The artificial intelligence changes at extremely fast rates the way digital content is created and distributed. Images, videos, voices and texts that until a few years ago required professional equipment and specialized knowledge, can now be produced within minutes by anyone who has a computer and access to modern AI tools.
The result is impressive. But at the same time it creates a new challenge: how can we know if what we see is real?
This discussion has begun to become more and more intense in recent months as the cases of publication on social media of photographs and videos processed with AI tools that show a false image increase. The a few days ago incident with the Minister of Health, Adoni Georgiades is such a typical case.
The reality is that technology has evolved to such an extent that many times even experienced users find it difficult to distinguish the original from the artificial.
What should be pointed out is that image manipulation is nothing new. For decades there have been photo and video processing tools. The difference today is that with GenAI existing tools - which are provided at little cost or even free - are not limited to processing existing content. But they can create entirely new faces, new voices, and new scenes that never existed.
A video may now show a politician making statements he never told or a businessman announcing false news about his company. Corresponding to tools «voice cloning» (voice clowning), a voice can be created almost identical to that of a real man.
The problem becomes even bigger in an environment of social networks where information is moving rapidly and often without control.
The signs to be noticed
Although modern AI models have improved significantly, there are still indications that can help a user understand that something is not true.
In videos, one of the most frequent signs is the abnormal movements of the face or lips. Small contemporaries between sound and image, strange expressions or excessive «Perfect» skin can be evidence of synthetic content.
Also, eyes and hands still make several AI systems difficult. Despite huge improvements, strange finger movements, wrong shadows or unnatural reflections appear in several cases.
In sound, artificial voices often have too stable tone, absence of physical pauses or a slight «mechanical» speech rate. Of course, as time passes, even these signs are reduced.
One of the most important defense tools against deepfakes, as the specific content is called, is but the verification of the source.
A video that suddenly appears in an unknown profile on TikTok or an anonymous post on X needs much more attention than content from official organizations or reliable news networks.
The search for the same video or news in more than one reliable source remains critical. Especially in times of elections, geopolitical crises or major events, deepfakes can be used for misinformation or manipulation of public opinion.
Platforms and detection technologies
Large technology companies are now trying to develop AI content tracking mechanisms. Tools such as digital watermarks, metadata authority standards and content prevention technologies attempt to «sign» digitally the original content.
Organizations such as Coalition for Content Prevention and Authenticity promote standards that allow the certification of the origin of an image or a video, recording whether it has been modified with AI.
At the same time, companies such as OpenAI, Google and Microsoft invest in watermarking technologies and recognition of synthetic content. However, this is a constant «equipment race», as they improve detection tools as they evolve and creation tools.
Η μεγαλύτερη πρόκληση ίσως δεν είναι τεχνολογική αλλά κοινωνική. Στην εποχή του GenAI, η ψηφιακή παιδεία αποκτά νέα σημασία. Οι χρήστες καλούνται να γίνουν περισσότερο επιφυλακτικοί, να ελέγχουν τις πληροφορίες που καταναλώνουν και να αποφεύγουν τη βιαστική αναπαραγωγή περιεχομένου.
Το παλαιότερο ερώτημα «είναι αλήθεια αυτό που διαβάζω;» μετατρέπεται πλέον στο «είναι αληθινό αυτό που βλέπω και ακούω;».
Και όσο η τεχνητή νοημοσύνη εξελίσσεται, τόσο πιο σημαντική γίνεται η ικανότητα να ξεχωρίζουμε το αυθεντικό από το κατασκευασμένο. Όχι μόνο για την προστασία από απάτες και παραπληροφόρηση, αλλά και για τη διατήρηση της εμπιστοσύνης στον ίδιο τον ψηφιακό δημόσιο διάλογο.
