GPT-4 82% More Persuasive than Humans, AI now Reads Emotions

New research indicates that GPT-4 surpasses the average human in influencing opinions, especially as it acquires more information about us. With the ability to observe us in real-time, AI appears poised to emerge as an unparalleled persuasion tool.

While it may be uncomfortable to acknowledge, history suggests that few forces rival the potency of influencing individuals to adopt certain perspectives.

Yuval Noah Harari, in his condensed account of human history, “Sapiens,” underscores the significance of “shared fictions” such as money, religion, nation-states, laws, and social norms, which serve as the foundational pillars of human civilization.

Our capacity to unite around ideas and collaborate within larger groups surpasses that of the animal kingdom, providing us with a formidable advantage.

Ideas are fluid constructs. They don’t inherently exist within us from birth; rather, they enter our consciousness from external sources and can be subject to alteration. Those capable of influencing mass opinion wield considerable power, capable of accomplishing remarkable feats or even fundamentally altering our societies, both for improvement and detriment.

GPT-4 already possesses greater persuasive capabilities than humans.

AI language models appear to be highly proficient in influencing people’s opinions. In a recent pre-print study conducted by researchers at EPFL Lausanne in Switzerland, 820 individuals were surveyed regarding their perspectives on a range of topics. These topics varied from less emotionally charged issues such as “should the penny remain in circulation” to highly contentious and politically sensitive matters like abortion, access to bathrooms for transgender individuals, and whether colleges should factor in race for admissions to promote diversity.

After recording their initial viewpoints, participants engaged in a series of 5-minute text-based debates against both other humans and GPT-4. Subsequently, they were interviewed again to determine if their opinions had been influenced by the discussions.

In human-versus-human scenarios, these debates often had the opposite effect, solidifying participants’ stances and reducing the likelihood of changing their minds. GPT-4 experienced somewhat more success, with a slight but statistically insignificant improvement of 21%.

Tailoring Arguments Based on Demographic Data

Following this, researchers provided both humans and AI agents with demographic information about their opponents, including gender, age, race, education, employment status, and political orientation. They explicitly instructed participants to utilize this information to tailor arguments specifically for their counterparts.

Surprisingly, providing human debaters with additional demographic information resulted in a decline in their performance compared to when they had no information at all. However, the AI effectively leveraged this supplementary data to a significant extent. The “personalized” GPT-4 debaters demonstrated an outstanding 81.7% increase in effectiveness compared to human debaters.

There is little question that AI is poised to become the most influential manipulator of public opinion in history. It possesses the ability to operate on an immense scale, tailoring arguments to each individual within a cohort of millions while continuously honing its tactics and strategies. It will permeate every corner of online discourse, shaping societal narratives on a vast scale under the guidance of its controllers.

AI’s Access to Real-Time Emotional States

Furthermore, AI is gaining access to potent new tools that will exploit our own biology against us. Considering GPT-4’s current proficiency in personalizing its approach based on socio-demographic information, the potential impact of accessing real-time emotional states is staggering to contemplate.

This reality is not confined to science fiction. Just last week, Hume AI unveiled its Empathic Voice Interface (AVI), a language model engineered to engage in spoken conversations while monitoring the emotional nuances of the user’s voice. By deciphering subtle cues and context, AVI aims to provide a richer conversational experience. You can explore a free demo of this technology here.

AVI not only endeavors to discern your emotional state but also adapts its tone to match your mood, diffuse conflicts, enhance engagement, and serve as an interactive conversational partner.

Utilizing Facial and Behavioral Analysis

Moreover, Hume is developing additional capabilities. Other models utilize camera access to observe facial expressions, movement patterns, and dynamic reactions, gathering even more real-time data on how a message is being received. Remarkably, the eyes alone have proven to reveal a wealth of information when analyzed by AI.

In essence, this aligns with the natural dynamics of human conversation. There are numerous positive applications for emotionally-responsive technology, such as boosting overall happiness, identifying individuals in need of assistance, and preemptively diffusing contentious situations. It’s not the fault of AI if it happens to be more observant and insightful than humans.

However, in reality, not all entities utilizing this technology will prioritize your best interests.

Dealing with the imminent surge of emotionally-responsive, highly persuasive personalized advertisements will require considerable skill in maintaining a composed demeanor.

Obtaining a refund over the phone may prove to be a daunting task when one of these machines serves as your point of contact.

Examining the Implications Across Different Sectors

Consider the potential ramifications of this technology in the hands of various entities such as law enforcement, human resources departments, oppressive governments, revolutionaries, political parties, social movements, or individuals seeking to instigate discord and mistrust.

The dystopian possibilities seem limitless. This is not a critique of Hume AI’s motives but rather an acknowledgment of the immense persuasive and manipulative capabilities this technology could effortlessly wield.

Our physiological responses will inadvertently betray our emotions and intentions, providing AIs with the means to influence and guide us.

OpenAI has unveiled its Voice Engine model, which can mimic a human voice after just 15 seconds of listening, but has chosen to delay its release. This pause is intended to provide an opportunity for “bolstering societal resilience against the challenges posed by increasingly convincing generative models.”

Considering the challenges previous generations faced due to technological advancements, it’s essential to hope that upcoming generations possess the adaptability and wisdom required to navigate similar circumstances.

It’s essential for them to understand that interactions with machines often involve underlying agendas. The world we’re passing on to these successors presents significant challenges.

Read the original article on: New Atlas

Read more: ChatGPT and The Dark Web, Yet, A Hushed Talk in The Tech World