LaMDA and CAPTCHA – Should we be worried?
Over the past few days, the internet went into a frenzy after a Google engineer claimed that LaMDA – a Google AI Neural Network is sentient. His proof was a series of chats the engineer had with LaMDA, which convinced the engineer that LaMDA is in fact, a conscious entity.
What this means is that computers can now pass tests designed to tell computers and humans apart – with flying colors. With CAPTCHA being such a test, what does this mean for CAPTCHA tests? Are we now forever doomed to mailboxes full of spam messages?
Is LaMDA conscious?
Whether LaMDA is in fact, conscious or not is, unfortunately, beyond the scope of this article. believe me; I would love nothing more than to spend a couple of days writing a long post about what it even means to be sentient or conscious. It’s a rabbit hole I am more than happy to slide down, but we’ll keep our focus on CAPTCHA this time around.
Should we be worried?
While LaMDA did effectively pass a CAPTCHA test, it does not mean it can pass any CAPTCHA test. AI machines are very domain-specific – they’re very good at doing one thing and nothing else. LaMDA is, in essence, conversational AI meaning it can converse with intelligence. According to the engineer, it can generate chatbots that are requirement-specific, making it something of a mother chatbot.
As such, while LaMDA is very good at conversations, to the point it can make highly-educated engineers believe it is sentient, it is not very good at picking out pictures of crossroads, for example. That’s the domain of Visual AI, which are systems built to recognize specific objects. While such systems do exist, they operate very differently from conversational AI neural networks. For all intents and purposes, they are two completely different technologies.
How is CAPTCHA defending itself?
CAPTCHA already has mechanisms in place to thwart AI image recognition efforts. This is why pictures are so grainy and hard to distinguish, even for humans. Such images are incredibly difficult to distinguish, even for modern visual AI systems.
CAPTCHA has also evolved from asking users to pick pictures. ReCAPTCHA V3, for example, focuses on user behavior rather than pictures, making it even more difficult for computers to pass the test.
AI systems are also expensive to build, run, and maintain. They use extraordinary amounts of power, making them unfeasible for spammers peddling scams to adopt.
In fact, many spammers are resorting to cheap labor – paying people on platforms such as Mechanical Turk to fill in CAPTCHA tests and send spam. This is more of a challenge to CAPTCHA than AI.
What will the future bring?
While AI is still in relative infantcy, we have to recognize that Moore’s Law tells us such systems will, one day, become commonplace as computing power becomes more readily available. Even refrigerators will one day include what we consider today to be cutting-edge AI. This, however, will not happen for some time, giving CAPTCHA enough time to continue evolving to counteract any technologies adopted by spammers in their relentless pursuits to annoy us.
The post LaMDA and CAPTCHA – Should we be worried? appeared first on WP White Security.
*** This is a Security Bloggers Network syndicated blog from WP White Security authored by Joel Farrugia. Read the original post at: https://www.wpwhitesecurity.com/lamda-captcha-should-we-be-worried/