, one weave on a tablecloth.

   And even getting that far has taken up an enormous number of computational cycles over the decades. It’s only even been possible the last couple of decades because the computers have gotten faster and faster. The algorithm is 50 years old and hasn’t gotten any better, because the model isn’t drilling down far enough and exposing the underlying principles.

7. One of the interesting intersects between Computer technology and Credentialism at the “Centres of Education”,is people are unable to reason.
   Given a result,from a calculator or computer programme,few ask the logical question,”Is this reasonable”?.
   Being numerically illiterate,few can do basic maths in their heads or on a napkin.
   The entertainment world has promoted Artificial Intelligence for decades,but reality is a little different.
   Everything works great,until that one step outside the predicted,programmed parameters.
   And that is how I came to realize,those obsessed with creating the “self driving car”,understood neither the limits of the technology,or the actual skills of driving.
   A double blind,hidden by pomposity,arrogance and group think.
   Now wonderful things are being created and every failure,when recognized and assessed takes us further.
   But the boasts,brags and using us as crash test dummies,are too much to stomach.