We can already make something that looks like them, sounds like them and answers back. Whether that makes it them is an entirely different question.
If you’d asked me a few years ago whether artificial intelligence could recreate someone after they’d died, I’d probably have filed it somewhere between flying cars and a fridge intelligent enough to notice the milk had gone off. In fact, I wouldn’t really have known what “artificial intelligence” was.
As it turns out, we got dead relatives first.
AI can already use photographs, videos, voice recordings, texts, emails and stories to create a digital version of someone who is no longer alive. Some can talk, some can appear on screen and some can answer questions the real person was never actually asked.
Which is where things get interesting. There’s quite a difference between hearing Great Grandma tell a story she genuinely recorded and hearing AI Great Grandma candidly tell you what she thinks about your new boyfriend. Especially if Great Grandma shuffled off her mortal coil in 1987.
What is a griefbot?
No one’s really settled on a name, so you’ll come across:
Griefbot. Deadbot. Thanabot. Digital ghost. AI afterlife. Digital immortal.
None of them seem particularly appealing.
“Griefbot” sounds like something that should arrive from Amazon with three attachments, a charging cable and somewhere to put H2O so it can cry convincingly.
Broadly, they all describe technology that allows us to interact in some way with people after they’ve died. But there’s an important difference between replaying the real person and letting AI invent new answers for them.
When the real person did the talking
At one end of the scale are systems that record someone while they’re still alive. They answer questions about childhood, family, work, relationships, regrets and memories. Later, the family asks a question and the technology finds the appropriate recorded answer.
So if Great Grandma tells you she once ran away to Blackpool with a saxophone player, there is at least a reasonable chance she really did. AI hasn’t simply decided the story needed jazzing up.
Then AI starts answering for them
This is the bigger leap. Feed AI enough of someone’s messages, emails, photographs, voice recordings, opinions and stories and it can begin to imitate them, potentially extremely well.
It can look like them, sound like them, know things about their life and respond to questions they were never asked. At which point the human brain is going to do quite a lot of the remaining work.
It looks like Dad. It sounds like Dad. It knows we went to Cornwall in 1996 and that Dad couldn’t stand coriander. You ask it a question. Dad answers.
Except he hasn’t.
That’s the huge catch….
An AI recreation doesn’t have Great Grandma’s memories. It has information about Great Grandma’s memories.
And when AI doesn’t know something, it has another rather unfortunate talent. It can make something up, very convincingly.
That’s irritating when AI invents the opening hours of a restaurant. It’s rather more serious when your dead mother tells you she’d always regretted something she may never have regretted at all. Particularly if it concerned your conception.
The better these recreations become, the easier it may be to forget where the real person ends and the invention begins.
Can someone be recreated after they’ve already died?
Potentially, yes. It depends on what they left behind.
Someone who died in 1972 leaving twelve photographs and a box of birthday cards presents a rather different challenge from someone who spent twenty years emailing, messaging, posting videos and arguing on Facebook.
Future generations may have quite the opposite problem. They’ll have far too much of us, including things we’d rather hadn’t made it into the training data.
So is it really them?
No. However convincing it becomes, it is still a simulation built from information about a person.
But knowing that intellectually may not stop it feeling real. If something has your husband’s face, speaks in his voice, knows where you spent your honeymoon and calls you by the private nickname only he used, the fact that you’re technically talking to software may become slightly academic.
That’s where grief tech stops being merely clever and becomes considerably more complicated.
But what would an AI actually need to know about you?
This is the bit I find more interesting.
Photos, emails and voice recordings can tell it quite a lot, but would they tell it what really makes you you?
Would it know the phrases you always use? Your opinions about money? What you do when everything goes wrong? The recipe you never wrote down because you’ve always just known how much to put in? Why you fell out with Auntie Jean in 1984 and whether anyone should believe her version of events?
What would you tell your children about love, mistakes, getting old or choosing friends? And on a more practical level, would anybody know where you’ve put the will, what that mysterious key opens or how to get into your phone?
Those aren’t necessarily the things we leave behind on Facebook. They’re often the things families realise they don’t know when it’s too late to ask.
Which is why I made The Answers
While I was writing Ghost Grandad, I kept coming back to one thought. You don’t actually need artificial intelligence to preserve the important bits of yourself. You can just answer the questions while you’re still here to answer them.
So I created The Answers.
It’s a free guided journal designed to capture the things families may one day wish they’d asked. Not a formal autobiography. Not “I was born in York in 1960 and attended St Whoever’s Primary School” unless you particularly want to record that.
It’s about the things that make you recognisably you. Your stories, your opinions, your mistakes, the things you believe, the advice you’d give, what makes you laugh and how you behave in a crisis.
And some distinctly unromantic practical things too, because preserving your personality for future generations is lovely. Making sure they can find the paperwork is also quite useful.
You can download The Answers free, print it, ignore half the questions, add your own, give it to somebody else or leave it in a drawer for your family to discover one day.
No AI required.
Although if somebody does decide to build an artificial version of you in fifty years’ time, at least you’ll have given it better source material than your Facebook posts after three glasses of Sauvignon Blanc.
Download The Answers free here
If somebody wanted to recreate you one day, what would they need to know that they could never learn from your photos, emails or social media?
You can answer that question in The Answers.