AI Review: Trust, Predictions, and Anomalies

In this edition, we tackle the age-old idiom ‘you don’t know, what you don’t know’. Might it be by anomaly detection, automated testing, or prediction, artificial intelligence is steadily moving forward, integrating into everyone's lives, and changing the way businesses learn.

AI UNDERSTANDING GOES BOTH WAYS

Technology and humans have always worked hand in hand to achieve all sorts of goals. Recently, with the advancement of AI and other forms of automation, we need to start thinking about ways to foster mutual understanding.  

The best way to achieve that is by creating environments that promote the behavior we are expecting to see, like a game, simulation, or test. This is what the United States Ministry of Defense has done with its pilots through some friendly human-robot competition. In the interest of improving human-machine teaming and building trust between the two, the Ministry has created an Air Combat Evolution program that will put on its first test between an F-16 fighter pilot and a machine-learning algorithm. The program manager hopes this will be a step forward in the road towards effective human-machine teaming in combat situations, a relationship that relies on huge amounts of trust.

Building trust is -for many- the ultimate frontier. While increasing the effectiveness of machines helps, having confidence ingrained in the way we work with technology will help solve harder problems faster and more accurately in a number of situations. To help better understand the relationship between trust and performance, a team from UCLA ran an experiment to see what method of explanation fostered the most trust between humans and robots. In this, they found that what was most effective for the productivity of the robot did not directly correlate to more trust from the human participants. Rather, the researchers needed to use a combination of performance-based components (in this case, haptics) and more symbolic components to see the highest levels of trust.

What does all this mean? We are still a ways from humans fostering complete trust in automated systems. Those designing and implementing AI into their businesses should spend time thinking about how they learn what AI is doing to ensure the most impressive results. 

PREDICTING THE FUTURE, UNDERSTANDING THE PRESENT

The phrase "It's tough to make predictions, especially about the future" is often attributed to Yogi Berra. Spoiler alert: it isn't his. Regardless of who said it first, it is particularly relevant whenever trying to understand what changes AI will bring in the future.

Yet, in this Wall Street Journal article, six experts opted to weigh in on the most significant challenges and opportunities artificial intelligence will bring. For some, the future looks grim without some serious human intervention. For example, questions of ethics easily arise when we look at the way human bias can infiltrate our algorithms. For others, there is the inevitability of failure in areas like the healthcare industry and stock trading. Still, many predict that within 15 to 25 years, we will have a sophisticated AI capable of much more than what we can even dream of today. 

With all of these challenges and opportunities, it can be difficult to discern where AI should step next in the short-term. Should we be focusing on human bias? Should we be creating more and more sophisticated systems? Ultimately, we should always look at these challenges through a lens that keeps the final customer, patient, or user in mind. This will aid us in crafting solutions that will evolve with our clients and help them cut through the noise to make the best call for their business.

ANOMALY DETECTION: A POWERFUL TOOL FOR YOUR COMPANY

Businesses across the board are continually seeing disruption and changes in their customers' behaviors, needs, and expectations. 

Sometimes there are clear trends that build up gradually and sometimes these disruptions happen suddenly. Those disruptions sometimes leave fragments of their behavior in ways that help identify them before they become too disruptive. This tactic is often deployed by companies that aim to prevent fraud, avoid unnecessary losses, and make the most out of emerging trends before they become reality. That's why anomaly detection is such a powerful tool.

In Chalapathy and Chawlas’ paper, "Deep Learning For Anomaly Detection: A Survey", they present the advantages and limitations and discuss the computational complexity of the anomaly techniques in real application domains. It identifies and explains the different aspects of deep learning-based anomaly detection. The challenge associated with detecting fraud in telecommunications, insurance, health, automobile, banking etc. is that it requires real-time detection and prevention.

At IV.AI, we use robust models to understand and predict those disruptions from language, because it is the frontline between companies and their customer support teams. But there are different methods. For instance, Ben-Gurion University researchers are using similar approaches with imaging technology to prevent malware in the healthcare industry. 

In a different study from Ben-Gurion University, a research team led by Tom Mahler, has developed a technique using artificial intelligence that analyzes the instructions sent from the PC to the physical components using a new architecture for the detection of anomalous instructions. They evaluated the new architecture in the computed tomography (CT) domain, using 8,277 recorded CT instructions and evaluated the CF layer using 14 different unsupervised anomaly detection algorithms.

Whenever working with big companies you always find large datasets that allow to model anomalies and prevent harmful scenarios. The only thing you need to figure out is how you want to make the most out of it. What if you spend, let's say, one hour every week in order to clarify, focus, and act on your data chest? That's how treasures are found.

If you don't want to miss the next AI Review please sign up here.