We are measuring the AI ​​badly, and we miss what matters most

There is a peculiar irony in how we evaluate artificial intelligence: we have created systems to imitate and improve human capacities, but we measure their success through metrics that capture everything except What makes them truly valuable for humans.

Technology industry boards are overwhelmed by impressive numbers in AI: processing speeds, parameters, reference scores, user growth rates. Silicon Valley’s largest algorithms adjust non -stop algorithms to overcome these metrics. But in this maze of measures, we have lost sight of a fundamental truth: the most sophisticated in the world is not worth it if it does not significantly improve human life.

Consider the story of early search engines. Before GoogleCompanies competed fiercely for the large number of indexed web pages. However, Google did not prevail not because it had the greatest database, but because it understood something deeper of human behavior, that relevance and reliability matter more than a gross amount.

Alas that generates confidence

The current alas landscape feels very similar, and companies that compete to build larger models, although the most nuanced elements of humans’ design that really drive and impact can be missed.

The path to a better evaluation of the AI ​​begins with confidence. Emerging search prove That users are deeper and persistent with AI systems that clearly explain their reasoning, even when these systems sometimes climb. This has an intuitive sense: trust, whether in technology or humans, grows transparency and reliability instead of pure performance metrics.

However, trust is just the basis. The most effective AI systems were genuine emotional connections with users, demonstrating a true understanding of human psychology. Research reveals a convincing pattern: When AI systems are adapted to the psychological needs of users instead of simply performing tasks, they become comprehensive parts of people’s daily life. This is not a superficial kindness: it is about creating systems that really understand and respond to human experience.

Trust matters more than technical capacity when it comes to AI adoption. An innovative Chat study AI Of about 1,100 consumers found that people are willing to forgive the failures of the service and to maintain the loyalty of the brand not based on the speed with which a AI resolves their problem, but if they trust the system that tries to help them.

Alas that gets you

The researchers discovered three key elements that create this confidence: First, the AI ​​must show a real ability to understand and address the problem. Second, he must show benevolence: a sincere desire to help. Third, it must maintain integrity through consistent and honest interactions. When AI’s Chatbots embodied those qualities, customers were much more likely to forgive service problems and less likely to complain -others for their experience.

How is a trustworthy AI system? The study found that simple things make a big difference: the anthropomorphization of the AI, programming it to express empathy through its answers (“I understand the frustrating that must be”) and be transparent about data privacy. In an example of an indicator, a customer who treated delayed delivery was more likely to remain faithful when a chat named Russell acknowledged his frustration and clearly explained both the problem and the solution, compared to an unnamed boat that had just declared facts.

This perspective challenges the common assumption that AI only has to be quick and precise. In healthcare, financial services and customer service, the most successful generative AI systems are not necessarily the most sophisticated: they are the ones that create an authentic relationship with the users. Take time to explain their reasoning, recognize concerns and demonstrate constant value for the user’s needs.

And yet, traditional metrics do not always capture these crucial dimensions of performance. We need frames that evaluate the systems of AI not only in their technical competence, but also in their ability to create psychological security, create an authentic relationship and, most importantly, help users achieve their goals.

New AI metrics

In Cleo, where we focus on improving financial health through an AI wizard, we are exploring these new measures. This can mean measuring factors such as users ‘confidence and the depth and quality of users’ commitment, as well as looking at whole conversational journeys. It is important for us to understand if Cleo, our AI financial assistant, can help a user with what they try to achieve with any specific interaction.

A more nuanced evaluation framework does not mean abandon performance metrics: they are still vital indicators of commercial and technical success. But they must be balanced with deeper measures of human impact. This is not always easy. One of the challenges with these metrics is their subjectivity. This means that reasonable humans may disagree with what looks like. Still, it is worth chasing.

As the AI ​​sticks more deeply into the fabric of daily life, companies that understand this change will be those that succeed. The metrics that have brought us here will not be enough where we are going. It is time to start measuring what really matters: not only the operation of the AI, but the good that helps humans to prosper.

Opinions expressed on Fortune.com Comments pieces are only the opinions of their authors and do not necessarily reflect the opinions and beliefs of Fortune.

Read -Ne More:

• CEO of Genesys: As the Ematic Ia can climb our humanity During economic uncertainty
• When ai is built ai: The upcoming big inventors could not be human
• ItAI cost collapse changes what is possible– “me
• I have spent years helping female founders to access capital.Now that they have and, they may not have to do it

This story originally presented to Fortune.com

Source link