teissTalk: Robot wars – using AI to combat the deepfake threat

teissTalk host Geoff White was joined by Dr. Janet Bastiman, Chief Data Scientist, Napier AI; Ansgar Koene, Global AI Ethics and Regulatory Leader, EY; Chuck Brooks, Adjunct Professor, Georgetown University.

 

Views on news

 

Intel has  introduced FakeCatcher, which it says is the first real-time detector of deepfakes —synthetic media in which a person in an existing image or video is replaced with someone else’s likeness. Intel claims the product has a 96% accuracy rate and works by analysing the amount of light that is absorbed or reflected by blood vessels in living tissue in milliseconds.

 

There are also many responsible and authorized use cases for deepfakes for employee training, education and ecommerce. Being dependant on lighting and skin tone, the technology has its own limitations. Even someone having make-up on can affect results. Open source deep fake solutions are now readily available, you don’t have to go to the dark web to access them.

 

Bad actors have their own developers and data scientists, and they keep up with the latest technological advancements. As a technique of spotting deep fakes, this technology has already been out for more than two years.

 

Bad actors will obviously test their deepfakes against any detection tools that they can get hold of and they can resort to tricks such as adding artificial make-up to a fake video and reduce to reduce the accuracy of the tool.

 

Detection tool’s test is when they are in use – too many false positives or not detecting a high percentage of fakes can be turn-offs.

 

How much deep fakes are a live problem for businesses?

 

While deepfakes already create problems for businesses, they shouldn’t shift the focus completely away from more basic techniques of fraud such as putting a different face on a photoshopped image. Software to doctor images is available at no cost.

 

Even if these images don’t pass, all criminals get back is a message asking for a better-quality image. Meanwhile, deep fakes are expected to become a problem in recruitment too. All the technology is already available to create a fake image and a fake voice for yourself and interview with it for a job.

 

The EU’s legislation on AI focuses on the safety, security and human rights of individuals. But it also has an element that if someone creates an automated system with the potential of misleading people by making them believe they are interacting with a human, they must notify their users about it.

 

The upside of conversations about how AI can serve criminals’ purposes is that relevant information and positive use cases will also get out into the public domain, which provides experts with opportunities to explain which fears about AI are unjustified. Having said that, deep fakes are feared to give a new boost to phishing, one of the main types of online fraud.

 

In a non-representative Teiss Talk poll, 82% of participants said that deep fakes are a problem NOW.

 

The panel’s advice

 

To combat deep fakes, you need to start with the basics, educating your workforce regarding what is possible. If they are in doubt, they can make deepfakes pronounce a complicated word and see if they make mistakes or use complicated backgrounds for video calls to make their deepfakes look less authentic (there will be aberrations as these backgrounds are generated on the fly).

 

If you use deep fake detection tools, always be aware of the GDPR implications of the use of personal data

.

To check out a video creating platform, click here.