Innovation is a magical thing. Some inventors who have made great achievements are not more diligent and hardworking than their peers, but they can often pick up unexpected treasures.
The most notable major innovation event in the world recently is the birth of the generative artificial intelligence model ChatGPT. It was developed by OpenAI (Open Artificial Intelligence Company), but it didn’t get special attention at first.
The four founders of OpenAI are all in their 30s and 40s. The CEO, Sam Altman, studied computer science at Stanford University and dropped out of school; the chief technology officer, Mira Muratti, is a young woman whose parents is an Albanian immigrant; President Greg Brockman went to Harvard University and MIT, but eventually dropped out; He immigrated to Canada and finally to the United States.
Two Americans without a degree and two foreign immigrants led dozens of R&D personnel to form a small company, adopted a technical route that big companies including Google were not optimistic about, and made the most amazing Shocking technology products. Can something like this be planned?
Why can’t great innovations be planned? The book “Why Greatness Can’t Be Planned” by Kenneth Stanley and Joel Lehman makes this issue clear. Their answer to this question comes from an AI algorithm.
For example, if you want to evolve beautiful pictures from some simple lines, or let a robot on paper walk out of a maze, or let a robot in a three-dimensional space learn to walk upright, what should you do?
The intuitive approach is to first set the evolution goal of the AI algorithm, and screen each step of the evolution. If it is close to the goal, it will add points, otherwise it will be eliminated. But in practice the effect of this approach is not good.
The algorithm Kenneth and Joel invented is called the “Novelty Search Algorithm”. This algorithm will randomly generate a set of solutions. By evaluating the novelty and retaining the solutions with high novelty, the solutions will undergo certain mutations like biological evolution, and so on, until the predetermined number of iterations is reached or the problem is completely solved. solve.
In the iterative process, this algorithm does not consider whether a solution is conducive to approaching the goal. It does not matter how weird and unreliable the output solution is, as long as it is novel, it will be kept. However, various experiments have proved that the solution found by this method can solve the problem best. It can generate the best-looking pictures, find the way out of the maze at the fastest speed, and allow the robot to learn to walk upright in the shortest time. Why is this?
One reason is that seeking novelty means seeking complexity. Simple solutions always appear first, and when you have tried all the simple solutions and want new ones, the ones that come out must be more complex ones. Complexity means having more information, which makes it easier to solve problems.
The more important reason is that the new scheme is a “stepping stone” to the updated scheme. It’s like you are hunting for treasure in a swamp. You have to step on more stepping stones to explore more places. You have to explore many places to be more likely to find what you want.
Take the example of teaching a robot to walk upright. If you focus on “walking upright” from the beginning, you will deliberately avoid the solution of making the robot fall over. But it is precisely these solutions to make the robot fall that can teach the robot to kick! If you learn to kick, you will naturally fall easily. But if you don’t kick your legs, how can you walk?
It is definitely a good thing for the novelty search algorithm that the robot goes from “will not fall” to “will fall”. The more the robot knows, the more advanced it is, and the more opportunities it has to acquire the skill of walking upright.
The pursuit of novelty ensures that there is a wide range of exploration, and good things will follow. Looking at the history of technological development, good things have never been deliberately planned according to a certain goal, but have been developed automatically one after another.
The Wright Brothers invented the airplane, relying on the bicycle technology at the earliest – countless people wanted to invent the airplane before, but no one thought that the “bicycle manufacturer” was the first to fly into the sky; microwave technology was originally used to drive the radar magnetron. Technology accidentally made the microwave oven; the first electronic computer used electronic tubes, but electronic tubes were not invented for computers at all.
Greatness is not the result of goal guidance, because the route to greatness is not always a straight line, and many times slow is fast-every time you just choose the next stepping stone, you can find treasure instead.
This is not to say that life should be aimless and go with the flow. Although the novelty search algorithm does not presuppose specific goals, it is guided by values, which are novelty and fun.
For example, a child thinks watching TV is very interesting at first, and the parents are very worried about it, thinking it is a waste of time. But the child will soon find that playing games is much more interesting than watching TV, so he will turn his energy to playing games. As long as his vision is high enough, sooner or later he will find that there are many more novel and interesting things in the world than playing games.
That’s right, it is not ordinary people who can really persevere in the pursuit of novelty and fun. They can always see the next stepping stone, and sooner or later achievement and utility will follow.
If you have identified what kind of treasure you want to get from the beginning, it will be difficult for you to get this treasure; those who finally get the treasure are just looking for the next stepping stone, and all they get are unexpected treasures.
To be novel is to be good, to be out of the ordinary is to be excellent, and to be interesting is to have fun.