AI mistakes are too risky: Here is how to fix them

Artificial intelligence is based on complex data analysis. One acquired skill of AI System is that it adapts to the given information. There are processes that analyse that data, predict and make decision based on it. There will be sensors that collect the data and actuators that make the changes in the real world based on the prediction or recommendation of the system. For example, in an auto heating system, the sensor collects the data on temperatures, AI system makes some complex data analysis and predicts or recommends a decision and the actuator actually carries out the decision recommended in real life like turning on the heating on or off or adjusts.

AI & Autonomy

Inherently all AI systems imply some amount of autonomy. That means it works without any human input.

Issues of designing for AI & Ways to address

• Understand how AI works in a particular case in question and then design for it.

• Direct manipulation and control are possible in a traditional system. But that may not be the case with AI systems. There is a direct mapping between an action and a result in the traditional system. Unless designed considering this direct manipulation and control principle, this may not be the scenario with AI systems because of its inherent autonomy and autonomous learning capabilities. Less control over the outcomes will lead to less trust of the system by the user.

• While designing AI systems, the designers should also need to take into consideration the Transparency and social norms which are currently not built into the systems.

• Even in a most intelligent and learned artificial intelligent system, there is probably a 3% error and some have even more errors. The problem with AI systems is that we cannot predict in which way the system is making an error because it’s based on complex analysis of data. When a real person makes an error, at least he is able to explain the rationale behind making that error and take you through the reasoning of how it went wrong. A machine can’t do that on its own.

The more verified data we feed, the more accurate the analysis becomes. The AI allows the user to train it in some a way that simplifies a lot of labour from the other side, which is recommended by the users of digital world.

“You will rarely get an Artificial intelliget system that gets it 100% right all the time”

Dr. Simone Stumpf, Senior Lecturer at City University London

Now enjoy the video

Featured Speaker

Dr Simone Stumpf, Senior Lecturer at City University London

Well-versed in academic and industrial Human-Computer Interaction (HCI) research and interaction design practice. Research Interests include:
– end-user interaction with intelligent/machine learning systems
– end-user development for everyday systems
– personal information management esp. activity-based interactions
– information-seeking particularly for images or music