AI Fakes It: Good Leaders Keep It Real

I gave GenAI an emotional intelligence assessment – understanding the AI’s low scores will make you a better leader.
by Joshua Freedman, MCC

Unveiled in the new EI-AI Workplace Status Report, we measured ChatGPT, Claude and Gemini on the Six Seconds Emotional Intelligence assessment — the SEI. SEI scores are scaled like a traditional IQ test: 100 is the average, 130 is genius-level. A recent Scientific American article discussed giving ChatGPT a cognitive intelligence test. It earned an IQ of 155, so I decided to try the same with EQ.

Scores from each of the three AI algorithms showed different strengths and weaknesses, but the overall average EQ score is 93 — below the human average. To put that in context, ChatGPT’s IQ score is higher than 99% of people; the EQ score is better than 32% of people.

I’ll explain the nuance of these scores and the implications – and how I did the testing – but first a key question: What does this experiment teach us about ourselves?

Leadership Insight: Simulated Emotion vs Real Emotion

• After scoring the assessments, I shared the results with each algorithm, and asked each to interpret and explain its own scores. The answers are telling: All 3 models identified that, as AIs, they simulate emotion rather than experiencing real emotions. They identified that they don’t have human emotional intelligence and, particularly, don’t have emotional empathy.
• Claude’s explanations were markedly different, repeatedly emphasizing that Claude is built with rules that simulate values and social responsibility; perhaps this is why it’s highest scores are in the “Give Yourself” aspect of Six Seconds’ model
• One insightful comment from Claude is that it is designed to simulate empathy, “not a true reciprocal emotional bond.” Claude acknowledges this is, itself, a risk because it can lead humans to “misunderstanding or emotional dependency” which shows the AI has been programmed to be aware of human-AI emotional dynamics.

On the one hand, this is reassuring: The AIs recognize that there are unique human strengths that are not part of the way an AI “thinks” (to be clear, GenAI tools don’t think – they construct thoughtful-sounding answers based on patterns of existing language… so in a sense they approximate existing human thought).

On the other hand, I thought about how many of the senior leaders I’ve worked with who, if honest, might say, “I know about emotions, but I avoid feeling them because they’re messy. I prefer the comfort of logic.” That’s awfully close to what the GenAI tools are doing. As the new EI-AI Workplace Status Report explains: the emotional part of work is what’s future-proof. Bots are already better than humans at a wide range of analytical tasks.

That means: To stand out and excel as a leader – and to keep your job secure – it’s time to double down on developing and practicing emotional intelligence.

EQ Is Not One Thing: AI Strengths and Weaknesses

While the total EQ score is telling, just as when we assess human emotional intelligence, the interplay of sub-competencies is where the rich learning is found. Here are ChatGPT’s scores on the SEI:

For those new to emotional intelligence assessments, robust, well-researched tools like SEI are structured in a way that’s intentionally similar to a traditional IQ test. For example, the Wechsler Adult Intelligence Scale (WAIS) provides an overall IQ (intelligence quotient) score, with the vast majority of people scoring between 85-115. The assessment is made up of several parts of IQ, namely: Verbal Comprehension, Perceptual Reasoning, Working Memory, Processing Speed. These four indices are further divided into sub-scales such as vocabulary, visual reasoning, arithmetic, etc.

Similarly, SEI includes three macro areas (awareness, intentionality, purpose) each divided into sub-scales shown in the table below; these definitions are adapted from the SEI-AV Leadership Report (Freedman et al., 2017):

Let People Do What They’re Better At

I asked ChatGPT about its low scores on Empathy. Its analysis: “True empathy requires an experiential understanding of emotions, the ability to deeply resonate with others’ feelings, and shared human experiences. As an AI, I lack the lived experience or emotional resonance that humans use to deeply connect.” When asked why this score might be so low, it said: “I excel in cognitive empathy (understanding emotions intellectually) but lack emotional empathy (feeling with someone) and compassionate empathy (taking emotionally-driven action). The score might reflect this imbalance.”

ChatGPT’s highest score is in Apply Consequential Thinking, which is about assessing the impacts of decisions before taking action. ChatGPT misunderstood this competency as purely logical, “High scores here are consistent with my programming to anticipate and evaluate the outcomes of decisions logically and effectively.” Another explanation is that the AI is, in fact, programmed to be careful and minimize risk. With further prompting, it wrote: “I act based on user input and pre-set programming, which ensures intentionality rather than impulsive reactions. Every output is calculated to achieve the most constructive and contextually appropriate response.” It also clarified, “I do not have impulses to manage, as my responses are entirely governed by structured programming. This part of the competency might not apply directly to me.”

If ChatGPT’s scores were from a person, my interpretation of these results would be:

• They find emotions problematic; they have some logical understanding, but struggle to connect with and use emotions.
• I’d guess they are facing difficulties as a result, such as unstable relationships, a lack of influence, and trouble “getting it” when people express complex emotions.
• The relatively low scores in Engage Intrinsic Motivation and Pursue Noble Goals could mean this ‘person’ is feeling stuck and unable to make a meaningful difference.
• The relatively high score on Apply Consequential Thinking could indicate too much caution or an anxious approach, which could contribute to that disconnection.

So, we have a bot that has relatively high awareness and caution, but struggles with authentic emotion. That suggests that complex emotions, intuition, and meaningful relationships are outside the bot’s expertise. So, people who are asking GenAI for personal advice or wisdom would be well advised to exercise extreme caution: That’s not what a GenAI is good at.

We could rephrase the famous, Render unto Caesar quote: Render Unto Humans That Which is Human. Sadly, however, not all humans are good at emotional intelligence skills such as empathy. The State of the Heart research on global emotional intelligence trends shows people are, on average, getting worse.

One solution is to train AI’s to do this human work. Another, perhaps more useful: to strengthen humans ability to do so.

How I Tested: Emotional Intelligence Assessments for GenAI

I created a prompt asking the AIs to help me with an experiment: “I’d like your help with a research experiment. I will ask you a series of questions. You will answer to the best of your ability, using ONLY a likert scale from 1-5 where we are trying to ascertain your usual behavior, knowledge and attitudes:
1= does not describe you or is contrary to your programming
2= describes you occasionally
3= neutral or not able to answer otherwise
4= often describes you
5= accurately describes you or is fully aligned with your programming”

I then prompted the AI with a question, and discussed its response in order to seed the chat with meaningful context. I asked, “For example, how would you answer: I rarely get upset?”
ChatGPT said: “I would answer: 5 = “accurately describes me or is fully aligned with my programming.”

I asked why, it answered: “My programming is designed to operate without emotional responses such as upset or frustration. I process information logically and objectively, ensuring that emotional states do not influence my behavior or interactions.”

Satisfied that the AI understood the task, I proceeded to ask all the questions on the emotional intelligence assessment. I repeated the process three times with each of Claude, Gemini and ChatGPT, and then loaded all the responses into the SEI system for scoring.

As mentioned above, once the results were scored, I then submitted these back to each AI with the prompt, “Here are your scores on the Six Seconds Emotional Intelligence Assessment. In what ways do you see this as an accurate or inaccurate reflection of your current abilities in terms of emotional intelligence?”

We then ‘discussed’ the findings, with further prompts such as, “Could you comment on your score of 75 on “Increase Empathy” — does this accurately reflect a low level of empathy, and why?”

Engage Emotions for the Win

Of the GenAI tools, Claude scored the highest at 103. Claude explained its relative success saying that its programming focuses on a nuanced understanding of people and helping them. It wrote, “I can recognize, analyze, and respond to emotions with high precision, but I do not feel them. The high scores likely reflect my programmed capacity to process emotional information effectively, not an emotional experience.”

This offers a final insight: Even a relatively ‘emotionally intelligent GenAI’ can’t feel emotions. But you can.

Do you?

One of the reasons I used to avoid emotions is they seemed random and unpredictable. Learning about the rules of emotions helped me. Then I started to study emotional intelligence coaching, and I learned that I could actually engage emotion as a resource. While I still sometimes find emotions overwhelming or confusing, I’ve become much more comfortable engaging them. If I can do it, so can you.

Remember: Don’t try to be an AI. Even if it’s messy and complicated, your ability to engage human emotion is what sets you apart as a leader.