Augmented Intelligence
A Human-Machine Marriage on the Way to Complete Intelligent Automation
- A talk by Simon Stiebellehner, Head of AI @ craftworks
Nowadays, everybody is talking about Artificial Intelligence (AI). It has become a major buzzword of the yet young 21st century. But are we ready for a largely automated world driven by Artificial Intelligence?
In his talk at the AI congress in Vienna, our Head of AI, Simon Stiebellehner, gave us insights into the current state of AI and why we should often rather focus on “Augmented Intelligence”. Augmented Intelligence uses AI technologies with the objective of supporting humans in their decision making. In the following blog post, we summarize Simon’s talk for you.
The “i” in AI: Intelligence or Illusion?
If you have ever done a Google image search for “Artificial Intelligence”, you probably found lots of images of robots. The further you scroll down the results page, the creepier these images become — it quickly goes from human-machine collaboration to robots taking over the world and defeating mankind. That escalated quickly.
Obviously, what you find in such a Google search does not reflect the true current state of Artificial Intelligence. I can assure you that there is no Terminator yet. So what is the current state of AI? Considering the great advancements we have made when it comes to voice recognition systems, such as Alexa, or applications of computer vision such as autonomous driving, it is safe to say that the current state of AI is quite promising. We have come a long way in a very short period of time.
Yet, our trust in Artificial Intelligence is limited. A study conducted by Volvo shows this very well. They found that, ironically, one thing that people really want to have in an autonomously driving vehicle is a steering wheel. This highlights the fact that people are not willing to hand over total control to an AI system. Have you ever asked yourself how much you trust in AI? Do you also want that steering wheel? Let’s see.
Imagine you feel a bit sick. You have a cough and you feel slight chest pain. So you are going to see your doctor. However, instead of a human doctor, some AI-powered robot welcomes you and asks you for your symptoms and an X-ray of your chest. Based on that information the machine then computes your diagnosis: “95% probability of being healthy and 5% probability of Pneumonia. Congratulations, you are healthy! Most likely”. How do you feel about that?
“Do you trust the decision of the AI system or do you prefer an additional human opinion?”
Since this is about one of the most precious things in life, namely your health, you most likely prefer having a human doctor look over it in addition.
Naturally, trust in AI is not only limited when it comes to health. At craftworks, we have extensive experience in developing Artificial Intelligence solutions for the industry (Industrial AI). Frequently, we observe that also in the industry clients have the strong desire to have a human have the last say when it comes to more or less critical decisions. Let’s dive into such a case.
Suppose you are the manager of a production line in a factory that produces expensive industrial parts. From time to time, the machines on the production line need to undergo maintenance work. This means that your production needs to be stopped for a certain amount of time. On average, this costs you around € 10,000 per maintenance. Very rarely it happens that a machine has some kind of defect. Defects typically cause longer downtimes, significant repair costs and, in certain cases, even damaged products. Therefore, a machine defect is quite expensive. Typically, such a breakdown incurs costs of about € 200,000. Obviously, as the manager, you want to keep costs low. Thus, you purchase a predictive maintenance solution that enables you to predict the occurrence of defects before they actually happen based on historical machine sensor data such as temperature, pressure or vibrations. This allows you to perform maintenance in time to avoid costly machine defects. Similar as in the previous case, your predictive maintenance system may not be fully confident about some predictions. For instance, imagine the system outputs that “there is a probability of 10% that a machine defect will happen in the next 24 hours. 10% is below the defined threshold — no maintenance is needed”. However, failing to perform maintenance could lead to a defect, which may cost € 200,000. Are you willing to take that risk?
“Do you trust the decision of the AI system or do you prefer an additional human opinion?”
Probably, you prefer having one of your experienced technicians have a quick look at the output of the predictive maintenance system and the machine in question to verify that everything really is fine.
Summing up, these examples highlight that many critical tasks will not be completely automated by AI in the near future. And there are good reasons for that:
- AI solutions are limited by the (human) specification of the model, data quality and availability.
- AI often seems like a black box. Typically, humans tend to mistrust black boxes due to the lack of understanding and traceability of decisions.
Augmented Intelligence
Augmented Intelligence helps us overcome named causes of mistrust in AI. It aims to enhance human decision-making abilities through the use of Artificial Intelligence technologies instead of focusing on complete automation of processes. For this reason, it is a highly collaborative approach — a “human-machine marriage” you could say. What does this mean concretely?
Let’s have a look at our previous industrial example from an Augmented Intelligence point of view. Now, instead of just predicting the probability of a machine defect, the system goes one step further and additionally provides an explanation for its prediction. When the predictive maintenance system computes a high probability of a machine defect, a technician is notified. A mobile app on his smartphone provides him with the prediction and the corresponding explanation: “There is an 80% probability of a machine defect happening in sector C in the next 24 hours because there was a spike in vibrations between 10:10 AM and 10:15 AM today”.
In contrast to the AI system, the technician is human. Thus, he is not limited by model definitions or data quality. He is able to combine the information provided by the predictive maintenance system with years of experience and a holistic view of what is happening around him. This combination allows him to make truly qualified decisions. In this case, the technician knows that there was construction work going on next to the machine in question between 10:00 and 10:30 today morning. He reasons that this activity is the likely cause for the spike in vibration level and, consequently, there is no robust indication of a machine defect happening in the near future. This analysis took the technician 5 minutes but saved € 10,000 of maintenance cost.
“Augmented Intelligence is collaboration: a human-machine marriage where each part does what it can do best to exploit synergies and achieve superior results.”
Artificial Intelligence technologies, specifically Machine and Deep Learning, excel at detecting patterns in large datasets. However, at the core, AI is based on models — statistical models, machine learning models or deep learning models. Per definition, models only incorporate a small fraction of the real world, resulting in severe limitations. In contrast, humans are typically very limited in their ability to find patterns in high-dimensional data. But, on the other hand, we have flexible minds, we can think outside of the box and learn from very few samples. Augmented Intelligence is the collaboration of humans and machines where each part does what it can do best to exploit synergies and achieve superior results.
It is important to note that Augmented Intelligence is not a compromise. In fact, it is an indispensable step towards complete intelligent automation through Artificial Intelligence — for two main reasons. First, as we have seen, taking an Augmented Intelligence approach helps us overcome the limitations of current AI systems that we touched upon earlier. Second, and even more importantly, it builds trust in AI technologies. Augmented Intelligence does so by putting great emphasis on explaining the reasoning of AI systems and thereby fostering understanding and building trust in AI. Trust in Artificial Intelligence is fundamental for its large-scale adoption.
In case you are interested in more details, technical insights about Augmented Intelligence (starting at 16:55) or a practical industry case (starting at 20:00), you can watch Simon’s talk here.
