Rules are about, ”If this occurs, that should happen” and ”If that occurs, this should happen”. When it comes to complicated problems, rule based analysis gets out of hand fairly quickly. We probably face a future where most simple problems already have been solved. The problems that remain are either complex by their nature or poorly understood by humans. Rule based analysis almost always require a well-defined boundary problem, such as a language grammar or algebra in math. More complex problems, like self-driving cars, does not belong to this category. Predictions and probability based software is about training the algorithms with massive amounts of data in order to shape the characteristics of the predictions. This can be done to recognise faces, identify tumours that can be hard to detect in X-rays, and to translate foreign languages. Google uses AI in their translation service, and in other services too. Sometimes the results are good, sometimes they are plain funny, but we always understand what a translation wants to say. The best part is that there is still a lot of progress to be made in AI assisted translation. That is true for AI in general, as we have only started to explore this branch in technology.
Economics talk about supplements and complements in business. If the price on coffee drops, the price on milk and sugar goes up, because people will drink more coffee. Milk and sugar are complements to coffee. But if the price on coffee goes down, so must the price on tea, otherwise people will stop drinking tea. Tea is a supplement in this simple example given by Ajay Agrawal, an expert in AI. For Artificial Intelligence, organisations will use software and data to predict decisions, the results are actions based on those decisions. Decisions are in this model a supplement, while actions are a complement. It means the value of making decisions will go down — probably dramatically — while the value of taking actions will rise. Think-tanks, management consultancies and upper-staffers must change their way of working, from an analysis-decision making approach to thinking more about strategies and finding creative solutions; activities that will address the current situation and be more competitive. We will have a future where most upper-staffers will agree on the situation, not because their manager says so, but because the facts are hard to dispute. The other side of the problem is that few upper-staffers will know what to do about the situation. Creativity will, therefore, be a key ability and a sought-after talent.
Similar can be said about Artificial Intelligence in Requirements Engineering. The value generated from making analysis with the help of Artificial Intelligence will drop, while the value of the artifacts a project builds will rise. Simply put, projects will build better stuff in shorter time. Today, the analysis of content and quality is relatively modest in most projects, if it's done at all, and the value generated by these activities are therefore close to zero. This is not what the complement-supplement model in economics says. The consequences from taking poor business decisions in general have one outcome only, disaster, that is why the value from decision making as an activity has been high. It is the value of making high grade decisions that will drop with Artificial Intelligence, the value of poor decision making has always been zero or negative. The same applies to Requirements Engineering. The value from taking actions based on analysis results will probably increase tenfold as progress is made in AI. The actual value created by these actions vary, as actions can be brilliant or foolish. If this direction is going to take place depends on whether the software community appreciates and understands what the introduction of AI can accomplish, or if they continue to stick to a business model that keeps them going until the money is spent.
Today, the most advanced analysis made in Requirements Engineering is rule-based, even if it's AI assisted. It will likely remain rule-based for the near future, despite that AI assisted analysis will increase in importance. As long as we address quality and information content, the problem is relatively well-defined, with a boundary that is defined by a grammar. Analysing problem domains in general is much harder, and AI assisted analysis will probably have a limited success in doing this, but the future may have other ideas for us. For Requirements Engineering, the progress in Natural Language Processing will be critical, and while the progress is good it doesn’t mean that AI can be easily applied to requirements analysis. AI is also a black box solution, in the sense that no one can really tell how it solves a problem. Failure in solving a task can also be a problem, and probably harder to fix than for traditional software. Because AI is trained by massive amounts of data, every AI application will be unique and can’t be copied. This means that the value of a company will depend more on how it succeeds in Artificial Intelligence and how creative it can be.