Bringing emotional intelligence to artificial intelligence

It doesn’t take much for a machine to beat me at backgammon. Level 2 will do it. If you ask Alison, it doesn’t take much for a GPS to find a better route to the downtown gallery rather than my more scenic navigation.

How do I feel about being outperformed by artificial intelligence in almost every field where it can be applied? Personally, I am not anxious about it: in fact, I think it’s rather fun. Others, less geeky about the tech, may feel uneasy or even threatened. But most importantly, we all feel something because we are emotional beings.

For their part, what do the AI apps feel about being better than us? Absolutely nothing.

This lack of emotional content is not a problem in itself. In fact, it may be one of the reasons AI is so  effective in some scenarios. Current AI achievements are purely transactional and most efficiently so. Yet, as humans, a great deal of what we do in business, education, entertainment and our personal lives is driven by emotion, not by analysis.

The IQ of AI may be impressive, but its EQ - its emotional intelligence - is entirely lacking. In one sense this may always be the case. Adam Smith long ago pointed out that we feel sympathy for others in distress because we can imagine how we ourselves would suffer under the same circumstances. Machines may mimic our behaviour, but they cannot share our experiences.

Teaching machines about emotions

As AI takes on roles that are increasingly integral to our day-to-day activities, we will come face-to-face with this lack of emotional awareness as a significant limitation. Without EQ, AI will be severely restricted in its ability to interpret complex human contexts, resonate with users, and build trust. As a result, AI's usefulness in sectors like healthcare, customer service, and social interactions could always be hindered.

We need to find a way to teach EQ to machines which can only work with objective data. But emotions are highly subjective, nuanced, and dependent on context. This poses immense challenges. Unlike factual information, emotional states are often ambiguous and layered and conveyed not only in words but by tone, body language and other subtle factors. For example, a raised voice could signal anger, excitement or determination: or the speaker may just be from New York.

Current AI training data has severe limitations in representing rich emotional diversity. Datasets skew towards majority demographics, losing important cultural nuances. Complex emotions are difficult to categorize and quantify. Even extensive data may not sufficiently capture the full human emotional spectrum.

Finally, EQ demands going beyond surface-level pattern recognition. It requires truly understanding emotional triggers and flows to respond appropriately. Yet machine learning models have limited capacity for reasoning about abstract concepts like empathy. They cannot intrinsically comprehend human needs for compassion.

Untapped - a unique approach

Recently I have been working with a team who have a compelling answer to this problem. Untapped.ai is a professional coaching company which combines highly skilled human coaches with AI approaches. On the human side they combine professional coaching with psychological and therapeutic techniques to add speed, depth and sustainability. (They call this acceleration. ) The AI meanwhile acts as a third eye in the relationship between the user and the human coach, offering a different perspective, which is uncensored and unbiased. The AI, subject to continuous supervised learning, generates scores on seven crucial core competencies.

The result is exceptionally effective (accelerated) coaching. But there’s more to it than that. Untapped have also curated an unparalleled coaching dataset capturing thousands of emotional interactions over many years. The dataset encodes real-world psychological data, rich in emotionally nuanced interactions.

The dataset is notable for its quality, scale and nuance, in addition to the strict governance you would hope for. It’s also unique in its diversity, capturing a wealth of otherwise underrepresented voices in AI training data, ensuring the information reflects a breadth of cultural backgrounds, communication styles, and emotional patterns.

It’s really fascinating to work with this team: the potential of the data asset is extraordinary: it feels like a precious asset. On every call we’re almost in awe of what is at hand. For anyone else to build a similar resource from the ground up, could involve years of experimentation and refinement: the slow pace is critical to allow the accumulation of authentic human emotions in deep context. It may never be possible to create such a dataset again.

The promise of AI integrating meaningfully into our lives hinges on its ability to understand not just algorithms but emotions. The gaps are evident — from limitations in data representation to the inherent challenges of teaching EQ to machines. However, innovative approaches like Untapped's exceptional coaching dataset offer glimpses into a future where AI could be as emotionally aware as it is intelligent. This isn't just a technological aspiration; it's a societal imperative, essential for AI to genuinely assist, enrich, and empower our human experiences.

Check out Untapped at www.untapped.ai or contact me for more information.