According to how the company sells itself and the content of the media reports,人工智能It seems to be everywhere. However, despite the hype, there is almost no production today.actualUse AI. Of course, companies like Google and Amazon use artificial intelligence in their products to help you write email responses or recommend products that might be of interest to you. However, most people will still spend the whole day on AI without getting help or delegating work. Why is that? Where are all the accreditation bodies we have been hearing?

Short answer? They have not yet been built. In this article, we'll describe the barriers to creating artificial intelligence products and how to build them in the right way.

How to identify opportunities

The irony is that despite the extraordinary progress in the field, the industry is working hard to find artificial intelligence applications. Impressive with Open AIGPT-2The language model is an example. Based on a few words, it can write an entire article with themes, themes and preferred styles. Although this is really unbelievable, what is its real application? If the AI ​​is good enough, it has the ability to replace the author or even the script or news author. But it has not yet arrived there. So how do we use its current power?

The GPT-2 example illustrates how the industry is going backwards. We are blinded by the tremendous progress in this field and are desperately looking for problems that can be solved by artificial intelligence. In fact, we should think in reverse: To adopt a people-oriented approach, we must first understand the user’s background and needs. In this way, AI is another tool we can use-a very powerful tool that allows us to imagine further than we thought.

Double diamond frame

We develop products in a way that observes and discovers user needs at the beginning of the problem-solving process. This is part of the end-to-end process we capture using something called a "double diamond framework."

The goal of the first diamond is to ensure that we are solving the right problems. We do this by uncovering the core challenges faced by users and turning these insights into ideas. Once this is understood and the idea has been defined, we will turn to the second diamond. The goal here is to solve the problem in the right way. We design and build experiences by implementing the build-measure-learning process so that hard data can guide our priorities and validate our assumptions.

In the table below, the first three columns highlight activities and results. The final column describes how to consider AI at various stages of the dual diamond framework.

While we should not conclude whether "AI" should be part of the first diamond solution (steps 1 and 2), be aware that the great features provided by this tool set can lead to better ideas. For example, whenever we encounter users who are frustrated by having to engage in tedious or repetitive tasks, we can consider artificial intelligence. AI is great for automating these tasks.

Please pay attention to the order in which we think about the process: we start with the users and their frustrations, design ideas for problem solving, and then think of AI as a tool for designing solutions.

AI product execution:

Once you have defined how to solve the problem and how to meet the needs of the user, the next step is to build the solution. Here are some things to considerGuiding Principles :

1. Start small:Build a solution that increases your chances of success. This may mean that there is discipline from the outset and it does not solve the big problems that require complex AI solutions. Instead, a more effective strategy is to prioritize the solution to less complex ones because it increases the chances of success. This approach also provides the added benefit of building AI power within the organization.

2. Enhancement and automation:The term AI is synonymous with automation because AI allows the user to delegate repeated or unwanted tasks that were previously done manually to the computer. But sometimes users prefer使用 AI instead of fully automating tasks. There are two reasons for this:

  • False positives or false negatives have a high meaning: for example, mistakes can jeopardize personal safety or increase the financial risk of use cases, and the benefits of fully automated tasks may not exceed costs.
  • Enhancement can be an effective way to automate step by step. The value of some automation solutions can reduce time and enhance the ability of developers and models to learn faster.

Data is a distraction

In the design process of artificial intelligence solutions, there is a time to worry about too little data. At this point, the wheels are stopped and the entire project may be derailed due to the focus deviating from the user's problem. Instead, try to move to blind exercises that accumulate large amounts of data, create data infrastructure, data quality and clean exercises.

This is not to say that the data is not important, but it should not be at the expense of the user experience. Data, models, and user experience are interdependent and cannot be resolved independently. Spending data alone makes it possible to get the wrong data and create an invalid infrastructure. Building the model and exposing it to the user can help inform the data collection process. Many questions about the volume, state, and predictability of data can only be answered by building machine learning models. Whether the insight of the model is needed, useful and actionable questions can only be answered after the target user has reached them.

Finally, AI applications should be designed to collect their own data for expansion. Autonomous cars are a good example. The Autopilot is essentially a four-wheel drive application for collecting data. Every time the driver intervenes and corrects the AI, it learns valuable feedback. This improves the autopilot's predictions, increasing the number of times people choose to use this feature. As a result, more autopilot data was collected. AI application designers should build this virtuous circle instead of pre-collecting data.

in conclusion

For everyone with a background in product development, what we show here may be very familiar. It seems that the software industry has once again experienced another cycle of “reinventing the wheel” in the development of AI solutions. Starting from user needs,thenConsidering how artificial intelligence meets these needs, adopting a people-centred approach is a powerful way to innovate and build a user-friendly experience.

This article is transferred from medium,Original address