Long ago, it seems that the year 2019 was the year that artificial intelligence truly went downstream. Here we are talking about the AI more than ever we did before, and the media coverage has nowadays become less neutral and more positive, which was reported according to the 2018 AI index Report. Out of the general public, 81% thinks that AI constitutes the next technology revolution. It seems like everyone has an interest to “do AI”, but many remain confused about where to start or what does that phrase actually mean.
A study of the MIT Sloan Management Review states that only 18% of the organization has adopted the AI extensively within their offerings and processes – a number that seems high to us is based on the experience in the field. Here are the top five things that anybody who wants to work with artificial intelligence (machine learning, deep learning, neural networks, GANs etc.)
needs to know. Using these ideas to go out and build great things or pushing the back on the things that you don’t think should be built.
Start with the outcome, nit the technology.
Hearing the same story, again and again, is that somebody comes in the team from the upper management and says that he had heard that this AI thing is pretty cool, so it’s better that we start using it up. Then the teammate starts scrambling to put together with some kinds of AI action plan that will satisfy the important person by injecting AI into any and every process possible. With the new shinning technology, it’s easy to get caught up in the hype cycle and gets jumped on the bandwagon without thinking where it must be going. But using this technology just for technology sale is never a good deal.
- Your project is only as feasible as your data is accessible.
The technology of AI uses data to recognize patterns. Machine and deep learning models are the only good until the data use to train them is this pattern recognition. The outcomes you want to achieve from your AI project should be parallel to the relevant data.
But here’s the million-dollar question: from where do we get good data to create an awesome machine learning model!? There are three basic types of data: data of our own, data that is publicly available and data that you can buy. Data that we own is great as because it’s free and whatever we use to train won’t be used by any of the other competitors if they don’t have accessed the same data. Data that is publicly available is also free, but the problem is that it is available to everybody, so chances of being copied by the competitor’s increases. Buying the data is not ideal. You should be sure that the data costs into accounts as you are developing the project plan.
- Your project needs broad stakeholders to buy-in.
The majority of investment of money and time needs to be an executive champion, but the AI projects need to have a broader cross-organizational buy in to be successful. Without participating from the people who are going to use those tools or impacted by its work – you will not be able to build up an effective AI application.
This is where things use to get a bit sticky, and the AI project on which you are working turns into an exercise in the change’s management as much as in the technology implementation. The team that has to be concerned is being supported by the customer is going to have their workflow or even worse that they might be thinking that you will be building a system that would ultimately be replacing them entirely.
- Identify potential risks as the start of your project.
Once it was said that every technology company should have a “Chief Skepticism Officer” whose job would be to make holes in the potential projects that ensure nothing truly disastrous is to be released in the world. In addition to the most awesome job that is being titled all the time, this role is actually essential for AI projects.
Anybody who works with the AI should be aware of the lookout for potential problems with their models and application. The ability of AI to amplify the social bias on a huge scale as the susceptibility of AI of being hacked by the adversarial attacks that prey on its weakness and the privacy concerns are associated with the training models on the individual data are all valid risks that you must thoroughly explore.
- An AI project is never finished.
No AI model survives its own encounter with the real world; your model which was trained at a point on the time of real-world data snapshots which is going to begin to degrade in accuracy and performing the moment is released into the wild.
This is called concept drift, and it can be a major issue if it is not kept in check. There’s no true end to these AI projects, but if you’re doing it in a good manner, then it’s just a life cycle of feedback and retraining which is continuous.