Systematically using artificial intelligence
Artificial intelligence (AI) is often equated with machine learning (ML). This way of thinking already limits us unnecessarily, because machine learning is very effective in many situations, but does not exploit the potential of AI by far. To really benefit from AI, we need a holistic approach.
MIT professor David Martinez is of the opinion that a systems engineering perspective is necessary in order to fully exploit the potential.
Using artificial intelligence today and tomorrow
Companies set AI primarily to gain automated insights from content: "Is there a cat in this video?". This area of application is powerful, well understood and already supports decisions and increases efficiency in many areas.
However, artificial intelligence can also be used collaboratively by teams consisting of humans and AIs working together. The use of AI should lead to context-based work and support decision-making processes that are very similar to human collaboration.
That sounds promising, but how do we get there? We need to look at the use of AI from a systems engineering perspective.
Successful use of AI with systems engineering
In A Systems Engineering Approach to AI David Martinez describes what this could look like. A systems thinking approach is central to this. This is because AI can only be used effectively in a context that consists of sensors, data sources, processes and interactions between people and machines. Specifically:
- Sensors and other data sources - We are collecting more data than ever before. Much of this data is unstructured. This presents a huge opportunity for AI - but also the danger of being buried under the mountain of data if we lose sight of our goal.
- Data preparation - The previous point makes it clear that data must be processed before other AI systems can process it further. In principle, this is a "simple" data transformation, but it requires us to have an understanding of what is signal and what is noise in the data stream.
- Machine Learning - ML is just one element in the system chain, albeit an important one. We need to find the right approaches here. Do we need labeled data, or can a self-learning ML system be used? This question can only be answered correctly in context.
- Man-machine teams - The previous point shows that the data produced by the ML system must be further interpreted and processed by humans. Here we often have feedback loops to further improve the ML system.
- Application - The aim of the AI system is to provide a benefit. For example, the analysis of an X-ray image could prompt the doctor to further adjust the therapy. If the result is not correct, then everything else was useless and must be reconsidered accordingly.
- Computer systems - The use of AI has only become possible thanks to modern computer systems and is therefore an important aspect of the overall system: Are calculations carried out centrally or on end devices? Which data volumes need to be moved where? Are there architectures such as graphics or Tensor processorswhich would be particularly efficient?
- Robust AI - In particular, this means that we have at least a rudimentary understanding of the overall system, can comprehend decisions and assess the tolerances of the results. For example, there have already been several scandals concerning discriminatory facial recognition.
Conclusion
This block is usually about systems engineering knowledge. Today it was about a domain in which we Systems Engineers can make a contribution. Artificial intelligence opens up enormous opportunities to drastically change our lives. We have the opportunity to support this with our work.
