I had a few interesting conversations last month during the AI-native workshop in Porto. A few things were talked about in depth, and I’ll list them here briefly as they also apply to some AI posts I’ve seen recently in the Mauritian techspace.

These are my thoughts. If I am wrong about something, I’ll be glad to discuss it.

Number one: “AI (genAI) is like a calculator. If you don’t use it, you’ll get left behind.”

I’ve heard this many times, echoed a lot in the techspace, and I don’t think we can compare genAI to a calculator. I haven’t seen a calculator that can:

  • create deepfakes of people for malicious purposes,
  • accelerate energy and water consumption (think datacenters),
  • disrupt critical thinking skills (skill erosion),
  • hallucinate results and create slop,
  • be trained on stolen data,
  • be trained on stolen creative work,
  • a training corpus built by exploited data labelers,
  • be used as a surveillance tool.

There is more to add, and we should remember that calculators were not used in the early stages of learning. Instead, the teachers made sure the students understood how to do algebra. But I’ll be honest too as I was initially excited about this analogy until I found out most of the things listed above.

Number two: “AI is inevitable. Get on board or get left behind.”

It’s the same people saying genAI is like a calculator who parrot that it’s inevitable. From what I’ve seen and heard, it’s some people whose paycheck depends on these genAI tools who say it’s inevitable like some sort of a sales pitch (or AI cheerleading). Before LLMs and genAI, decision-making in pricing, recommendation systems, loan approvals, job application screening, and much more was already influenced by algorithmic operations. The conversation on this topic should be more on “Yes, AI is pervasive, not inevitable. So the focus should be on responsible AI.”

To add, AI luddites and others would love The Resist List at airesistlist.org.

Number three: Don’t board the hype train without listening to the AI experts instead of the AI fans.

You do not need a PhD in AI/compsci to be an AI expert. Listen to people who understand the technology and avoid if possible the AI fans who see all of this as magic and want AI everywhere. Because, they don’t see the multiple points of failure, the issues of scaling with increasing users, and so on. Look towards people asking questions over the people hyping AI. Or demand evidence before blindly adopting any AI-powered solutions.

Number four: “You cannot secure LLMs with guardrails.”

There’s a NIST article stating that “a fixed set of guardrails placed on AI is not universally robust against adaptive adversarial prompts”.

So if you keep adding guardrails or rule-based boundaries every time a threat or vulnerability appears, you end up with an infinite number of guardrails. I have to point out that guardrails are not totally useless and we do need them until there are better ways of securing these systems.

So, these are my brief thoughts on these things after hour-long conversations. I get the hype around it but the problem is hype with no questions attached is… problematic. If there are things I missed or got wrong about, or you want to have a conversation on these topics, feel free to reach out.
And I hope this post made you aware of a few things worth questioning.