Why AI will have the faults of human intelligence

Artificial intelligence (AI) has come a long way over the past decade, from early chatbots to today’s virtual assistants that integrate with multiple applications to “learn” about us. 

However, as impressive as modern AI systems seem, they still have a long way to go before achieving true human-level intelligence. This article chronicles my experiences building various AI tools to assist with business functions and looks at what’s in store for marketers.

AI’s long road from fiction to reality

In the movie “2001: A Space Odyssey,” the astronauts come to the frightening realization that Hal, the AI supercomputer, runs every aspect of the space station. 

A discrepancy between the ground computer and Hal begins the process of the astronauts believing that Hal may be going rogue. When they ask Hal why there may be a difference between the two computer systems, Hal responds, “It can only be attributable to human error. This sort of thing has cropped up before, and it has always been due to human error.”

Scaling the use of AI

Our first experience with AI began in late 2019 when we began experimenting with an IBM Watson application. We used an AI tool to profile buyers’ personalities to help a client narrow down the 17 personas that came from the product marketing group to 4 actionable personas. 

From there, we built a business on the tool, which we call personality-based marketing. The enabling AI tools we use, Crystal Knows and xiQ, can quickly determine an individual’s personality using DISC segmentation. 

Our second experience was right when ChatGPT 4 was launched and we added Jasper to help us with content and image generation. We used the tools to help generate content for websites and images to give our creatives a head start. 

Honestly, we have had mixed results. The tool being prompt-driven, made it challenging to get the output we desired. And given everything else going on in our business, we really didn’t have the time to continue to learn and perfect the prompts. Plus, each new release of the tool brought a whole new set of best practices on prompting.

This led us to the third round of experimentation with an AI platform called Cassidy. The new tool is potentially a game changer for us and possibly the beginning of a modern-day Hal. Unlike previous tools, Cassidy will learn about us on its own. 

It has read our website and is integrated into our G-Suite, Chrome browser and Slack application. It will understand and communicate in our brand voice, act as an assistant, and execute its own workflows. This can be a true game changer for an organization like ours, a fast, efficient agency with little to no overhead. 

Dig deeper: A deep dive into how marketers use MarTechBot

The promise and pitfalls of ChatGPT

The future of AI is bright, but it’s also going to be bumpy. We know there will be shortcomings in this new wave of technology. For example, based on what we have now observed using the first-generation tools, we know that biases exist in the tools. 

The second generation taught us that the output is only as good as the input. The age-old saying “garbage in, garbage out” is still true despite how smart the technology may be. 

Many believe that ChatGPT is getting worse, not better, as it becomes more widely distributed. A discussion thread on the OpenAI Developer Forum from November entitled “ChatGPT is Getting Worse and Worse Every Day” includes 182 comments, all of them in agreement that things are getting worse with each release. 

Developers point to increasing error rates, low retention of previous commands and outputs, and a general lack of support. They even go so far as to say that ChatGPT 3.5 performed better than the most current release of 4.0. So we can, for now, put aside our concerns that Hal will be taking control of the ship. 

Dig deeper: How to fight bias in your AI models

The road ahead

We have built business development and project manager assistants with a knowledge base created by integrating our proposal and project drives. These AI assistants will be able to help our team increase efficiency and, one day, automate the proposal writing process. 

It’s also integrated into my Chrome browser, so it reads my email, attends meetings and archives and indexes the websites I visit. Perhaps someday soon, it will begin drafting and responding to my emails — on its own.  

Until that day, it will require our time to shape it into what we want (or need) it to be. And that’s the point: we will make AI into what it will be, and many have done so already. And, even though we have moved from “machine learning” to machines that learn, it is a good reminder that “artificial intelligence” is actually better described as “artificial human intelligence.” 

A technology built by humans and trained on what humans have made will not be perfect — just like its creators.