L'ascesa dei deepfake: come l'IA alimenta nuove minacce alla sicurezza e alla privacy

  • 2 settimane fa
In questa puntata di Sci-Tech, esploriamo la rapida rinascita dei deepfake, guidata dai progressi dell'IA generativa.
Transcript
00:00Many of us have received dodgy emails asking for money, promising things too good to be true.
00:09Now artificial intelligence could be used to try and dupe us.
00:13A deepfake is an AI produced copy of a human being.
00:17And this technology makes fraud of a new kind very very possible.
00:22One bank is fighting deepfakes with deepfakes.
00:27For scammers it is a powerful tool they can use to steal your money.
00:35Welcome to this edition of SciTech from London where we're delving into the rise of deepfakes.
00:40Video, audio and pictures that appear to be real but are not.
00:44Just like that introduction there, that wasn't me.
00:47In fact it was my avatar created using artificial intelligence.
00:52Would you be able to tell the difference?
00:56My avatar was made with my consent through a platform run by Synthesia.
01:01It produces AI generated content for legitimate purposes.
01:06Like training videos, marketing and customer service.
01:10With strict safeguards in place.
01:13We would never generate an avatar of someone without their explicit consent.
01:16The other one is control. We apply content moderation on the videos that are created.
01:20And the third is around collaboration.
01:22So working with industry groups, working with governments and policy makers
01:26to make sure that the technology is developed and used responsibly.
01:32Deepfakes that have malicious intent copy the image of real people
01:36from photos, videos and audio to misrepresent them or an organization.
01:42Professor Michael Woodridge, an AI expert at the University of Oxford,
01:47says the technology is rapidly improving.
01:53If you can take it to the point where you can be in a call with somebody
01:57that you think is your boss looking at somebody's video
02:00and it's an AI produced video on the other end,
02:04then it starts to become very difficult to know what is real and what is fake.
02:08In a survey published in July, just over 40% of respondents age 16 plus
02:14said they've seen a deepfake in the last six months.
02:18Less than one in ten were confident in their ability to spot one.
02:24One bank has now launched a campaign deploying its own deepfake content
02:29to raise awareness of the dangers.
02:32As you might have guessed, this isn't me, this is a deepfake
02:35created to warn you about deepfakes.
02:38Technology companies are also tackling deepfakes.
02:42This new phone feature pits AI against AI,
02:47helping users to spot what's real and what's not.
02:51The on-device AI carries out a facial scan of a person in an image or a video.
02:55The algorithm analyzes things like the composition,
02:58whether or not there are any pixel imperfections,
03:00or whether or not there are any anomalies to the facial features
03:03of the person in that image or video.
03:05And crucially, it allows the user to see in real time
03:08whether that person is real.
03:10Let's take a look at an example specifically.
03:12So if we have a look at this lady, what are your thoughts here?
03:17It's difficult to tell, isn't it?
03:19I'm not sure I would know whether that was a deepfake or not.
03:22Let's have a look. If I click on the magic capsule here,
03:24you can see right away that the on-device AI is doing its work.
03:27It's carrying out a facial scan to determine if this person is real.
03:30And within seconds we have an answer.
03:32Here we can see that a suspected face swap has been detected
03:35and the likelihood of this person not being real is 93%.
03:39And this alerts the user that, firstly, the person is not real,
03:42but secondly, and crucially, they could be at risk of being caught up
03:45in some sort of scam or fraud.
03:47So basically now you're a technology company fighting technology.
03:51Well, ultimately for us this is about ensuring the privacy
03:54and security of our users.
03:56I think we all know that technology has been developing
03:58at quite a pace for some years now, thanks in part to AI.
04:01And inevitably there will be people who are determined
04:04to exploit that through the creation of so-called deepfakes.
04:07And on an industry level, I think there's an onus on all of the key players
04:10to play their part in stemming the rise of deepfakes moving forward.
04:15Fraud is not the only concern.
04:17Deepfakes are also fuelling intimate image abuse online.
04:21One victim whose real identity is protected by this avatar
04:25was deepfaked by a former friend.
04:28I felt confused and taken aback.
04:31A documentary about this kind of deepfake abuse
04:34spawned a campaign group now demanding tougher regulations
04:38to block publishing sites.
04:41I have spoken to over 30 survivors of this
04:44and in every single case their lives have shrunk.
04:48It has had an impact on their job, on their career,
04:51on their sense of their body, on their sense of their identity.
04:54The stakes could not be higher.

Consigliato