Getting a new phone this month meant moving from thumb authentication to face authentication. It seems to work pretty seamlessly when I pick up my phone – the devices are getting smarter. Also, this month I’ve noticed Gmail regularly Captcha-asking for additional authentication to prove I’m not a bot. Is agentic A.I. causing more issues? I don’t know. But it made me think about how verifying my identity has changed over my lifetime.
Everything was done manually when I was young. No computers or internet. I don’t remember how I was verified when I first entered primary school. How did the school and teachers know I was not an impostor? In my most recent videochat with my mother (a former schoolteacher, long retired), I asked her and she told me about the systems they would use. My first major verification that I vaguely recall was the primary school national exams. Apparently, the government sent us letters with an entry slip and a unique number; and this allowed me to take my exams where I think I had to carefully write the unique number on all my exam papers. I vaguely remember teachers drilling us to do this. And while I don’t recall exactly, I think our own teachers were also the ones who verified us because we’d been in their classes for a whole year.
After the age of twelve, when I had secondary school examinations, we all had to bring in our identity cards, and place them on the corner of our desks for each exam. Each of us had a specific desk that had our name on it, and the invigilators who were not our teachers, walked around (clipboard in hand) to verify each of us via the identity cards. I still experience this process when I’m at the airport, visiting the bank in person, checking in to a hotel, picking up my badge at a conference, or any other situation where I need to verify who I am because the verifiers don’t know me and wouldn’t recognize my face. If you lived in a small village and never had to leave, everyone knows everyone and verification is easy. But in an era of urbanization, global travel, and not knowing your neighbors, verification becomes trickier.
The age of the internet has made authentication even more challenging. Are you who you say you are? How does the system know? There are logins, passwords, two-factor authentications, additional questions for information unique to you, and now voice and face verification. These are going to get more stringent as A.I. makes it easier to “fake” more characteristics. We’ll be increasingly up the wazoo in verification.
“Authenticating” is the title of the second chapter in Brian Christian’s The Most Human Human which details his experience as a confederate in an annual Turing Test competition. He’s trying to prove he’s the human against an A.I. competitor. The chapter opens with a story about a man with phonagnosia. He cannot recognize anyone’s voice and growing up had assumed his voice was distinctive because everyone else recognized his voice but he couldn’t recognize anyone else’s which made phone calls an interesting case of guess-the-identity for him. He couldn’t voice-verify. In another vignette, someone easily breaks into the email account of a public figure simply by selecting “I forgot my password” and then verifying information based on internet searches.
The meat of this book chapter, however, is about what might be unique between a human conversationalist and a chatbot. Many of the successful chatbots in the early competitions were able to steer the conversation to avoid the tricky out-of-book situation, which is a reasonably strategy when the conversations are timed in a speed-dating like format. Successful bots typically had a single programmer devoted to developing the bot’s personality so that it would seem like a single coherent individual. The bot felt like a singular You. Today’s A.I. large language models however were developed with the opposite philosophy: use weighted statistics from millions of disembodied conversations. Apparently, this is why A.I. translators are weak on long literary novels which require a singular coherent voice throughout, but do just fine on shorter snippets.
When you’re conversing with a chatbot today, you’re conversing with a multitude of voices averaged into a response. Your interlocutor is Legion for they are many. They contain multitudes. You are no longer talking to an individual but a host of ghosts. I’ve never had an extended conversation with a chatbot (I have better things to do with my time), and my queries have usually been specific and chemistry-related; I dabble in exploring if chatbots can help my students gain a better understanding of chemistry. So I don’t personally know if I would ever feel that a chatbot feels like a friendly human; I know some of my students do enjoy their chatbot chats. And it may be that sufficient familiarity and multiple chats provides its own authentication, for better or worse, now that chatbots can access their memory store of their personal conversation from you and draw from it. I suppose this is what Personalization is all about. At some point the chatbot might feel like a You. But that’s because of you.
No comments:
Post a Comment