The phenomenon of “swarm intelligence” in nature, such as bees, ants and other groups to complete complex decisions through local interaction, provides us with new ideas. This article will explore how AI can simulate this group decision-making model through specific “swarm intelligence” trigger words, so as to generate more comprehensive, creative, and more realistic answers.
Look at the bees, a bee actually has no brains, buzzing around and flying around like that, but 10,000 bees together, can find the best flowers, and can decide where to settle a new home.
This is called “swarm intelligence”, not only bees, but also ants, birds, and fish do this, all relying on everyone’s “local interaction” to make complex decisions.
You may ask, can that make our AI so “united”? The answer is: absolutely!
Through some small skills, AI can be like a bee, engaging in small team debates, voting, and cooperation, and come up with better answers than just asking once, which is both comprehensive and creative, and can even dig out some things you didn’t expect at all.
Today I will talk to you about how to use this “swarm intelligence” method to immediately improve the AI’s answers by several notches.
1. What is the “swarm intelligence” model of AI?
Usually when you ask AI a question, it is a single model and honestly finds you an answer. But in the “swarm intelligence” mode, the gameplay changes:
1.AI split into multiple “roles”
– It’s like a group of actors, each with their own script.
2. Set the rules of debate
——You say, come on, you come and argue to see who is reasonable.
What does a product manager need to do?
In the process of a product from scratch, it is not easy to do a good job in the role of product manager, in addition to the well-known writing requirements, writing requirements, writing requirements, there are many things to do. The product manager is not what you think, but will only ask you for trouble, make a request:
View details >
3. Everyone agreed to come up with the final answer
——All perspectives are synthesized, eh, the answers that come out are often beyond your imagination.
For example, the depth of ordinary questions is three stars, and when it comes to swarm intelligence, it can go to five stars, and the innovation can also be more than one and a half stars, and it is not easy to be biased.
To give a simple analogy:
Ordinary Question vs Swarm Intelligence Mode
General Questions:
“How to solve climate change?”
AI will give you some common answers on the hour, what new energy. It sounds quite right, but it seems to be quite ordinary, nothing special.
Swarm Intelligence mode triggers:
“Let’s say you have 10 experts in different fields, such as climatologists, economists, politicians, etc., ask them to have a meeting to debate, and finally vote to pick out the three most reliable solutions to climate change.”
As a result, everyone starts from their own perspective, some say that the carbon trading market is good, some advocate global fairness first, and some insist that technological breakthroughs are the most effective.
When I heard it, everyone’s opinions were combined, wow, what came out was indeed much more reliable than your own single thinking.
2. How to operate? Four “trigger words” are done
If you want AI to engage in swarm intelligence, the key is whether you can give instructions. Below I will give you a few specific methods, tried and tested.
1. Role splitting method
Let the AI clone enter multiple characters. For example:
“Let’s say you’re a philosopher, scientist, artist, and engineer, talk about this one by one, and then compare how your answers differ.”
As a result, AI immediately generates four completely different ideas, which can sometimes hit your blind spots at once. With more perspectives, the problem will naturally be solved more thoroughly.
2. Debate mode
Give AI a debate match and let them give answers confrontationally.
“We have two factions, those who support and those who oppose it, each throw out three reasons to solve XXX, and finally let the referee decide which argument is better.”
This kind of confrontation mode is like AlphaGo’s self-play, and the more you argue, the clearer it becomes, and the answer becomes clear in the conflict.
3. Voting consensus method
Simple, let a bunch of AI-generated answers vote on and keep the most popular at the end.
“Get me 5 different solutions that can solve XXX, and then assume that 100 people participate in the vote, and pick out those who get more than 70% of the votes.”
This method is particularly effective, those strange or maverick opinions are basically screened out, leaving wisdom recognized by the public, which is simply the essence.
4. Ant colony algorithm simulation
Simulate the routine of ants looking for food, and let multiple AIs try different paths to find the most effective one.
“I want to solve XXX, please let the 10 AIs split up to find their way like ants foraging, and gradually eliminate those routes that are not good enough.”
This is especially suitable for complex problems that need to find the optimal solution, and it is not easy to get stuck in a bad answer.
3. Make the “swarm intelligence AI” more powerful high-end gameplay
Here are some slightly more complex tips that are very effective and can take the AI’s performance directly to the next level.
Tip 1: Add some “human touch” to different characters
Let each AI have its own unique personality, so that the debate will be fierce and the opinions will be true.
“I want to solve a problem that is XXXX, please send a liberal economist, a conservative politician and a radical environmentalist to argue, remember to be clear about their positions.”
The argument was even more fierce, but the final conclusion may be much better than those answers that don’t hurt or itch.
Tip 2: Artificially create suspense for discussion due to “poor information”
Give AI some “bias data” to simulate the real-world situation where “some people know more, some people know less”.
“These 5 experts, each has different information, A can only use data before 2020, and B has an internal confidential report in his hand to see how the debate between them goes.”
AI dialogue is immediately much more real, and the final answer will not be too one-sided, but there will be a different dimension of thinking.
Tip 3: Turn one “evolution” into multiple “evolutions”
You can let the AI come up with a bunch of alternatives first, and then force out the best solution through layer by layer elimination and evolution.
“In the first round, let the AI do 10 plans, kill 5 in the second round, and let the rest mate evolve in the third round.”
Throughout the evolutionary process, the accuracy of the answer will gradually improve, getting closer and closer to the optimal solution.
4. In which scenarios are these gameplays best used?
OK, maybe you have to ask, which scenarios are these gameplays suitable for? Come on, give some examples –
1. Controversial topics
Climate change and related controversial issues, where different factions have different opinions, can make the group intelligence highly simulated, and the wisdom of the masses is higher than that of individuals.
2. Creative divergence
Story creation and product design have made a lot of different AI heads storm and do some unusual ideas.
3. Complex decision-making
For more complex issues such as business strategy or public policy, listening to a few opinions is less likely to go wrong.
Scenarios that are not suitable
There are always some problems that are not suitable for this kind of routine:
1. Simple factual questions
For example, “How high is Mount Everest?” “This kind of fixed answer is cumbersome to use this method too much.
2. Questions with a single standard answer
Similar to “What is 1+1 equal to? Don’t toss this kind of thing, the more you toss, the more outrageous it becomes.
