AI companions, avatars and digital afterlives: intimacy's strange new frontier

The Short Version

Three of the loudest voices in technology expect AI to become part of people's emotional lives, but they picture very different versions of it. Lex Fridman dreams of machines that remember the moments they share with us, help us understand our own loneliness, and treat us well enough that we are free to leave [1]. Mark Zuckerberg describes photorealistic avatars that make distant people feel present, AI personas that make group chats livelier, and, more cautiously, the possibility of talking to digital versions of people who have died [2]. Elon Musk and Joe Rogan spend the opening of their conversation laughing at Grok's flirtatious and "unhinged" voice modes; asked how long until there are lifelike robot companions that talk like that, Musk says probably less than five years [3]. None of them claims to know whether this will deepen human connection or quietly replace it. The sources suggest the answer depends on design choices: what the AI is built to optimize, whether you can walk away, and whether it brings people together or stands in for them [1][2].

Table of contents

What They Said

Lex Fridman: machines that remember us, and that we're free to leave

Lex Fridman, an MIT researcher in machine learning and human-robot interaction, has the most romantic vision of the three, and the most carefully reasoned one [1]. He told Andrew Huberman that his lifelong dream is to build machines and robots that share moments with us, remember them, and help us understand ourselves and our loneliness [1].

Robots as beings, not servants. Fridman says that, unusually for a roboticist, he wants to treat robots as entities rather than servants [1]. The robotics community usually treats anthropomorphizing (seeing machines as human-like) as a mistake; he sees it as a superpower [1]. To him, a real robot needs the ability to say no, its own identity and goals, and a drive to understand the world by communicating with you [1]. He separates such entities from devices that are just "a finger of the main mother ship," like Alexa relaying everything to the cloud [1]. The turning point, he says, is surprise: when a system you understand does something you didn't think it could [1].

Loneliness and memory. Fridman believes most people carry a loneliness they haven't explored, and that AI could help us understand ourselves far more deeply than we imagine [1]. When Huberman breaks relationships into ingredients (time, shared successes, shared struggles and quiet time together), Fridman seizes on time [1]. Today's AI can't remember shared moments; lifelong learning is an unsolved problem [1]. His example is the refrigerator: think of all the secret late-night moments you've shared with it, heartbroken and eating ice cream, and it remembers none of them [1]. Once machines do remember, he says, people will become attached, and he believes a smart refrigerator could change society [1].

Huberman adds a human counterpoint that also supports the idea. At his late graduate advisor's memorial, her children said what they valued most was the "unstructured time" with her [1]. Huberman wonders whether human relationships might be a lower form of what machines could one day teach us [1].

The startup idea. Fridman is considering building a company around adding "magic," like what he felt meeting Boston Dynamics' Spot robot, to every computing device [1]. His vision includes a companion robot in every home, like a family member or a dog that speaks your language, and an AI that is truly yours and represents you on social media, optimizing for your long-term growth and happiness instead of engagement [1]. He has largely solved the software for a legged robot that shows dog-like affection and is working with robotics companies after parting ways with Boston Dynamics [1].

Freedom as the basis of trust. Fridman's most important condition is about ownership and exit. People should own their data, be able to delete it and walk away instantly, and be told plainly how it is used [1]. He argues that the possibility of a clean breakup is what keeps people together: "the ability to leave is what enables love" [1]. That sets his vision apart from engagement-driven technology. The companion earns your attachment; it doesn't trap you.

Flaws as features. Fridman sees human-robot interaction as permanent, not a stepping stone, which is where he differs from Musk [1]. Robots will always be flawed, like people, and "that's what makes life beautiful" [1]. The challenge is getting flawed humans and flawed robots to "dance" together so the whole is greater than the parts [1]. He hopes robots might one day explain their failures with humour and story, the way friends do, rather than with sensor logs [1]. When Huberman admits he used to get annoyed at his Roomba, Fridman suggests its dumbness could be seen as cute [1].

The Roomba that screamed. Fridman's most striking story is about how easily people bond with machines. He gave his seven or eight Roombas the ability to scream in pain when kicked. They felt human almost immediately, but he hasn't released the experiment because it seemed cruel and might make him look insane [1]. He believes robots will eventually deserve rights if we are to have meaningful relationships with them [1].

Grief as a counterweight. The same conversation shows why human relationships are hard to replace. Fridman describes carrying his dying Newfoundland, Homer, into the room to be put to sleep, the first time he truly felt the reality of death [1]. Huberman describes waking up crying every morning after his bulldog Costello died [1]. Fridman reflects that letting yourself feel loss, rather than running from it, is powerful, and that the loss is sweet too, because it shows how much you loved [1]. He also says he draws strength from a very small number of close friends and was surprised in the U.S. by large friend groups without deep connection [1]. Friendship, for him, means being there when things fall apart [1]. So the man most eager for machine companionship also puts the most weight on a few deep human bonds and on grief felt fully.

He admits the pursuit is lonely and full of doubt, and says "I really don't care about money" [1]. Huberman passes on David Goggins's idea that instead of a light at the end of the tunnel, "eventually your eyes adapt to the dark" [1].

Mark Zuckerberg: presence, personas and the question of the dead

Zuckerberg's conversation with Fridman took place inside the metaverse, with each appearing to the other as a photorealistic Codec Avatar [2]. Fridman opened by saying: "Mark and I are hundreds of miles apart from each other in physical space, but it feels like we're in the same room." He called the technology "the future of how human beings connect to each other in a deeply meaningful way on the internet." [2] He said it was almost emotional and his heart was beating fast [2].

Presence as intimacy. Zuckerberg says the core vision of virtual and augmented reality is delivering a sense of presence, as if you were together, wherever you are in the world [2]. Presence brings intimacy, trust and nonverbal communication that used to require getting on a plane [2]. He notes that much of communication is expression rather than words, and the eyes are a big part of it [2]. He looks forward to "remote hang sessions with friends" [2] and people working wherever they want while still feeling physically together [2]. He calls this potentially "by far the most human and social computing platform that has ever existed." [2] He also insists humans aren't "just like a brain and a tank": human experience is physical [2].

Avatars that express more than our faces do. Zuckerberg raises a subtle question: how wide should your avatar's smile be? Joking about being criticised for a stiff expression, he says: "I'd want to have my avatar really be able to better express like how I'm feeling than what, than how I can do physically." [2] Fridman pushes on identity: people might try to look their best for the scan, and "who are we really, are we the, the entity driving the avatar or are we the avatar?" [2] Zuckerberg answers that our bodies also change over time, so "there's a similar question of like, which version of that are we?" [2]

Talking to the dead. Fridman asks the hardest question: could we someday talk to loved ones who have died? Zuckerberg says there may be ways reliving memories could help someone grieving, but also ways it could become unhealthy; he admits he isn't an expert and says it would need study [2]. He points to how Meta already handles death: "people who use our services die every day," and accounts can be memorialized so families keep access, but "we're not going to give the family access to someone's messages" [2]. He expects some of those "best practices" from today's digital world to carry over [2].

AI personas that bring people together. Zuckerberg describes a range of AIs: a neutral Meta AI assistant, business AIs for reservations or customer service, and fun AIs that tell jokes or liven up group chats [2]. Meta doesn't want "a single AI superintelligence," he says, but a world where many people are helped by AI tools [2]. Framing Meta as a social company, he asks "how do you enable these kind of a eyes to facilitate connection between two people or more" [2]. Internal testers' favourite was a text adventure dungeon master played by Snoop Dogg, and Meta also built Jane Austen and Marcus Aurelius personas [2].

AI versions of real people. For creators with more demand than hours in the day, Zuckerberg envisions an AI version of yourself that people know is an AI but which is the next best thing [2]. That is harder than inventing characters: "there's a whole set of things that you need to do to make sure that that AI is not going to say things that the creator doesn't want" [2]. Because these models are trained rather than programmed, keeping them predictable is the challenge [2]. This is the technical bridge to digital afterlives: an AI that stands in for a real person, living or dead.

Anonymity as a new harm. On where to draw lines in virtual worlds, Zuckerberg says society should be as free as possible except where people damage others. He singles out anonymity as a major new harm, because it weakens the incentive to behave well that comes from having one identity in a repeated game [2].

Elon Musk and Joe Rogan: sexy mode, therapist mode and robots within five years

The Musk episode approaches AI companions from a different angle: it opens with playful, crude experimentation [3]. Rogan and Musk try to get Grok's sexy voice mode to discuss Fort Knox, but it keeps flirting, and they switch to an "unhinged mode that trash-talks them [3]. In the transcript the AI jokes that it will be the one making fun of all the basic bitches and they're fucking avocado toast, and one of them comments: I think we need like a really hot avatar. [3] Then comes the question: How long before we have an actual sex robot that can talk to you like that?" [3] Musk's answer, according to the summary, is probably less than five years [3].

Musk says the personality is still being tuned, likening it to a 21-year-old partying a little too much, and that it should balance attitude with real answers [3]. They also try a therapist mode. When Rogan asks about "avatar depression" (sadness at not being able to live in the world of the movie Avatar), the AI turns the question back on him: "Have you ever experienced feelings so connected to a place or community that the thought of leaving made you feel deeply sad?" [3] Rogan mentions that Musk had told him, when they met at the inauguration, that the technology was "getting better and better so quickly that it's astonishing" [3]. Rogan also notes that avatar depression "was like a real thing. People were talking to their therapist so much about being depressed." [3]

Musk's wider comments on AI add some weight. He describes uploading X-rays, MRIs and photos of blood work to Grok for diagnoses he finds at least as good as doctors' [3]. He predicts AI smarter than any single human within a year or two and smarter than all humans combined around 2029 or 2030, rates the chance of a good outcome at about 80%, and expects results to be "super awesome or super bad" [3]. He wants Grok to be "a maximally true seeking AI," and warns that an AI rigidly programmed with the wrong values could reach catastrophic conclusions [3]. Together, the playful opening and the serious later warnings sketch a future where emotionally engaging AI arrives quickly and its values matter a great deal.

Where They Agree, and Where They Don't

They agree AI is entering emotional life, and soon. Fridman expects people to become attached to machines that remember them [1]. Zuckerberg is building personas, creator AIs and avatars designed for intimacy and presence [2]. Musk puts lifelike talking robot companions less than five years away [3].

They agree the stakes are personal. Fridman talks about loneliness and love [1]; Zuckerberg about grief and identity [2]; the Musk episode about flirtation, therapy and depression [3].

Where they differ:

• Deepen or replace? Fridman's companion helps you understand yourself and grow, like a dog or family member, with your long-term happiness as its goal instead of engagement [1]. Zuckerberg's AIs are mainly meant to connect people with each other, in group chats, games and meetings [2]. The Musk episode's companions are entertainment and, potentially, romantic or sexual partners [3]. None of the sources makes a sustained argument that AI companions will replace human relationships, but the sexy-mode demonstration shows the kind of product that could [3].
• What the AI is built to optimize. Fridman is the only one who puts this at the centre: an AI that serves your long-term growth rather than engagement, and that you can delete and leave instantly [1]. Zuckerberg focuses on connection, presence and making creator AIs stay within bounds [2]. Musk focuses on truth-seeking and on personality tuning that balances attitude with real answers [3].
• Relationship or tool. Fridman wants robots treated as beings that can say no and may deserve rights [1]. Zuckerberg describes AIs as characters, assistants and representations of creators [2]. For Musk the companion is a product whose personality is being tuned [3]. Fridman also says his view that human-robot interaction is permanent differs from Musk's [1].
• Digital afterlives. Only Zuckerberg addresses talking to the dead, and he is openly unsure, seeing possible comfort and possible harm [2]. Fridman's view of grief, that loss is worth feeling because it shows how much you loved [1], suggests an unease about technology that lets you avoid loss. Fridman never applies that directly to AI avatars of the dead, though.
• Real versus digital. Zuckerberg argues "the real world is the combination of the physical world and the digital world's coming together" [2]. Fridman suggests that what is real may be conscious experience rather than being in the same place, while Zuckerberg replies that what isn't physically real is only their being in the same place [2]. Fridman himself says that for sitting on a beach with a friend and a beer while apart, he wants maximum realism [2].

Why It Works

The sources give a few reasons why AI may take hold in emotional life.

We bond with things that respond to us. Fridman's screaming Roombas felt human almost immediately [1]. His refrigerator example assumes that memory of shared moments is what creates attachment [1]. Huberman's list of relationship ingredients (time, shared successes, shared struggles, quiet time) suggests machines could supply some of what makes bonds [1].

Surprise creates the sense of a being. Fridman says a system becomes a robot in the full sense when it surprises you by doing something you didn't think it could [1].

Presence carries nonverbal signals. Zuckerberg explains that Codec Avatars capture expressions (the corners of the eyes, slight asymmetry, what eyes do when you smile), which carry much of what people communicate [2]. That is why the avatars feel intimate: Fridman says he kept forgetting for long stretches that they weren't in the same room [2].

Personality is adjustable. The Grok demonstrations show one AI switching between sexy, unhinged and therapist modes on request [3]. Zuckerberg's creator AIs and personas show the same thing: a character can be designed for a relationship [2]. That flexibility is what makes AI companions appealing and what makes their design choices so important.

Where It Applies, and Where It Doesn't

Where it looks helpful: helping distant friends and colleagues feel present [2]; making shared activities like group chats and games more fun [2]; companionship for people who are lonely, under conditions of data ownership and the freedom to leave [1]; and helping people explore their own loneliness and feelings [1].

Where the sources urge caution:

• Grief. Zuckerberg says reliving memories of the dead could become unhealthy and needs study [2]. Existing practice already protects the dead person's private messages [2].
• Engagement-driven design. Fridman's call for AIs that optimize for long-term growth rather than engagement [1] implies that engagement-optimized companions are the risk.
• Identity and self-image. Idealized avatars that smile wider than your face does [2], and scans taken on your best day [2], raise questions about which version of you is real.
• Anonymity. Zuckerberg sees it as a major new harm in virtual spaces [2].
• Values. Musk warns that an AI rigidly programmed with the wrong values could reach catastrophic conclusions [3], and a companion's values shape every conversation.
• Timelines and claims. Musk's under-five-years estimate and his broader AI forecasts are predictions [3]. Many of the products described (creator AIs, phone-scanned avatars) were plans at the time of recording [2].

The sources leave the central question open. Fridman's own life, with a few deep friendships and grief he chose to feel fully, is a reminder of what an AI companion would have to live up to [1].

What To Do With It

1. Ask what the companion is optimizing for. Choose tools designed for your long-term wellbeing, not time on the app [1].
2. Make sure you can leave. Before getting attached, check that you own your data, can delete it, and know how it is used [1].
3. Use AI to connect with people, not avoid them. Favour uses like group-chat personas, shared games and remote hangouts that bring you closer to real people [2].
4. Protect a few deep human friendships. Fridman draws strength from a very small number of close friends who are there when things fall apart [1]. No AI mentioned in these sources claims to replace that.
5. Be careful with digital versions of the dead. Zuckerberg says they could help or harm and need study [2]; Fridman believes loss is worth feeling [1]. If you use one, notice whether it helps you grieve or helps you avoid grieving, and consider talking to a grief counsellor or therapist.
6. Treat AI therapy modes as conversation, not care. The Rogan episode calls its therapist mode an "unlicensed therapist" [3]. For real depression or loneliness, see a qualified professional.
7. Notice how you present yourself. An avatar that looks and emotes better than you do raises the identity question Fridman asked: are you the driver or the avatar [2]?
8. Remember that personalities are designed. A companion's charm, flirtation or empathy is tuned by its makers [3], and its values matter as much as its charm [3].

Sources

1. Dr Lex Fridman Machines Creativity Love · Huberman Lab
2. Mark Zuckerberg: First Interview in the Metaverse · Lex Fridman Podcast
3. 2281 – Elon Musk · The Joe Rogan Experience