I particularly like the structure of this video. The clubgoer over and over drifts into philosophical arguments about the issue at play, only to run into dead ends. How can he talk about the issue...
I particularly like the structure of this video. The clubgoer over and over drifts into philosophical arguments about the issue at play, only to run into dead ends. How can he talk about the issue in a way that actually gets through to the person he's talking to?
The structure of the video is especially clever when you consider that it is essentially what a machine learning model goes through during its training. It sees a great multitude of possible...
The structure of the video is especially clever when you consider that
it is essentially what a machine learning model goes through during its
training. It sees a great multitude of possible paths, but most of them
are “wrong”, and the model slowly learns the ones that are “correct”.
(I refuse to call ML models “algorithms”. They are, in my
understanding, much closer to equations than to algorithms proper.)
They're definitely algorithms. "Algorithm" is a really broad thing: In ML model is trained using an algorithm, and this is quite clear given the basic structure of training represented as...
I refuse to call ML models “algorithms”
They're definitely algorithms. "Algorithm" is a really broad thing:
a process or set of rules to be followed in calculations or other problem-solving operations, especially by a computer.
In ML model is trained using an algorithm, and this is quite clear given the basic structure of training represented as high-level 'rules':
Look at training example and have model predict
Evaluate prediction
Use evaluation to update model
Repeat steps 1-3 N times
Even once trained, a function score(model, input) -> prediction must be run to get anything useful from it. That function, here called score, is also an algorithm.
I conceptualised it as someone save-scumming their way through a dialogue tree in a videogame but now that you point it out, ML is basically the same idea.
I conceptualised it as someone save-scumming their way through a dialogue tree in a videogame but now that you point it out, ML is basically the same idea.
Something that Oliver doesn't mention in this video, but I think is an important bit to mention, is that machine-learning algorithms are inherently unjust. They're just statistical models. They...
Something that Oliver doesn't mention in this video, but I think is an important bit to mention, is that machine-learning algorithms are inherently unjust. They're just statistical models. They have no real comprehension of the greater world so to them correlation and causation are the same thing. They have no place in governance and are dangerous every time they're used to make meaningful decisions about people's lives.
Of course, that all changes if computers gain the skill of general intelligence. But that's an invention with potentially world-ending consequences.
I particularly like the structure of this video. The clubgoer over and over drifts into philosophical arguments about the issue at play, only to run into dead ends. How can he talk about the issue in a way that actually gets through to the person he's talking to?
The structure of the video is especially clever when you consider that it is essentially what a machine learning model goes through during its training. It sees a great multitude of possible paths, but most of them are “wrong”, and the model slowly learns the ones that are “correct”. (I refuse to call ML models “algorithms”. They are, in my understanding, much closer to equations than to algorithms proper.)
They're definitely algorithms. "Algorithm" is a really broad thing:
In ML model is trained using an algorithm, and this is quite clear given the basic structure of training represented as high-level 'rules':
Even once trained, a function
score(model, input) -> predictionmust be run to get anything useful from it. That function, here calledscore, is also an algorithm.I conceptualised it as someone save-scumming their way through a dialogue tree in a videogame but now that you point it out, ML is basically the same idea.
It's essentially a depth-first-search across the problem space. Graph exploration algorithms are one of the simplest forms of AI.
Something that Oliver doesn't mention in this video, but I think is an important bit to mention, is that machine-learning algorithms are inherently unjust. They're just statistical models. They have no real comprehension of the greater world so to them correlation and causation are the same thing. They have no place in governance and are dangerous every time they're used to make meaningful decisions about people's lives.
Of course, that all changes if computers gain the skill of general intelligence. But that's an invention with potentially world-ending consequences.