The website Talk to Transformer lets you to play with a new machine learning model that has been trained to write complete sentences. You type a snippet of text, hit “Go,” and artificial intelligence takes over, riffing off the sentence you’ve entered to generate grammatically-correct text in response.
Think of Gmail (and now Outlook) and its suggestions for how to respond, only way more sophisticated. The model, a neural net called GPT-2, has been trained on a massive corpus of texts to enable it to write a plausible follow-up.
I’ve been having some fun with it. In fact, GPT-2 has written the last sentence of every paragraph in this blog post that follows. Does it write better than me? Maybe. Does it write better than any editor I’ve ever encountered?
It’s a neat trick. And that’s the problem. It’s just a trick. Impressive as it might be, the machine does one thing: it predicts the next word, based on statistical probabilities. It can’t sustain a thought, it can’t reason, it can’t conceptualize. And when the next word doesn’t come, you can’t hold onto a question or an idea. It stays very much in the present.
The current state of AI is prediction without understanding. The current state of AI has not yet reached a state where the machine can achieve the level of human understanding necessary to properly and successfully interact with its target. We believe there is more than enough room for improvement in this area.
Developments in AI, impressive as they are, will soon hit a dead end if we continue to regard intelligence as a computational problem. Thinking and feeling are not math. Pattern detection and prediction will take us only so far and neither will approach the capability of a human. Just as computer programs cannot be downloaded into human brains, human minds cannot be re-programmed and deep learning will not be able to solve very complex problems.
We have hold of the wrong metaphor. The brain is not a computer. We are focused on the neuron when the real magic is the connection between neurons. The brain is physical stuff, but it is immaterial relationships between material neurons that give rise to consciousness. The interval between two musical notes is not physical, it’s a relationship – the immaterial soul of music. The neurons we are targeting here are the ones that connect us to the spiritual universe, not the one that tells your tv remote which channel to turn on.
The big challenges and the big opportunities are not the property of the first discipline that comes to mind. The next big break in AI, or climate change, or economic inequality, or space exploration, will depend as much on the humanities as science. They are outside the realm of the current discipline. I’m looking for the broken pieces. The area where I have some hope for both the humanities and the sciences is in education.
That’s why universities are the most exciting places to work today. Where else will we discover the better metaphor, unlock the invisible music between two pitches? The university in this painting is a child learning to use a spoon, a child of a like mind on a quest to find the best spoon. The youth’s quest has driven the material into form, coloured with a painterly sense of pathos.
Oh wow – okay.
Philosophy and physics: The future belongs to institutions that combine unlike disciplines to ignite such sparks the world has never seen.
CoolOps Blog 