001: Are CAPTCHAs still useful?
On verifications and validation
Hello, world. This is not written by AI, but by someone that has been tinkering with it.
Are CAPTCHAs still useful?
When we’re verifying that we’re human and not a bot, online services ask us to pick out images, or decipher gibberish text. Can’t generative AI solve these now? Or is it too expensive for spammers to use AI to solve them?
When someone is creating a video or photo to prove that it’s recent, they hold up a newspaper. Because it shows information that would not have been known or created the date or time in question. So that’s using world knowledge as verification. It’s using someone else’s creation as verification.
Is there an equivalent if I, as the reader, wanted folks online to prove that they’re not a robot?
This is a blurry line because there’s proving that it’s not an automated system creating, publishing, and responding to others without human intervention. Then there’s proving whether the content has been generated and augmented or edited and customized by a person. What if it’s just a person using generative AI to create or edit their images or content? How much would I care?
I think what matters is whether there was human intervention. If there isn’t a human involved, I want to know it’s a robot/system.
Speaking of CAPTCHA, it stands for: Completely Automated Public Turing test to tell Computers and Humans Apart. And it’s defined as “a type of security measure known as challenge-response authentication.”
So the challenge and response part reminds me of crime shows and movies where one person is asking another whether they’re a cop. And the challenge-response can sometimes be in the form of, commit this violent crime, to prove who you are, or at least cross a moral boundary that you can’t undo.
So maybe that’s how challenge-response interactions will go with robots and scammers. Ask them to do things that are still not possible by generative AI systems. Ask for anything that current systems will trigger the “sorry, as an LLM, I cannot…” but that will only work with the legit systems. There will be modified systems that will be comfortable going past the LLM moral line.
Maybe instead of moral, it’s asking it to tell a joke, or have it solve common sense logic questions that don’t work as well in LLMs yet. So this challenge-response line will move over time.
Maybe instead of text and logic, it’ll be multi-modal challenges, or time delayed, or multi-channel challenges. I’m talking about LLMs, but also reminded of scams and how they try to use different information to prove that they’re legit in order to gain your trust.
What are the other ways to handle challenge-response situations?
Let’s phone a chatbot
Here’s the part where I asked an LLM about challenge-response systems. In the process of writing my prompt to the chatbot, I asked for digital and physical world examples, and interactions between human to system, and human to human, this was a distinction that I hadn’t thought about until I had to break it down for the system.
So here’s the list of examples from the bot:
Digital (Human-to-System) Examples:
Password Authentication
Two-Factor Authentication (2FA)
Biometric Authentication
Challenge-Response Tokens
Digital (Human-to-Human) Examples:
Secure Messaging Apps
Digital Signatures
Non-Digital (Human-to-Human) Examples:
Challenge Coins (Military Tradition)
Secret Handshake
Safe Words or Phrases
Non-Digital (Human-to-System) Examples:
Physical Access Control
Voice Recognition Systems
What’s interesting from this list it’s that the physical examples remind me of doors and locks. Locking a door and having a key is a challenge-response interaction. If you don’t have a key, or something else to break down the lock, you’ve failed the challenge. and the door laughs at you. Same with our day-to-day logging into systems. Digitally, the system asks us for passwords before granting access.
At this point, everything seems like a challenge-response interaction. It’s a bit of a stretch, but even job interviews are challenge-response interactions. The hiring team’s goal is to verify and evaluate your skills before they let you into the company to join a team. Dating—it’s also a series of challenge-responses interactions to determine values, personality, and fit. It’s a matching process, where the challenge-response isn’t as direct at times.
The challenge-response at events is typically, “what do you do for work?” If it’s relevant, the response is accepted, and the conversation continues, otherwise, folks move on from each other. Maybe this is why the question is off-putting, because everyone is constantly challenging each other and being challenged by some rando with a drink in their hand.
Phoning in a friend (chatbot), here’s a list of topics and softer challenge questions you can use, besides asking what they do for work. Make sure to ask every question before you move on from the conversation.
Hobbies and Passions:
"What do you enjoy doing in your free time?"
"Any favorite hobbies or activities you're passionate about?"
"Do you have any special interests or things you love to do?"
Media and Entertainment:
"Have you read any good books or watched any interesting movies/shows lately?"
"What type of music are you into?"
"Any favorite podcasts or YouTube channels?"
Travel and Experiences:
"Do you like to travel? Any favorite destinations?"
"Have you been on any memorable trips recently?"
"What's the most exciting experience you've had?"
Food and Culinary Preferences:
"Are you a foodie? Any favorite cuisines or dishes?"
"Do you enjoy cooking or trying new recipes?"
"Any favorite restaurants or cafes you recommend?"
Fitness and Sports:
"Do you have a favorite sport or physical activity?"
"Have you tried any interesting workout routines or classes?"
"Any favorite sports teams you follow?"
Goals and Aspirations:
"What are your career goals or aspirations?"
"Is there something you've always wanted to learn or achieve?"
"Any personal projects or goals you're currently working on?"
Volunteering and Causes:
"Do you engage in any volunteer work or support any specific causes?"
"Is there a social or environmental issue that you're passionate about?"
"Have you been involved in any community projects?"
Technology and Gadgets:
"Are you into technology? Any favorite gadgets or apps?"
"Have you tried any interesting tech-related activities recently?"
"Do you have a favorite type of technology?"
Art and Creativity:
"Are you into any form of art or creative expression?"
"Do you enjoy visiting art galleries or museums?"
"Any favorite artists or art styles?"
Learning and Education:
"Do you enjoy learning new things? Any favorite subjects?"
"Have you taken any interesting courses or workshops recently?"
"Is there something you've always wanted to learn more about?"
36 questions that lead to love
Each time I think I’m done writing this post, something comes up. I’ll end this post with this: the 36 questions that lead to love; (NYTimes paywall link) or if you want to see the questions, here’s a link to the study, the questions are on page 12 of the PDF.
It’s organized into three sets, that escalate in intensity and go deeper. What’s interesting is that the study also includes two sets of 36 questions. One that’s meant to generate closeness, and one that’s small-talk. Fascinating to see the contrast. I’ll dig into them and report back juicy details.
Challenge question to sign off: what are you doing with your life? ‘cause chatbots don’t have goals and intentions. yet.