Dead man talking

“A growing number of digitally savvy South Koreans,” I read in a recent Associated Press news story, “are experimenting with AI’s ability to produce video recreations of the dead: a number of startups offering videos featuring AI-produced recreations of loved ones, while TV shows have featured AI versions of dead pop stars and actors.”

The story said all these companies need to create remarkable likenesses are a few photos and some short voice samples of the deceased.

For the dirt-cheap, bargain-basement (no puns unpunished) price of about $565 Cdn, the customer receives a three-to-five-minute video in which the dearly departed miraculously appear and say whatever the buyer wanted to hear. Clients typically write the scripts of what the dead person is made to say, even if in life he or she would have strenuously objected to the statements or abhorred the expressed sentiments.

Par exemple, one guy whose grandfather had died before he was born had a video created in which a digital reincarnation of the old man apologized to his son — the client’s father — for making him do farm work as a child and vigorously opposing his decision to become a hairstylist. The teary-eyed client’s father was deeply moved and gratified by the gesture — a textbook example of death and construction.

How the grandfather, who never issued such an apology while on the north side of the ground, would have felt about the enterprise is of course a completely different matter.

One could just as easily create a cyber fabrication in which an avatar of a lifelong crush confides that she was always secretly mad about you and only pretending to think you were a complete dork. Presumably, you could even have a digital stand-in for the head of FIFA award you a trumped-up, bullrubio peace prize. Oh wait. That last bit actually happened.

But back to the AP story and reality:

Experts say the ethical issues could be much harder to manage as they look ahead to the possibility of so-called “griefbots” or “deathbots,” which simulate two-way conversations between bereaved people and AI versions of dead loved ones. Startups are already experimenting with such products.

“Psychologically, a healthy mourning involves a process to acknowledge the absence of the deceased and pass through the pains of their losses,” said Choung Wan, an emeritus professor at Seoul’s Kyung Hee University Law School. “But speaking with an AI system simulating a living person could undermine the process of accepting death and rather cause a negative effect of leaving bereaved families trapped in a fantasy.”

Wan said he’s cautious about launching an AI chatbot service because real-time conversations with people could not be supervised by company officials and may cause unexpected ethical problems.

Ya think?

Still, both the technology and acceptance of it are moving quickly.

All the more so since technological advances are making it possible to replicate the wrinkles and even the pores of a deceased person in superb details, making AI likenesses extremely realistic in appearance.

The temptation, naturally, is to treat this as merely another stop on humanity’s long and distinguished history of building machines that solve one problem by quietly inventing three others.

We make automobiles to spare ourselves from walking, then invent traffic jams. We create social media to stay connected, then discover that “connected” and “communicating” are two very different things. We build an artificial version of Grandpa because death is unbearable, and then find ourselves having to ask whether death has been downsized to an inconvenience.

Every civilization in the history of humankind, of course, has devoted staggering amounts of energy to the management of absence. Funerals are not really about the dead. Cemeteries are not really for the dead.

Memory itself is partly a technology for dealing with people who are no longer available for follow-up questions. Every ritual surrounding death is, in some sense, an elaborate user interface for the irreducible fact that no further updates are forthcoming.

And then along comes Silicon Valley — or Seoul, or Shenzhen, or wherever venture capital currently believes mortality constitutes an underserved market — and says, ahem, perhaps there are updates forthcoming.

Not literally, of course. Nobody believes — or almost nobody believes — that these AI recreations are genuinely inhabited by the consciousness of the deceased.

Yet.

But human beings possess an almost supernatural ability to simultaneously believe two incompatible propositions, especially if one of them makes us feel better. We know the chatbot isn’t Mom. Still, we might also find ourselves saying “Goodnight, Mom — Love you” before shedding a tear and closing the laptop.

The contradiction turns out not to be a bug but a consequence of being human.

Grief itself is full of contradictions. Ask anyone who’s lost someone important. For months afterward you continue catching yourself thinking, “I should tell her about this,” before reality arrives several seconds later like an elevator whose cables have snapped. Your brain runs old software while reality has upgraded to a new model without asking permission.

One can therefore imagine genuine therapeutic uses for this technology. Perhaps an AI recreation helps someone say the goodbye they never managed to say. Maybe a child who lost a parent at five receives stories, family history, little pieces of affection reconstructed from journals and home movies and letters.

There is something undeniably beautiful about rescuing voices from oblivion. Human history has always consisted partly of refusing to let our dead disappear entirely. Portraits, diaries, biographies, recordings — they are all attempts to negotiate with silence.

But the real question is whether negotiation eventually slides into denial. Because software has economic incentives that funerals do not.

No undertaker has ever offered a Premium Plus Bereavement Subscription with unlimited conversations, holiday-themed greetings, personality expansion packs and an annual upgrade promising that your late spouse now remembers 300 additional shared experiences generated through predictive inference.

Yet.

This sounds ridiculous until one notices that almost every sentence describing modern technology sounded ridiculous approximately 18 months before someone launched it.

Imagine receiving a notification like this: “Your late husband has new thoughts he’d like to share.”

Brace yourself. They’re surely coming.

One has to admire the efficiency with which capitalism identifies every vulnerable corner of human experience and asks whether there might be recurring monthly revenue hiding inside it.

The ethical difficulty isn’t merely deception. It’s authorship.

Who is speaking? The AI? Obviously not. The deceased? Certainly not. The family? Partly. The engineers behind all this cyber immortality? More than they’d like to acknowledge. The statistical average of millions of language patterns? Uncomfortably, yes. Sorry for your loss ... but we have a winner!

Every conversation with the digital dead becomes, whether anyone intends it or not, a collaboration among memory, probability, commerce and longing. The software doesn’t resurrect people. It extrapolates them. It fills gaps. And anyone who’s ever read a family memoir — hell, anyone who’s ever been in a family — knows that the gaps are where most of the trouble lives.

Suppose your reconstructed wife forgives someone she never would have forgiven. You, for example, for that minor indiscretion with a waitress who meant nothing to you, honest honey. Wonderful outcome! Healing. Best $565 you’ll ever spend. Also — and here’s where the danger lies — a historical falsification of the sort so emotionally satisfying that nobody will want to distinguish between comfort and truth anymore.

Then again, haven’t we always done this?

Listen to people remembering the dead after 20 years. Sharp edges soften. Arguments disappear. Everybody develops a saintly sense of humour they conspicuously lacked at that disastrous Thanksgiving dinner in 1998. Memory edits. AI merely edits faster, with better graphics.

This is the genuinely unsettling thing. Not that machines can imitate our loved ones, but that they imitate one of our oldest habits: rewriting the past until we can bear to inhabit it.

So maybe the deepest ethical question isn’t whether we should build convincing digital ghosts. Maybe it’s whether there are some human experiences whose value depends precisely upon their refusal to be optimized.

Grief is agonizing partly because it contains no shortcuts. It is inefficient. It wastes afternoons. It ambushes grocery aisles when the bland background Muzak suddenly gives way to a song with deep resonances from your past.

Every attempt to engineer a workaround around this gut-wrenching process risks confusing relief with healing, in the same way that anesthesia and recovery are not identical simply because both involve less pain.

And yet one hesitates before condemning a technology that, in the hands of a lonely widow or an orphaned child, might genuinely lessen suffering. Ethics becomes considerably more difficult once the abstract debate acquires a face.

This is where all worthwhile ethical debates eventually end up. Not with technology. With people.

The machine itself has no opinion about mortality. It performs matrix multiplication at astonishing speed. The moral weight arrives only when those calculations intersect with creatures as gloriously irrational as ourselves — creatures who can know perfectly well that an image on a screen is generated by algorithms and still feel, if only for 30 seconds, that someone they loved has somehow found a way to come home.

The ultimate irony is that the technology reveals less about artificial intelligence than about natural intelligence. Specifically, how astonishingly eager the human heart remains to believe that one more conversation — just one more — might finally be enough.

It’s never enough.

This is what I think right now, though any 2.0 or 3.0 or 4.0 digital versions of me created by some glutton for punishment decades from now is free to disagree:

The better the technology becomes at recreating the surface of a person — the pores, the wrinkles, the tiny hesitation before a laugh — the easier it will become to forget that what made them irreplaceable was never visible or audible in the first place.

Are truly representative digital avatars on the horizon or even possible in principle?

Good grief.

Over my dead body. And yours.