Giving systems emotional intelligence
Artificial intelligence (AI) is finding a home in many applications, from industrial automation to autonomous vehicles. Perhaps its most personal impact, however, is when the AI must interact with humans in providing information and services. For human interaction, the next trend for AI to embrace is the addition of emotional intelligence.
When the Amazon Echo first came out in 2014, I viewed it as a perfect example of an IoT device. It required a minimum of onboard hardware capability, achieving most of its impressive capability in understanding and responding to human speech by using cloud-connected resources. This allowed it to be inexpensive yet powerful and allowed continual capability upgrades to occur without requiring changes to onboard hardware or software.
Now, it is the face (or perhaps, voice) of AI for many consumers. Its popularity has allowed the general public to become comfortable communicating with machines by voice rather than a keyboard. But while today’s AI does an outstanding job of word recognition and analysis in its vocal communications, its dependence on words alone is an inherent limitation.
According to Dr. Rana el-Kaliouby, CEO and co-founder of AI company Affectiva, only about 10% of human communication depends on words. As she pointed out in her presentation at the Global Altair Technology Conference (ATC 2020), some 90% of human communications involve vocal intonation, body language, and facial expressions. AI using speech recognition alone misses out on all of that.
This is a limitation for more than the Echo. Speech recognition AI is now widely used in a variety of business settings, such as telephone support systems. Many businesses long ago replaced human operators as the first level of telephone support with automated systems that take callers through a menu of options. At first those systems had callers use touch-tone key presses to navigate through a fixed, and sometimes lengthy, decision tree. Increasingly, though, speech-recognition AI is giving callers the ability to state their concerns verbally to more quickly navigate through the system, allowing for a much more complex set of response options to become available.
But it all feels so cold and mechanical, dampening a company’s customer service reputation and often frustrating the caller. A more human form of communication is desirable, which AI cannot provide when it depends on words alone. That is set to change, however.
According to Dr. Ayanna Howard, founder and CTO at Zyrobotics, providing AI systems with the ability to infer their user’s emotional state and respond accordingly is one of the technology’s emerging trends. Speaking at ATC 2020, Dr. Howard pointed out that an “emotional” AI that can sense and respond to the user’s emotions holds great promise for increasing the user’s performance in human-AI collaborative efforts. An early study by researchers at Stanford University and Toyota, for instance, determined that something as simple as adjusting a car voice system, such as for a navigational aide, to react to emotions could improve driver safety. The study showed that matching the car’s voice – energetic versus subdued – to the driver’s emotion – happy versus upset – resulted in drivers being more attentive to the road and having fewer accidents.
Dr. el-Kaliouby also sees promise in bringing emotional intelligence to AI. She pointed out that the best way to build trust starts with empathy, so users are more likely to trust an AI system that is capable of sensing and reacting to their emotions. This can lead to more effective interaction, as well as helping the system recognize potential problems. Rising user frustration, for instance, might signal to the AI that there is something wrong in the process that needs correction.
The creation of emotionally-aware AI is already underway with products now reaching the market. Affectiva, for instance, offers automotive AI systems with in-cabin sensing that monitors, among other things, the emotional state of the car’s occupants so that it can improve rider comfort by adapting music, lighting, temperature, and the like. It can also help improve driver safety by recognizing such states as drowsiness, anger, and distraction. A similar offering from Sensum provides an empathic AI engine for automotive developers to give them a head start on developing systems that respond to how users feel.
Figure 1 AI systems are being developed that can infer human emotional states and respond appropriately, a kind of artificial empathy. Source: Affectiva
For voice-only systems, such as the Echo, the Japanese company Empath offers an API that lets developers add emotional detection capability. Their cloud-based software service accepts a WAVE file and returns an assessment of the speaker’s mood: joyful, calm, angry, or sorrowful. Current applications using the capability include management tools that check employee’s moods from their speech to help improve motivation, and smart call centers that can visualize customer and caller emotions to help improve the conversion rate in outbound telemarketing.
There is still much to be done for AI systems to acquire more than basic emotional intelligence and empathy. Still, even a basic ability to sense and respond to user emotion will go a long way to making the technology more comfortable to use. Emotional AI is a trend worth noticing.
Rich Quinnell is a retired engineer and writer, and former Editor-in-Chief at EDN.
This article was originally published on EDN.