When an AI project turns into a disaster

"Entities must not be multiplied beyond necessity."

Willem Occam, English philosopher, 14th century. 

Remember Occam’s razor during AI projects. 

"Entities must not be multiplied beyond necessity." Entities can be any work that we do. We should not do the same thing twice. Without necessity. Everything that is made without necessity is useless. 

Popularly, the principle is sometimes paraphrased as "of two competing theories, the simpler explanation of an entity is to be preferred. In AI projects, that principle can take on a form. "Of two competing models, the simpler model of an entity, or work, is to be preferred."

Keep the orders that you give to AI. As simple, clear. And minimal. As you can. The orders that you give should involve only things that are necessary for work. Orders should be given using proper language.  So, grammar must be followed in that order. And don’t write the text. Just give orders to the AI. About what it should do. And leave the work to it. Don’t introduce complete work to AI if you want to use it as a helper. 

Remember Occam’s razor. Give only necessary information to the AI. Don’t turn orders unnecessarily complicated. And don’t fill orders using information that doesn’t belong to the mission. When AI writes business letters, don’t tell how the sun shines. Just give orders that involve only the information that the AI requires for that mission. When orders are long, there is a possibility. There are more mistakes than in short texts. 

And then. Beginning is always. The most difficult part of the project. 

Sometimes, or almost always, the AI project turns into a disaster. Then we start to find out what went wrong.  The answer can be that the AI is misunderstood. People had very high expectations about the AI, and when the AI didn’t meet everything that people expected, that was considered a failure. The biggest mistake about AI is to treat it as a real expert. In that case, people can give AI orders. To write some reports or texts. Then they simply. Copy-paste that text to an Email. Without even reading it before. We should treat the AI as a secretary. And if we use AI to make some reports. We should read that text before we send it forward. 

The second thing is this: we should train the AI. When we first take the AI into use. Or start the AI project. We must train the AI. We cannot use AI as a production tool without training that system. Even the simplest AI algorithms, like grammar checkers, require training. That they make things as we want. Training the AI means. People use AI in their everyday work. The AI requires access to data so that it can learn things. Without that access. The AI program, or project, will fail. So, if an AI program faced failure, one of the reasons for that could be that people simply didn’t use it. If people don’t use AI, the AI doesn’t learn things. 

And that causes failure. So, when we start our AI project, we must ensure. That people use it. And especially. At the beginning of the project. AI doesn’t have enough data. It can create things automatically . About its users. And in the beginning. The untrained AI can make multiple errors. We must realize that AI is not ready when we create usernames and passwords for those systems. Those systems need training. That they work correctly. They learn only when we use them. Same way. Workers don’t learn things if they don’t do everyday jobs. If we don’t allow them to get access to the necessary data. They don’t learn anything in their work. 

The AI requires clear orders to accomplish its duty. If orders are not clear. Or they are not properly written. That can cause misunderstanding. Training the AI requires time. And as all workers AI requires access to information, so that it can learn how to cooperate with humans. And at the beginning of the project. Work can be done faster and probably easier. Using the old-fashioned products. Same way. When computers came to offices. Qualified typists. It could make their work faster than computer users. But then. The computers beat the workers who used paper. 

Automatic grammar error detectors helped the work of people who were not so qualified typists. But when the nets came to the offices. Computer users can share their documents all over the network. They could make error correlation on the screen, and that didn’t require physical papers. Same way. As hackers can attack networks, people can steal physical paper. If a company uses millions of papers each year. Physical papers cause very big bills. 

The problem is: how to motivate people to use AI. Why should people use AI if it takes their jobs? That is one of the things. That people should think about. What would you do with an AI project if you knew that the successful AI project would cause you to be fired? The problem with working life is that. We should be over-effective. We must work all the time. We have no time to take breaks. The attitude that the AI offers the possibility. To fire half of the workers is dangerous. 

So, when people start their AI project, they should follow two golden rules. One is that workplace AI is for work use. The second one is this: follow Occam’s razor. If every part. Of the use of the AI. Keep orders that the AI gets as simple as possible. The AI requires clear and precise orders. Occam’s razor is the idea that Willem Occam introduced in the 14th century. That golden rule is to remove everything unnecessary from the AI and the orders that it gets. The use of AI should also be connected to the work. If the AI gets information that doesn’t belong to work. That can cause a mess. The AI doesn’t think as we do. It doesn’t separate the information. That is connected to working life. From the information that the AI requires for love letters. If AI is trained. By using the wrong type of information. The result is catastrophic. 

The AI project mean. The new tool is coming to the workplace. People need preparation. And they must understand that AI is a tool. The AI makes things that people order it to make. People need instructions for that project. They need to know what kind of orders they can give. And they must know what types of writing the AI requires. AI requires clear and direct orders. Those orders must involve only the necessary information. We should follow the orders. The philosopher Willem Occam wrote in the 14th century. People should not unnecessarily complicate things. Things should be introduced as simply as possible. That principle is called Occam’s razor. 

People should remove everything unnecessary. From the orders that they give to the AI. That means when we, users, want to use the AI. To write a letter to customers. We should give direct orders for writing text to the customer. We should not tell the AI. About things. Like how nice the weather is. The AI requires only orders about things. That should be in the customer service letter. Everything that doesn’t belong to those letters should be removed. And that causes one thing about the AI. Should the company limit the AI that they brought only to work use? The problem is that AI uses all data.  It gets. For learning. And if people write things like love letters using those AIs. That can cause misunderstandings. People should remove everything else. Than the things. They need to work. From the workplace’s AI solutions. 

https://en.wikipedia.org/wiki/Occam%27s_razor