top of page
  • Writer's pictureKatarzyna Hasnik

Episode #3 AI answers: Why AI has a problem with facial recognition? Revolut - Fail.

Updated: Aug 17, 2023

Revolut AI Facial Recognition - exposed weakness. This is what AI thinks about facial recognition.

Open AI ChatGPT Revolut AI Facial Recognition Fail

Open AI ChatGPT Revolut AI Facial Recognition Fail

Some folks think AI is a solution to everything, some are fearful...well, I am too sceptical and too balanced about it to go each side. AI has a place, just like most of the inventions of humankind.


But again, I was exposed to the AI facial recognition weakness. like again. This simple task would be solved by a human in a fraction of a second, even a child could handle it quickly.


AI is in trouble. A lot of trouble and a man is needed to take over.


No validation can be performed for several inputs. Uploaded many different in good light correctly aligns ID, passport, ID card and driver's license. All are approved with beautiful UI task completion confirmation. Then the selfie was taken multiple times in good light and again I got task completion positive confirmation. I've passed all steps.


But the AI can't get it...


I changed my hair colour from brown to blonde... and as a result, the AI is unable to recognize and verify my face, regardless of the correct details. Dyeing hair is a problem for AI and who is behind coding those scripts for facial recognition software?


It is my first interaction with Revolut, as they finally reach the Irish Banking Market. It's an extremely poor verification process that has been going on for days now.


Maybe I shouldn't be their customer with my personal and business bank? Maybe I should stay in a hard-core old-school bank where I can go and sit on the comfy armchair whenever I want. Maybe it's good to have ATMs and branches where money can be launched… maybe one day all innovation will just pass and we'll all go back to HUMAN TO HUMAN INTERACTION. Invaluable in this case.


It will take 1 second to approve my photo ID with my face in the local branch by a human. AI cannot do the job 100% and it is stuck at 60%. Repeating the process several times, taking photos of various IDs, and pulling my hair into a ponytail didn't help making multiple selfies don't help.


No other reason was indicated. It's all because I dye my hair.


Reportedly e.g. IBM phased out facial recognition as early as 2020. Maybe it's time for Revolut too. Get SMARTER, NOT NECESSARY AI'TER... ;)


Revolut has scored 1 star so far...we'll see how things go with Human to Human customer service. It's way too slow right now. Internet Banking should be fast and smooth, but if it was, I would have nothing to write home about. ;)


These user real-life scenarios are good stories. As they are real, contra adverts.


Poor AI, you have too much credit to solve or destroy the whole world. You are far from every path, which is fearful and overwhelming, over exciting people give it. They do live in Black Mirror or Utopia fantasy.


However, if dyeing hair is a problem now, what will happen if I will get facial reconstruction due to accidents or anything even more than this? Would I be that time not exist at all for AI, 10%, 20%, 30% only of verification? Maybe 1% or 0%.


If there will be no Human Labour at all, how would I pass this ever...


That's why More AI, more Human needs to be involved in TAKING CARE OF AI.


BabySITTING AI - new job title. :D


AI is just a tool, annoying, and here's what AI will say about itself. Yes, per se, because the AI is not a person, even if given a name. It's not even a pet. This is a tool. Sometimes these tools go out of the window if they are no longer in use. All in all, I'm waiting for something smart, much more, like teleporting people and objects to another place to save time, effort, soil, and trees and treating car/train/plane transport only as an additional option, a vintage style of travelling. I will be excited about this. Artificial intelligence is not fast enough, at least for treatments in hair salons.


Should I dye my hair brown so the AI can verify me?


Funny, let the AI fight to get more companies to recognize that facial recognition is not only against privacy but also a waste of money in many cases, like the one with Revolut. But of course, the AI is not to blame here, but the contributions of those who coded it. Because behind every innovation, there is a HUMAN.


Open AI ChatGPT Revolut AI Facial Recognition Fail

Me: Why AI has a problem with facial recognition?


ChatGPT: AI's problem with facial recognition can be attributed to several factors:

  1. Bias and Fairness: Many facial recognition systems have been found to exhibit biases, especially with respect to race and gender. This is often a result of biased training data that disproportionately represents certain demographics. If the training data is not diverse and balanced, the AI model can learn to perform poorly on underrepresented groups.

  2. Data Quality: The performance of a facial recognition system heavily relies on the quality and diversity of the training data. If the data is of low resolution, taken under varying lighting conditions, or from different angles, the AI model might struggle to accurately recognize faces.

  3. Variability in Appearance: Faces can look very different due to factors like age, facial hair, makeup, and accessories. AI models need to handle this variability, and if the training data doesn't adequately cover these variations, the model's performance might suffer.

  4. Privacy Concerns: Facial recognition technology raises serious privacy concerns. People are worried about their personal data being used without their consent or knowledge. This has led to public outcry and regulatory actions that limit the deployment of such technology.

  5. Adversarial Attacks: Facial recognition systems can be tricked by adversarial attacks. By introducing small, carefully crafted perturbations to an image, an attacker can cause the AI system to misclassify or fail to recognize a face.

  6. Context and Environmental Factors: Recognition can be influenced by environmental factors such as lighting, weather conditions, and camera quality. An AI model might struggle to generalize well to different contexts.

  7. Uncontrolled Environments: AI models trained on controlled datasets might struggle when faced with real-world scenarios, where lighting, poses, and backgrounds vary widely.

  8. Ethical and Legal Concerns: The use of facial recognition technology has raised ethical and legal questions, particularly in terms of surveillance and potential misuse. This has led to bans and restrictions on its deployment in certain contexts.

  9. Rapid Technological Evolution: The field of AI, including facial recognition, is evolving rapidly. This can lead to situations where the technology is deployed before it's mature enough, resulting in lower accuracy and unintended consequences.

  10. Human-Like Discrimination: AI systems are often trained to mimic human decision-making processes, which can include biased decisions. If human assessors have exhibited biased decisions in the past, the AI model might inadvertently learn and replicate these biases.

Addressing these challenges requires careful curation of diverse and representative training data, robust model architectures, ongoing testing and validation, transparency in deployment, and adherence to ethical and legal guidelines.


Are you ready for AI to replace human labour?


Open AI ChatGPT Revolut AI Facial Recognition Fail
44 views0 comments

Comments


bottom of page