I’ve been working with artificial intelligence systems since the 1980s. Back then, AI was considered a revolution in leveraging computer systems to achieve unheard-of capabilities.
Things are much the same today with generative AI (genAI). But to avoid the mistakes made during the first generations of AI systems, businesses must understand what AI is valid for and what it is not.
It’s 1988, all over again
Attempting to compare 1980s vintage AI, including Lisp and M1, to today’s machine learning and genAI capabilities is a bit unfair. Back then, AI systems cost many millions of dollars and had much less AI functionality.
However, many mistakes made AI go away, while other, more direct solutions were leveraged for businesses. The most apparent mistake was the misapplication of AI for use cases where AI provided little value.
Even with my teenage brain, I knew that transactional systems, such as sales order entry, were not good fits for AI. Nevertheless, I was given the order to build such things, knowing full well that I was killing an ant with a sledgehammer. An expensive sledgehammer at that.
This is largely why AI fell out of favor for most businesses. Years later, it has now returned as deep learning, machine learning, and machine learning doing generative AI.
While the technology has vastly improved and is way cheaper, I see the same dumb mistakes occurring now. Errors that will not align businesses to get the most value out of genAI could drive pushback in a few years as costly genAI systems built by expensive people genuinely don’t return the expected value.
These are self-inflicted wounds and wholly avoidable if businesses put a bit of thought into strategically using this technology. What are the killer business apps for genAI? What are the good and bad use cases? How can businesses pick the right path? How can we avoid the mistakes of 30 years ago?
To figure out what not to do with genAI, it’s helpful to look at what genAI does well and find the use cases that match these capabilities. Simple enough.
For our purposes here, I’ll pick the top three. There are many other good use cases, so don’t push back on me for only listing three. After all, this is a blog, not a white paper.
Natural language generation
First is natural language generation, or NLG. If you have ever tried to pass off a report, letter, email, or other written content created by ChatGPT, you already know this one.
Businesses can use this capability to generate tremendous value, including providing better customer experiences through personalized communications, either written or through a chatbot.
This will be a job killer, and many customer service positions, for example, will be replaced by NLG automation. However, businesses will benefit by doing a lot more with fewer humans. They can provide better customer experiences that solve problems much quicker.
For example, call a technical support line today, even ones with interactive voice response (IVR) systems, and you’ll find out quickly that your ability to solve your problem depends entirely on the knowledge and communication capabilities of the person on the other end. What if that somebody had the understanding and reasoning of 10,000 experts and could provide a response that much quicker and more useful to you, the customer? Also, what if that interaction could cost the business 20 cents instead of $20?
You can see where this is going. If done correctly, NLG can offer better value and an enhanced customer experience for a lower price. It will displace people, and so we need to consider the ethics. However, I see businesses moving in this direction quickly.
Recommendation systems are the ability for genAI-enabled systems to personalize recommendations in e-commerce, streaming, and content platforms. This is nothing new, and I’ve been working on these well before genAI showed up, but now we can take them to a new level of effectiveness. They have the most ROI for any business that sells things.
Have you ever wondered how an e-commerce site can recommend products to you, even before you’ve provided any information? The older versions of these could increase sales by 20% to 40% just by determining the sex, age, race, hobbies, and occupation of the person using the site and then recommending specific products and services that person would most likely need.
With the advent of genAI, we can achieve a sinister level of effectiveness by communicating with customers using dynamically generated interactions that are very fine-grained. Once the systems figure out that you’re interested in, say, cycling assessors, you’ll see a unique font, subliminal message, color scheme, custom images, and even a specific price point of products, all dynamically targeted at releasing endorphins, putting you in the right mood to drive more sales. Be prepared to be manipulated for the good of the bottom line. Again, ethical questions arise.
Anomaly detection is identifying irregular patterns or outliers in data for applications such as fraud detection or system monitoring. Here, genAI will help us spot data patterns that show trends, explain what those trends mean, and adjust processes to gain the most business value.
This goes beyond the next generation of genAI-based anomaly detection, such as using historical data patterns to find likely banking fraud or predict which systems may be headed for an outage. This is like Minority Report minus Tom Cruz. Your next loan application could be denied due to “pre-crime.” This use case also leads to many ethical questions we need to ponder.
Of course, there are dozens of other solid uses for genAI. The problem is that many businesses won’t consider those but instead jump to situations where genAI will drive cost and risk and generate little or no value. We need to get smart about this stuff quickly before businesses kill themselves through self-inflicted wounds.
Copyright © 2024 IDG Communications, Inc.