Be Unreasonable
“Fantasy remains a human right: we make in our measure and in our derivative mode, because we are made: and not only made, but made in the image and likeness of a Maker.” — JRR Tolkien
Reasoning models are deservedly all the rage. Ask a question, the model will think, and you'll get a good answer. Do this enough and you’ll notice patterns. The answers are good but they're rarely great. They'll uncover existing ideas and facts you didn't know because the models are the most complete encapsulation of all human knowledge ever imagined. They'll make all sorts of wild connections. They'll respond to input and put things together. But most of what we are getting is synthesis. They won't come up with something truly new.
We call our models Generative AI. It's better to call them Synthetic AI.
This is wildly useful! Most of what we do as knowledge work is putting together ideas and information in the right way. A lot of software engineering is just plumbing things together. Spreadsheets are collating. We can reason our way through all of this. It's an algorithm. A process. Just like arithmetic: we follow the steps and get the answers.
That's the nature of reason, and now we have reasoning machines. We can outsource all of this synthesis and focus on the 5% that's left. That's the transformative stuff that only comes along every once in awhile.
This stuff is un-reasonable.
Orville and Wilbur Wright were unreasonable. They were bicycle mechanics in Ohio. They had no formal engineering experience, no funding, and no support. As a teenager, Wilbur had a head injury during a hockey game from a bully who would later become a serial killer. He remained housebound after this and cared for his tubercular mother instead of attending Yale. Orville dropped out of high school to start a printing business.
After they had their ideas about airplane control mechanics, they discovered Kitty Hawk North Carolina by writing to the Weather Bureau for the windiest places in America. Then they travelled there via multiple trains and boats, few roads, and no Google Maps. They lived in a tent on the dunes with mosquitoes and seagulls for months. They did this for 3 years in a row, ignoring the expert calculations of the time, even building their own engines and wind tunnel, and still failing hundreds of times.
Lord Kelvin had declared flying machines impossible. Even if he was wrong, experts thought Sam Langley had the best shot - he was the educated Secretary of the Smithsonian and had government funding for his engineering project. But hardly anyone believed in him either. The Times opined that human flight may be possible in 1-10 million years.
No LLM would have recommended their path to flight — tent living, homemade wind tunnels, ignoring the experts and all accumulated knowledge. Only two unreasonable brothers, through grit and ingenuity, could have done that.
When you call LLMs from their API you can dial up parameters like temperature and get more variety and unpredictability in the output. AIs have some capacity for creativity, but they are tuned for synthesis. They're tuned to be reasonable.
Users see hallucinations as a problem when we're asking for knowledge and synthesis, which is most of the time. But out on the envelope of novelty and creativity, hallucinations are a feature. When LLMs hallucinate we shouldn't think of them as being wrong. They point us back to the special sauce that always marks the progress of humans and the unreasonableness of it. It is novelty. Outrageous grit and determination. Ingenuity and intellect. Feral creativity.
But always, always, always this involves human agency and unreasonableness. Galileo, Semmelweis, Van Gogh, Steve Jobs — every breakthrough in history has been totally unreasonable, just like the Wright brothers.
As AI takes off, we can give up a lot of the reasonable work if we choose to, and focus only on the unreasonable stuff.
Should we? The jury is still out: we don't know the relationship between all of the synthesis work we do and our agency and creativity. Joanna Maciewjewska said: "I want AI to do my laundry and dishes so that I can do art and writing, not for AI to do my art and writing so that I can do my laundry and dishes." On the other hand, Gustave Flaubert said: "Be regular and orderly in your life, so that you may be violent and original in your work." It's been thousands of years and the formula for ingenuity is still as fragile as a snowflake on a stove. We don't know if all the tedious stuff is the training we need for a cognitive muscle we can't afford to lose or if Joanna is right and we're wasting our creative output on the laundry.
But for the first time in history, we can find out. We can run the experiment. Machines are giving us 10x more reason. We must match reason by being unreasonable. By having 10x more agency.
Agency is about more than just resources. It's the willingness to pursue daunting things despite reasonable objections.
The human job is to be unreasonable. This is the nature of genius, what Tolkien saw as our divine inheritance. It is the ability to make a-new, sometimes entirely new worlds, as we are made in the image and likeness of the Maker.
What impossible things demand to exist? What new worlds are you refusing to make? What would you build if failure was unreasonable?
This is our purpose. This is why we've made these machines. Why we have our place above animals and below angels.
Make it. Be unreasonable.