I’ll be joining a panel discussion at the Globant Quanta NXT 2024 conference on August 29.
The topic of discussion is :

“Gear up for AI Impact on Businesses: Present and Future.”

I told them I have “a contrarian opinion” on AI, and they said that’s ok, it’ll make the conversation more interesting.

You can find out more about the conference and register at:
https://more.globant.com/quanta-nxt-2024

Short Term

While AI has definitely been over-hyped, and has not lived up to expectations in terms of actual capabilities or revenue potential (and the recent market dip of NVDA stock reflects that), I’ve been pleasantly surprised with the limited features that do exist.

The combination of Large Language Models (LLMs) with Generative AI and training with existing data (the internet) has provided a new interface to general knowledge that is at least as revolutionary as search engines in the 1990s and the conversational textual interface is possibly the most interesting user experience development since the invention of mouses and windowing systems in the 1980s.

I think there is huge potential here if we treat AI as such, a knowledge base with a conversational user interface. You can see the power and of that — when it’s accurate and has information relevant to the domain. Basic knowledge on a variety of topics is readily available in a way that it never has been before, and some topics (such as computer programming) contain a wealth of information that allow you to explore details and minutia with code samples. It’s a great force multiplier for knowledge workers.

However, the lack of depth and the tendency to hallucinate false details are real problems that need to be addressed. Teaching an AI model to say “I don’t know” would be great, but probably not an easy task given the way generative AI works — and that the source data cannot be fully vetted.

The inaccuracy and imcompleteness of AI, with it’s very limited scope are challenges that many people are becoming more aware of, the more they rely on AI.

It can be frustrating for experts using AI to waste time on incorrect or incomplete response, butthe dependence on AI for those with limited knowledge — unable to spot it’s inaccuracies does more harm than good.

Medium Term

Down the line, over the next few years, I’m actually optimistic to see the practical use of generative AI. Especially for the application of expert systems trained on specific data sets and given rules on specific domains. Logistics, Accounting, and Medical records are fields with huge potential. Every industry and large organization has loads of information locked away in email, CRM, project management, ERP and other systems waiting to be exposed if localized AI models can be trained with their proprietary data.

I look forward to seeing a lot of information become more easily accessible, searchable, and processable through natural language and business logic processing.

Long Term

Long term, I don’t see AGI or any extreme breakthroughs. LLMs and generative AI provide an interesting way to access data, but generating boilerplate text or pointless images or impersonating audio are not going to provide value.

My biggest fear is that AI will become a closed system and used for only trivial things. We used to have an open internet, but now we have walled gardens like Instagram & Tiktok where people post pictures of themselves on vacation or videos of themselves dancing and lip-syncing to bad music.

Much of the promise of computers and the internet has fizzled out and the potential has disappeared or been drowned out by the dross, and I expect the same for AI.

Right now, the biggest advantage AI has over search is that search engines have been gamed by SEO and useful information has been deliberately censored or casually buried beneath corporate advertising. Once the equivalent of SEO takes over AI and AI is used more and more to generate SEO, the value of AI will lessen and the useful knowledge will become even harder to find.

I expect there to be a brief bloom of increasingly accurate expert systems and creative methods for leveraging AI, followed by a dumbing down and closing off of functionality by dominant players, much as we’ve seen with other technolgies in the past.

Last week, I went to Columbus Ohio to give a presentation on “Pair Testing and Other Radical Ideas”.
The conference was great, I meet a lot of nice people, and my talk was well received.

My favorite comment was from an another tester who said my talk kept him awake after lunch.

Here are the slides:

It’s not the same as delivered live, with lots of audience participation and feedback, as well as audio clips from relevant music “back in my day”.