Can AI preserve your most precious memories? | Pau Aleikum Garcia

[00:00:00] [SPEAKER_03]: TED Audio Collective

[00:00:02] [SPEAKER_03]: This is TED Tech, a podcast from the TED Audio Collective. I'm your host, Sherelle Dorsey.

[00:00:16] [SPEAKER_03]: Our speaker today is Pau Aleikum Garcia. He's a media designer and founder of Domestic Data Streamers, a research and design studio that builds change through data, community, and arts.

[00:00:29] [SPEAKER_03]: In one of their signature projects, Pau and his team use generative AI to bring memories to life. It's really impressive tech. I was able to see them showcase it at TED 2024.

[00:00:40] [SPEAKER_03]: But what makes the project so special is not the precision of the images. It's their soft, blurry quality, which mirrors the fluid and often imperfect nature of human recollection.

[00:00:51] [SPEAKER_03]: The work's emotional depth is outstanding. And even more important, it's using technology as a tool to reconnect with the lost pieces of ourselves.

[00:01:02] [SPEAKER_03]: This project can even support patients battling with Alzheimer's and Parkinson's.

[00:01:08] [SPEAKER_03]: Pau's work carries so many implications for the future of AI, and he lays it all out in this talk.

[00:01:14] [SPEAKER_03]: But before we dive in, a quick break to hear from our sponsors.

[00:01:18] [SPEAKER_04]: How will humans and machines work together in the future?

[00:01:37] [SPEAKER_05]: We spend so much time discussing how the world's changing.

[00:01:40] [SPEAKER_05]: It would be absolutely absurd to believe the role of the CEO is not.

[00:01:44] [SPEAKER_04]: This is Imagine This, a podcast from BCG that helps CEOs consider possible futures for our world and their businesses.

[00:01:53] [SPEAKER_04]: Listen wherever you get your podcasts.

[00:02:00] [SPEAKER_02]: You've probably heard about artificial intelligence and chat GPT, but do you know the person in charge?

[00:02:06] [SPEAKER_00]: On our podcast, Good Bad Billionaire, we tell the stories of how the world's billionaires made their money.

[00:02:11] [SPEAKER_02]: We're telling the story of Sam Altman, the boss of OpenAI who make chat GPT.

[00:02:16] [SPEAKER_02]: He became a billionaire this year, but his wealth has nothing to do with artificial intelligence.

[00:02:20] [SPEAKER_00]: He actually got rich investing in other tech startups.

[00:02:23] [SPEAKER_00]: Listen to Good Bad Billionaire to learn how he did it and whether he's good or bad.

[00:02:27] [SPEAKER_02]: That's Good Bad Billionaire, wherever you get your BBC podcasts.

[00:02:30] [SPEAKER_03]: And now, Pal Alekum Garcia takes the TED stage.

[00:02:34] [SPEAKER_01]: A year ago, I was in front of Carmen, a 90-year-old woman from Barcelona.

[00:02:41] [SPEAKER_01]: I asked her, Carmen, what is your earliest memory?

[00:02:46] [SPEAKER_01]: Now, for someone who has almost seen a century unfold, this can be a difficult question.

[00:02:51] [SPEAKER_01]: I was half expecting her to say, son, my earliest memory is of you asking me this question.

[00:02:59] [SPEAKER_01]: But the truth was that it was from 1941.

[00:03:03] [SPEAKER_01]: She was six years old and her mother would pay another family so they could enter their house and go up to their balcony.

[00:03:12] [SPEAKER_01]: What was particular about that balcony was that it was facing La Modelo prison.

[00:03:16] [SPEAKER_01]: During that time, during the Spanish dictatorship in Spain, it was a political prison.

[00:03:22] [SPEAKER_01]: And her father, a doctor for the anti-fascist front, was a prisoner there.

[00:03:26] [SPEAKER_01]: So the only way they could see each other was from that balcony to the window of the prison.

[00:03:34] [SPEAKER_01]: And that was her earliest memory of him, between bars, through that street.

[00:03:41] [SPEAKER_01]: I asked her, Carmen, would you like to have an image of that memory?

[00:03:47] [SPEAKER_01]: And she said, yes, of course.

[00:03:48] [SPEAKER_01]: I would love to show to my family what I experienced, the things that I went through,

[00:03:54] [SPEAKER_01]: so they can remember where we all come from.

[00:03:57] [SPEAKER_01]: And this is how we started this experiment, trying to transform a memory into an image.

[00:04:02] [SPEAKER_01]: In this case, using generative AI.

[00:04:05] [SPEAKER_01]: We started by creating a description of the memory, a prompt to say so.

[00:04:10] [SPEAKER_01]: And then Carmen generated tens of images.

[00:04:13] [SPEAKER_01]: You could see her going through all of them.

[00:04:15] [SPEAKER_01]: But it was not until she saw these two images that something clicked in her.

[00:04:21] [SPEAKER_01]: She pointed at the image and said, yes, this is me and my mother in the balcony.

[00:04:25] [SPEAKER_01]: I can see the haircut.

[00:04:27] [SPEAKER_01]: It was as if she had found something long-lost.

[00:04:33] [SPEAKER_01]: From there, we did something else.

[00:04:35] [SPEAKER_01]: We used that image as a starting point to generate a video,

[00:04:39] [SPEAKER_01]: but this time not only showing her and her mother,

[00:04:43] [SPEAKER_01]: but also showing her father looking back at them from the other side of the street.

[00:04:47] [SPEAKER_01]: When she saw it, there was a long silence.

[00:04:51] [SPEAKER_01]: And later she told me that it had been a long time since she had seen an image of her father.

[00:04:59] [SPEAKER_01]: These are not factual recreations of the past.

[00:05:02] [SPEAKER_01]: This feels a bit more like dreams.

[00:05:04] [SPEAKER_01]: In fact, they are.

[00:05:06] [SPEAKER_01]: They are visualizations based in memories with their imperfections and undefinition.

[00:05:12] [SPEAKER_01]: That's how we discovered that blurry undefined images work much better than hyper-realistic ones

[00:05:19] [SPEAKER_01]: when we are trying to reconstruct memories.

[00:05:22] [SPEAKER_01]: That's why early generative AI models work much better than the ultimate state-of-the-art ones.

[00:05:28] [SPEAKER_01]: It's not the factual accuracy that moves us,

[00:05:32] [SPEAKER_01]: but the emotional truth that we find embedded into some of these systems.

[00:05:37] [SPEAKER_01]: And we call this project synthetic memories.

[00:05:39] [SPEAKER_01]: It's one of the several projects we are doing at Domestic Data Streamers.

[00:05:44] [SPEAKER_01]: A team of researchers, designers, and engineers exploring the impact that generative AI can have in society,

[00:05:49] [SPEAKER_01]: going from media literacy to cultural heritage, and, of course, memories.

[00:05:55] [SPEAKER_01]: Now, memories are the architects of our identity.

[00:06:00] [SPEAKER_01]: Memories remind us who we are.

[00:06:04] [SPEAKER_01]: And visual memories are very important.

[00:06:06] [SPEAKER_01]: We all have a picture of someone we love in our wallet, in our phones, in our homes.

[00:06:13] [SPEAKER_01]: Visual memories shape our sense of self, and they can shape our sense of belonging to a specific place.

[00:06:19] [SPEAKER_01]: They can teach us things from the past.

[00:06:21] [SPEAKER_01]: And because all that, they can actually make us understand, in a deeper level, how we react to things.

[00:06:31] [SPEAKER_01]: Unlike organic memories, which are formed and stored in the human brain,

[00:06:36] [SPEAKER_01]: synthetic memories are visual memories from a person's past,

[00:06:39] [SPEAKER_01]: which have been never documented or lost.

[00:06:42] [SPEAKER_01]: And that we can now, using Gen AI,

[00:06:45] [SPEAKER_01]: transform from a text or an oral description into an image.

[00:06:49] [SPEAKER_01]: Now, there are many reasons why we can lose visual memories and memories itself,

[00:06:54] [SPEAKER_01]: but the most common one is the one that comes with aging,

[00:06:58] [SPEAKER_01]: and diseases like Alzheimer's or Parkinson's.

[00:07:01] [SPEAKER_01]: I myself have lived through that in my family,

[00:07:04] [SPEAKER_01]: and I guess many of you have had similar experiences.

[00:07:07] [SPEAKER_01]: It's a terrible disease, as you not only see the memories fade away,

[00:07:12] [SPEAKER_01]: but also the character and identity of someone that you love.

[00:07:16] [SPEAKER_01]: So, two years ago, we started to create these meetups to meet people from very different disciplines,

[00:07:24] [SPEAKER_01]: trying to figure out how synthetic memories could be used in different social setups.

[00:07:29] [SPEAKER_01]: And then I met David, a social worker in a nursing home.

[00:07:33] [SPEAKER_01]: He was taking care of a group of Alzheimer's patients,

[00:07:36] [SPEAKER_01]: and he was using a therapy called reminiscence therapy.

[00:07:41] [SPEAKER_01]: Now, reminiscence therapy is a kind of therapy that uses music

[00:07:44] [SPEAKER_01]: and old photographs from a person's past

[00:07:48] [SPEAKER_01]: to trigger these very visceral, emotional memories.

[00:07:52] [SPEAKER_01]: Think of it as kind of lubricating an old rusty lock,

[00:07:56] [SPEAKER_01]: making it much easier to open the doors of certain forgotten rooms that we have in our head.

[00:08:01] [SPEAKER_01]: And it has been proven to, for certain people,

[00:08:05] [SPEAKER_01]: help very much in the prevention of depression, of cognitive decay,

[00:08:09] [SPEAKER_01]: and in some cases even improve cognitive abilities.

[00:08:12] [SPEAKER_01]: So, we designed a pilot experiment using synthetic memories within reminiscence therapy,

[00:08:18] [SPEAKER_01]: and for over two months, 12 patients did both individual sessions and group therapy.

[00:08:23] [SPEAKER_01]: And we could see a direct relation between their level of engagement and their cognitive abilities,

[00:08:28] [SPEAKER_01]: pointing out that this could not only be any way of doing therapy,

[00:08:32] [SPEAKER_01]: but also it could be another way to detect early signs of cognitive decay.

[00:08:40] [SPEAKER_01]: Now, this image was generated with Nuria,

[00:08:45] [SPEAKER_01]: a 96-year-old woman from a nursing home.

[00:08:49] [SPEAKER_01]: And she was telling me about one of her favorite memories of her husband.

[00:08:54] [SPEAKER_01]: It was during Sundays he would use to read the newspaper in the living room.

[00:08:58] [SPEAKER_01]: And she would always play this old, jumpy gramophone with some music just to annoy him.

[00:09:05] [SPEAKER_01]: When she saw the image, she was so happy.

[00:09:09] [SPEAKER_01]: Her eyes light up.

[00:09:10] [SPEAKER_01]: But after a while, she looked at me and she said,

[00:09:13] [SPEAKER_01]: but, Pao, we have a problem here.

[00:09:16] [SPEAKER_01]: This man over here is not my husband.

[00:09:18] [SPEAKER_01]: And I said, how come?

[00:09:20] [SPEAKER_01]: And she said, as laughing,

[00:09:22] [SPEAKER_01]: Well, my husband was much more ugly.

[00:09:26] [SPEAKER_01]: Pointing once again that love is not just blind, but also inner sighted.

[00:09:32] [SPEAKER_01]: This is far from being a cure of Alzheimer's,

[00:09:35] [SPEAKER_01]: but it could be a way of making the journey a bit less daunting.

[00:09:39] [SPEAKER_01]: Another patient told me,

[00:09:40] [SPEAKER_01]: It feels a bit like finding your glasses in your head after searching for them everywhere.

[00:09:46] [SPEAKER_01]: It's a small but very significant relief.

[00:09:48] [SPEAKER_01]: And from a medical perspective,

[00:09:50] [SPEAKER_01]: it could be a new way to temporarily enhance cognitive abilities,

[00:09:54] [SPEAKER_01]: making it much easier for family members and caregivers

[00:09:59] [SPEAKER_01]: in moments of distress or disconnection.

[00:10:01] [SPEAKER_01]: So the pilot experiment was promising enough

[00:10:06] [SPEAKER_01]: to partner up with the AgeWell Network here in Canada,

[00:10:08] [SPEAKER_01]: together with researchers from the University of Toronto

[00:10:11] [SPEAKER_01]: and the University of the British Columbia.

[00:10:13] [SPEAKER_01]: And during the summer, we will be exploring and designing an ethical framework

[00:10:18] [SPEAKER_01]: to protect both family members and patients,

[00:10:20] [SPEAKER_01]: as well as comparing this therapy to other therapeutic interventions

[00:10:23] [SPEAKER_01]: and exploring how we can make it truly accessible

[00:10:26] [SPEAKER_01]: for both families and institutions all over the world.

[00:10:30] [SPEAKER_01]: But as you can imagine,

[00:10:33] [SPEAKER_01]: synthetic memories in this project can go beyond the research on dementia.

[00:10:37] [SPEAKER_01]: There are many other reasons why we can lose visual memories.

[00:10:40] [SPEAKER_01]: There is the loss that comes from sudden displacement

[00:10:44] [SPEAKER_01]: due to war, political persecution or natural disasters,

[00:10:49] [SPEAKER_01]: situations in which our photo albums, hard drives, phones, diaries

[00:10:55] [SPEAKER_01]: can be lost, damaged or just left behind.

[00:10:58] [SPEAKER_01]: Synthetic memories,

[00:10:59] [SPEAKER_01]: it's a way to recover a part of this past heritage that has been lost

[00:11:04] [SPEAKER_01]: and that right now is at risk of being forgotten.

[00:11:07] [SPEAKER_01]: It's a new way of reconstructing a past that has been hidden to our eyes.

[00:11:14] [SPEAKER_01]: Now, soon we will open the first public office

[00:11:18] [SPEAKER_01]: for visual memory reconstruction in the city of Barcelona

[00:11:22] [SPEAKER_01]: with the support of the city council.

[00:11:23] [SPEAKER_01]: The idea is to make a city-scale experiment,

[00:11:27] [SPEAKER_01]: a place where any citizen will be able to generate their own memories,

[00:11:32] [SPEAKER_01]: to reconstruct a part of their past together in community,

[00:11:37] [SPEAKER_01]: understanding which are the subjectivities of the past

[00:11:40] [SPEAKER_01]: from different communities.

[00:11:43] [SPEAKER_01]: This image was sent to me not long ago by David, the social worker.

[00:11:49] [SPEAKER_01]: And he did it with a dementia patient

[00:11:52] [SPEAKER_01]: that for a very long time had not spoken with his daughter.

[00:11:57] [SPEAKER_01]: They were in kind of bad terms.

[00:11:59] [SPEAKER_01]: And he told David that he was terrified about horses,

[00:12:04] [SPEAKER_01]: but because his daughter loved them,

[00:12:06] [SPEAKER_01]: he would bring her to ride every week during her childhood.

[00:12:10] [SPEAKER_01]: He asked David to generate this image

[00:12:12] [SPEAKER_01]: of both of them riding in this kind of dreamy landscape.

[00:12:16] [SPEAKER_01]: This was not a memory.

[00:12:17] [SPEAKER_01]: This never happened.

[00:12:18] [SPEAKER_01]: It was a remembrance.

[00:12:20] [SPEAKER_01]: It was a way that he wanted to remember his daughter.

[00:12:25] [SPEAKER_01]: With his permission, David sent this image to his daughter,

[00:12:29] [SPEAKER_01]: who because of that,

[00:12:30] [SPEAKER_01]: after almost 10 years without speaking to each other,

[00:12:33] [SPEAKER_01]: came back to Barcelona to speak back to him.

[00:12:36] [SPEAKER_01]: Afterwards, she told David that when she saw the image,

[00:12:41] [SPEAKER_01]: she realized that her father still had good memories of her.

[00:12:46] [SPEAKER_01]: An image can work as a witness to the world,

[00:12:49] [SPEAKER_01]: We know that.

[00:12:50] [SPEAKER_01]: But it can also change the way we see it.

[00:12:53] [SPEAKER_01]: We have no tools that can help us understand

[00:12:56] [SPEAKER_01]: a time that was not ours.

[00:12:58] [SPEAKER_01]: Experiences from others that have never been shown in images.

[00:13:02] [SPEAKER_01]: And bridge realities that were hidden to our eyes.

[00:13:06] [SPEAKER_01]: Montserrat Roig, a beloved Catalan writer,

[00:13:10] [SPEAKER_01]: once wrote,

[00:13:11] [SPEAKER_01]: Si hay un acto de amor,

[00:13:13] [SPEAKER_01]: Aquest es la memoria.

[00:13:15] [SPEAKER_01]: If there is an act of love,

[00:13:17] [SPEAKER_01]: that is to remember.

[00:13:19] [SPEAKER_01]: And that is truly my hope,

[00:13:21] [SPEAKER_01]: that we all get the opportunity to love and be loved in that way.

[00:13:27] [SPEAKER_01]: Thank you very much.

[00:13:27] [SPEAKER_03]: That was Pau Aleikum Garcia at TED 2024.

[00:13:42] [SPEAKER_03]: TED Tech is part of the TED Audio Collective.

[00:13:45] [SPEAKER_03]: This episode was produced by Nina Lawrence,

[00:13:48] [SPEAKER_03]: edited by Alejandra Salazar,

[00:13:50] [SPEAKER_03]: and fact-checked by Julia Dickerson.

[00:13:53] [SPEAKER_03]: Special thanks to Maria Latias,

[00:13:55] [SPEAKER_03]: Farrah Degrange,

[00:13:56] [SPEAKER_03]: Corey Hajim,

[00:13:57] [SPEAKER_03]: Daniela Balareso,

[00:13:59] [SPEAKER_03]: and Michelle Quint.

[00:14:00] [SPEAKER_03]: I'm Sherelle Dorsey.

[00:14:02] [SPEAKER_03]: Thanks for listening,

[00:14:03] [SPEAKER_03]: and talk to you again next week.