It’s mid-2024 & AI Chatbots are everywhere. Call me a maverick rebel, but have we gone a bit overboard with chatbots? Does everything really need a chatbot & what situations do chatbots work best in?

Well, how did we get here?

I’ve been around long enough that when I started using computers in the early 1980s, they mostly used “command line” interfaces. This meant to get your computer to do anything, you had to type in instructions like so:

C:\>FORMAT A: /S /V /F:360KB /C "My Bootable Floppy Disk"

Unsurprisingly that was complete gibberish to most people. Personal computers didn’t really take off until Apple & Microsoft built their own versions of Xerox’s PARC GUI (Graphical User Interface) - descendants of which we all use today. The GUI is so ubiquitous you don’t even think about using your mouse or touch screen to operate your devices.

GUI have ruled for decades, receiving a boost from the development of modern touchscreen smartphones in the late 2000s. The past two decades have seen tech companies try to push beyond the humble GUI with mixed success. We’ve seen smart glasses, VR headsets & most successfully (to date) voice assistants like Alexa, Google Assistant & Siri.

This is not my beautiful house

Voice assistants are excellent examples of chatbots that reached mass market - with around half of Americans using them. But they certainly come with their own challenges. It’s socially awkward to talk to your device in public, but even in the privacy of your own home voice assistants are often clunkier than what they’re supposed to replace.

We have a smart speaker setup in our home & often, I want to adjust the volume - which I can do by:

1. Manually pressing the buttons on some (but not all) of the speakers.
2. Adjusting the volume via an app on my smartphone.
3. Saying: “Hey Google, volume down”

As many of you will know, trying to adjust the volume with voice commands can be super annoying. Unless you know your desired percentage/ level of speaker volume you might be in for a disruptive “conversation” about speaker volumes when all you want to do is listen to music or a podcast. In most cases, I just use the app or buttons to adjust the volume, as it easier & the GUI gives me a level of control far beyond shouting at my smart speakers.

More complicated tasks bring another level of frustration - I can ask my speakers to play music via Spotify, but the accuracy is hit & miss. A recent example is the upcoming singer/songwriter “Ciara Lea” - I enjoy her single “Feel It”. But as she isn’t well known, Google Assistant/ Spotify will often play a track by popular R&B singer “Ciara” or a different song called “Feel It”. And don’t get me started on playing a live/ remixed version of a song when I want to hear the original or confusing Underworld’s live album “Everything Everything” with the band “Everything Everything”.

Things changed at the end of 2022 when OpenAI released their ChatGPT chatbot to the world. Suddenly there was a chatbot that could converse at close to a human level. A good chunk of the tech world lost their minds at the possibilities.

Where does that highway go to?

The tech industry has a long history of over-promising & under-delivering. Elon Musk has been promising fully self-driving cars for almost a decade & even with billions of kilometres of Tesla driving data, they still haven’t become reality.

As per-usual the AI gold rush kicked off with a lot of clever & creative people predicting AI-powered revolutionary tech was just around the corner. Playing with the earlier chatbots (who had a bit more personality/ anarchy in their models) it was hard not to get swept up in the excitement, that this would change everything.

18 months down the road & most of the AI products on the market are… well… chatbots.

There’s a saying that if the only tool you have is a hammer, you tend to see every problem as a nail. Substitute “chatbot” for “hammer” in the current tech ecosystem.

Am I right? Am I wrong?

It’s hard to look at a newsfeed without seeing examples of chatbots getting things totally wrong - both humorous & concerning. Right now the internet is abuzz with the advice given by Google’s Gemini - including putting glue on pizza, eating rocks & that doctors recommend smoking during pregnancy.

The core of this issue is that general knowledge chatbots have been trained on the internet & the internet is famously full of misinformation. Not only is the internet full of misinformation it is also full of USA-centric content. This means chatbots often display a USA-centric worldview. For my US readers, a good example is the chatbot's insistence on using the imperial measurement system - which is used by three countries (USA, Libya, Myanmar) out of almost 200 countries. For over 7.5 billion people using a “fluid ounce” for measurement makes as much sense as using a “hogshead” or “furlong” - yet, here we are.

My God! What have I done?

But wait - there’s more! As of mid 2024 Chatbots struggle with maths & logic:

For example, I asked ChatGPT-4o & Gemini this question: “What is the 19th letter in this sentence? What is the 123rd letter in this sentence?” - a question a young child could answer.

ChatGPT thought the 19th letter was “e” - which is true if you count the spaces & numbers as letters. It then went on to tell me that the 123rd letter of the second sentence was also “e”.

Gemini told me the 19th letter in the sentence “The quick brown fox jumps over the lazy dog” is “r” - which definitely isn’t the sentence I gave it. But correctly identified the second sentence was too short to have a 123rd letter.

I use ChatGPT, Co-pilot & Gemini daily, but often, it feels like talking to an idiot savant with all the knowledge but none of the understanding.

Same as it ever was

Chatbot reliability aside, there’s another issue that I foreshadowed at the beginning of this article.

C:\> We’re back using command lines to interact with computers.

We already know that people find GUIs far easier to use, yet we’ve just reinvented the command line - except this time with natural language.

The tech industry's experiment is “how much value do you need to add before people will either use a command line or talk to their devices” - at this point, my guess is “a lot.”

How do I work this?

Chatbots definitely have their uses. Especially around getting advice or information on a specialist subject.

In this scenario, the is built using a large language model to understand human language, with a strongly structured information taxonomy for the actual knowledge.

The key to reliable chatbots is well-structured knowledge from a verified source. Much as some companies would like to scrape the internet & call it knowledge we’re better off looking to academics, researchers, subject experts, analysts & librarians to help build knowledge tools for the future.

As a side note, we can still use the technology underpinning chatbots to build AI-powered interfaces that aren’t chatbots.

Into the blue again

Generative AI has opened up another exciting chapter in our tech journey. But this new chapter has been over-hyped & the current over-reliance on command-line chatbots is distracting from opportunities to develop improved AI-power interfaces. There is an underlying issue of what the AI is being trained on & how we feed knowledge into the AI models so they can deliver wisdom instead of glue on pizzas.