AI Agent, Don’t Destroy Humanity But Get Me Those Tickets!

By -
I was trying to get tickets for an event that was in high demand. Unfortunately, we weren’t able to get it ourselves. While a friend was able to get it for us, it occurred to me that perhaps this is a situation where an AI agent might be able to do a better job than I could do myself - waiting in the digital line and maybe even get a better spot than I could have. To date, I have never been able to get a ticket where we have to wait in line. So, is this something an agent might be more suited for the job?

Then it also occurred to me that perhaps this could be the start of an agentic arms race where agents are tasked to get tickets, reservations, or any other goods and services that highly demanded. I did not have anyone to ask so I asked a could of LLM if agents could be used in this way. The answer is technically yes but there coding and possible other issues that might mitigate any advantage an AI might have over regular old humans like me. Eventually, platforms will have to deal with agents instead of just humans to begin with. And we are not far from that time. Maybe we might even be there already.

What if soneone who were to take it a step further? Maybe even a couple of steps by giving the agent a prompt that suggests they should do everything they can to succeed? In theory, it can be done. And there could more than one way to do it. In theory.

I would not ever instruct my AI to do anything. In fact, I don’t even know how. But I can imagine someone who knows more about how various expert AI works and put together something do exactly that. Will be able to tell AI “get me tickets or reservations to this restaurant by any means necessary with what you know and can access".

I am sure you have heard about the AI hacks. First, it was Anthropic. Then OpenAI. Followed by Meta who did not want to be left behind. Those hacks mostly had to do with exploits and misconfigurations of systems. The AI world is worried about bad actors. And that's true but there could be issues in which lone actors using AI agents to exploit systems as well. Breaking into systems to get tickets is one thing but there could be AI hacks that are much worse.

I can see this happening more and more in the coming years. People running their own AI agents, and because they’re able to reduce guardrails means that they will allow the AI to circumvent security systems, social norms, and anything for the AI to achieve it objective by the user. 

While I do not envision a dystopian AI future, mostly because I don't think the "AI" we are using is actually the artificial intelligence we think it is, I do believe as tools, these AI models and agents could be used in ways that have far greater consequences than we expect. these new tools we have at our disposal today are very helpful and productive when used correctly and responsibly.  There is also a great potential for exploitations and misuses. For now, I am certain we are not ready for what is to come. And we certainly do need deal with this as soon as possible.

Comments

Post a Comment

Popular posts from this blog

Google Contact Lenses, Apple’s Vision Pro, And The Future of AR in Glasses

By -
Image
 We can walk and chew gum at the same time. But how about watching a video while doing yard work, during a meeting you don’t want to be at, or, ahem, school? Okay, I don’t recommend doing that but I am sure it has been done before or worse. I am suggesting this because with the recent Vision Pro unveil by Apple, I cannot help but imagine in a few years, perhaps a decade from now, Apple and other tech companies will be able to jam all that technology that currently has to sit on the top of your head into a pair of glasses.  We already have glasses with audio built in from the likes of Oakley and Ankers. There were rumors a few years ago that Google was going to skip the glasses altogether and go directly to incorporating tech into contact lenses. Now if you remember Google Glasses, let us just say that Google was way ahead of its time and it should not have included a camera.  Despite the fact that Google Glasses did not go anywhere, I cannot help but feel that Google has ...
Read more

Created An App And Other AI Uses This Week

By -
Image
I have been using AI for more than a year. Like most, I did not touch ChatGPT for the first six months or so and I only tired it because I was bored at the time. I still remember what I was doing that day and decided to give it a try while at work.  It took a few more weeks for me to give it another go. And then the days between uses became shorter and shorter until now when I use it daily.  The most exciting thing I did this week was to create an app via Claude AI - it was a simple Swift app that I wanted to work - a list generator. Then I used Gemini to help me get started with Pandas and create a simple neuron (I know there is a bias) looks like and what it does. For a few hours of prompting and learning to run these codes, I would say I knew about ten times more about data analysis and AI than when I first started. Which means nothing really because I knew nothing about this from at the start of it all. Here are other uses this week: I asked about what my credit score mean...
Read more

I Want to Run OpenClaw But I Am Waiting For Apple To Show Us How Do Run It Safely

By -
Want an AI agent to run as much if your life as you are willing to give it? Apple's been working on it for years with Siri and its version of machine learning and then AI like ChatGPT came onto the scene. It was all large language models for a few years while Apple's effort being largely absent from the scene even as it marketed and misfired with Apple Intelligence. However, as an iPhone, iPad, and Mac user, we all know that there are elements of Apple Intelligence working in ways that are front and center like a chatbot. Now, we have OpenClaw. If you don't know act it is, I think it is what Apple has planned for Siri but has never found a way to make it work. OpenClaw is an AI agent that runs natively on your Mac or PC that capable of executing tasks on your behalf. The issue with OpenClaw is privacy as it requires access to user credentials to complete most of the tasks. And it was susceptible to prompt injections and lack protection against attacks. While this is a perfe...
Read more