Human Factors in Artificial Intelligence

• How would you define Human Factors?

Human Factors deals with all human interactions with the environment. As such it covers human interactions with technology and how human abilities and limits have to be taken into account in the design of these technologies.

• What are the applications of HF in Thales's avionics activities?

Thales deals mainly with technologies both in civil and military applications, and for more than 30 years has been a leader in the application of HF for design.

HF is everywhere, in all forms of technology designed by Thales because these technologies will be used by human beings and, as such, they need to take HF into account. Novel emerging technologies such as AI, machine learning and deep learning need HF in the same way because users will need to understand and know how to deal with them to attain efficient performance.

People need to have confidence in the system and they need to be able to anticipate the outcomes of their interactions with the system, whether they be from AI or not. Users need to grasp enough of the system so they’re able to make it work in a high-performance manner.

• What are the benefits of the HF in the systems developed by Thales?

The benefit in HF is improving technology to a level where human-machine interactions become efficient... and ultimately even pleasant!

“The best technology ever, if not usable by human beings, is the worst of technology”

The benefits of having HF in the design process is to be able to develop technologies that range from “at least usable” to “optimally performing” as users interact with them. And the more HF you put in the design of the system, the more you migrate from barely usable to optimal performance of the human-machine system

• Are there any specifics in HF for AI-based systems?

Today’s AI is the first trial at imitating human intelligence.

As such, it poses the new problem of interacting with a new form of intelligence we know nothing about, like an alien! An alien that would have abilities but that are different from humans. It means we need to create new models of interaction – that we don’t have yet – since all the models of interactions we have are based on human-related models, or animal-related models. We don’t yet know what type of AI models will be needed to interact with such systems. We need to develop those new interaction models: AI models for humans to interact with AI, and human models for AI to interact with humans. That is the core of Human Factors.            

• What are the limits of Human Factors?

HF represents a cost in the design process, and as a cost, it needs to be justified. The problem with HF is that it can justify its cost only if it was not applied.  When HF is applied properly, the result is better but it’s hard to get everyone to agree it’s better because of HF.

In fact, experience gained from failure is often the way of showing what the HF benefits would have been.

• What is the future of HF?

HF is everywhere, it’s undeniable. Every system that we design has so far been human-made for humans. But, with the advent of AI, it’s becoming interesting to try and understand whether AI needs specific “AI factors” to be designed, as complementary to HF.

But if you think that AI is ever to be at the service of human beings then HF will apply as is. It’s just another system designed for humans.

• What about explainability?

Imagine a human being is encountering an alien for the first time. Then, humans will have some sort of trials to communicate with that alien to make sense of any kind of its communication. That is explainability:  trying to make sense of what the alien is saying so we can interact with it in a proper manner.

Don’t you think that somehow when your smartphone isn’t working properly, it’s acting like an alien? You can’t make sense of it and it doesn’t react to what you want. That’s when explainability should kick in. Even more if you design a system that has similar or some intelligent potential, then you need to have an efficient way to understand it and to interact with it. That’s the basis of explainability.

• Are Human Factors a way to keep control on AI?

HF is way to put some human measure in the development of AI: I’d say seeing AI as an assistant system to human beings is a safe way to design AI.