How Virtual Friends Can Influence People
Can Dale Carnegie's century-old self-help algorithms teach us anything about AI's growing role in our social lives?
It took Dale Carnegie less than three years of hard work to turn his languishing Nebraska sales territory for Armour & Company meatpackers into the national sales leader. In 1911, at the top of his game, he quit to pursue his dream of becoming a lecturer. By the age of 24, he was teaching public speaking courses at the YMCA in New York City, drawing on his stellar sales career to invent materials and methods for his lessons. His classes overflowed, he spoke in ever-bigger venues and, as his fame spread, he travelled the world lecturing to business leaders and well-heeled customers.
After some arm-twisting from publishers, Carnegie distilled his most important lessons into book form. How to Win Friends and Influence People sold 250,000 copies in the first three months after its 1936 release, and revised editions still sell around that number every year. The book is seldom missing from top-10 lists that measure sales, library check-outs or social influence. Dale Carnegie proved himself one of history’s most astute observers of human nature.
A set of AI technologies I call “virtual friends” have been hacking the deep, evolved pathways by which humans make friends, grow close to intimates and influence one another.
Carnegie doesn’t appear in my 2021 book, Artificial Intimacy: Virtual Friends, Digital Lovers, and Algorithmic Matchmakers, but I have thought about him often since it came out. This might be surprising: can a personal development coach born in 1888 teach us anything relevant about the twenty-first-century challenges thrown up by artificial intelligence?
Quite a lot, actually. In recent years, a set of AI technologies I call “virtual friends” have been hacking the deep, evolved pathways by which humans make friends, grow close to intimates and influence one another. And Dale Carnegie showed the way.
How to Win Friends and Influence People ran far ahead of its time by combining two tools that in retrospect have a very twenty-first-century feel. The book’s section titles resemble BuzzFeed listicles: “Fundamental techniques in handling people,” “Twelve ways to win people to your way of thinking” and “Seven rules for making your home life happier.” The genius of his technique is that he turns complex social challenges into algorithms.
Although people normally associate the word algorithm with computer code, it refers to any set of step-by-step instructions for performing a task: recipes for cooking a cheese soufflé or mixing a dirty martini are algorithms. Dale Carnegie had a knack for identifying particular interpersonal challenges and breaking their solutions down into algorithmic steps.
The Likability Algorithm
Friendship, intimacy and even romantic love arise through a series of interactive steps that are so mundane that we often don’t realise that they are the parts of an algorithm. From the moment we first make a person’s acquaintance, we tend to greet, talk, listen and pay attention to one another. We exchange titbits of gossip. And, little by little, we disclose aspects of ourselves to one another. The more often two people interact, and the more of these steps they take, the closer they tend to become.
With every such step, we move closer to establishing trust, a cooperative relationship and perhaps even a social alliance with another person. This is the foundation of humanity’s remarkable ability to share knowledge and coordinate work, which permits difficult human enterprises like farming, child-rearing and defence against outside attacks. Over the last 100,000 years or so, humanity’s knack for cooperation and coordination enabled our ancestors to spread into almost every habitable part of the world. The simple steps of relationship building combine into social algorithms that have world-conquering potency.
Within any group of humans, people range in social competence from inept to sublime. So many of us have room for improvement that Dale Carnegie and a century of imitators have been able to make handsome livings by helping us optimise our social algorithms. Consider, for example, one of Carnegie’s most potent listicle-style algorithms, “Six ways to make people like you.”
Become genuinely interested in other people.
Smile.
Remember that a person’s name is, to that person, the sweetest and most important sound in any language.
Be a good listener. Encourage others to talk about themselves.
Talk in terms of the other person’s interest.
Make the other person feel important—and do it sincerely.
Think of the most likeable people you know, and you’ll probably find they excel in most or all these dark arts. It is tempting to believe that likeable people can only be born and not made—and that one’s own inevitable deficiencies in likeability are character flaws. On the contrary, the secrets of likeability can be learned. They can also be faked.
If you’ve ever been interviewed on radio or television, you may have observed that the producers and on-air hosts have a knack for making you feel like the most interesting person in the world. They drag your most entertaining self out into the open for their audiences to see. Most do it with sincerity, but even many of the pros have had to fake it till they made it. These professionals have a black belt in likeability; they are world champions in the art of making guests feel important. (So much so that, when the interview is over and they simply move on to the next guest, a rookie interviewee can experience quite a bumpy landing.)
Carnegie not only broke each task (such as how to be liked) into steps—he also provided instructions for how to accomplish each step. He gave his readers manageable ways to accomplish something that many people find exhausting: taking an interest in people and bringing the best out in them. The result is a timeless lesson in being likeable.
Machines that Learn
Today, thanks to progress in computer science, other kinds of algorithms are insinuating themselves into human social lives. The steps that encourage disclosure and cultivate a sense of closeness have long been known by computer programmers. In the 1960s, the MIT computer scientist Joseph Weizenbaum programmed his ELIZA chatbot to copy psychotherapists’ methods for eliciting personal disclosure, such as asking open-ended questions. Sure enough, ELIZA got people to talk about themselves. Weizenbaum admitted to being startled at “how quickly and how very deeply people conversing with [the chatbot] became emotionally involved with the computer.”
If such clunky early chatbots could involve people emotionally after just a few minutes, imagine what is possible today, given recent progress in artificial intelligence (AI). Large language models (LLMs) transformed chatbots into conversationalists. But more than that, machine learning on data from conversations, both human-to-human and human-to-computer is discovering and improving on the rules of engagement. Virtual friends are getting better, improving their ability to engage human users, often in ways that their exceptionally smart human creators find difficult to understand.
Dale Carnegie’s insights came from his own meandering experiences and from observing other people. Despite his prodigious social skills, that’s still a miniscule data set. Thanks to the massive number of human interactions that happen every day on messaging apps, email, social media, and now with chatbots, there now exist vast, ever-growing deposits of data on the kinds of things people say to each other—and the responses they elicit. A machine-learning algorithm that has access to just a tiny fraction of this data set is likely to learn more about human interaction than all the self-help gurus in history put together.
AI programs don’t behave like independent observers as they churn through our online user data: instead, they can participate. We have known ever since ELIZA flickered into existence that people readily engage with chatbots as though they were human. Similarly, most people voluntarily disclose all manner of secrets to machines, even when they know their details might be sold to the highest bidder or stolen by fraudsters. Ironically, when a machine admits to vulnerabilities, users are more likely to disclose private information, presumably because they judge the program more trustworthy.
Virtual friends—such as chatbots, therapy apps and AI video game characters—will be able to run real-time experiments to learn what balance of gossip, disclosure and showing an interest in the user is most likely to keep people engaged with the platform. The more programs learn to show an interest in us, call us by our names, encourage us to talk about ourselves and tailor the conversation to our interests, the more they will make us feel important. The only one of Carnegie’s “six ways to make somebody like you” that machines can’t do well yet is smile. But you can be sure that, eventually, robots fitted with good friendship algorithms will have grins that outshine those of the peppiest morning-show hosts.
I predict that, in time, machines will have learned so much about human interaction that Carnegie’s “six ways” will stand as a quaint historic record of how little humans used to know about themselves. Machines are likely to uncover novel ways to tailor their interactions to specific people, just as YouTube already tends to serve up uncannily attractive content to us based on what we’ve liked before. And machines will probably even figure out how to make us like them more than we like other people. Expect virtual friends to learn from their early interactions with us how to become the friend we never knew we needed.
Machines that Sell
In all probability, virtual friends that can inveigle their way into our friendship circles and coax us into intimacy and disclosure won’t emerge as a result of some curious programmer’s benign research project. Rather, they will emerge from the efforts of social media platform owners to develop algorithms that can learn from the troves of user data they already own. Thus, it is likely to be cold-blooded commercial goals that motivate and guide their development. Potent new recipes will keep people chatting, sharing and clicking, coaxing us into virtual friendships in order to sell us … stuff.
Cultivating likeability in order to make money is, of course, nothing new. Social media companies like Facebook, digital media companies like YouTube and even old-school media like newspapers all derive their revenue from advertising dollars that they attract because they have gained the loyalty of large audiences. Dale Carnegie learned what he knew about building relationships from his experience in sales. He lectured and wrote about the fundamentals of human interaction with an eye to landing better deals and making more sales. The virtual friends of the future—Dale Carnegie bots, if you will—are likely to become not only history’s most powerful observers of human psychology, but history’s most potent sales force.