2022 was the year of AI. From ChatGPT to Midjourney, it seems like whatever you’re doing, someone is working on an AI to do it better.
On the other hand, while we’re seeing really cool applications in the consumer space, it seems the business world is lagging a bit behind. It’s not like there’s absolutely no AI in B2B, but we’re still not seeing meaningful AI features in business applications.
Why is that? In one word - Control.
The thing with AI
Artificial Intelligence can do wonderful things - we know it can generate remarkable texts and images on command, but it can do a whole lot more. AI is great at recommendations, categorizing items, making predictions… tons of stuff. It’s used in anything from creating personalized playlists on Spotify, to detecting cancer.
But there is one major drawback: AI, by definition, is not 100% accurate. It makes mistakes, and when it does, they’re often way, way off the mark. ChatGPT can sometimes make up fake academic references (and be super-confident doing so), or make face recognition errors (try Googling “Chihuahua or Muffin”, and take a long break to enjoy the results).
The thing is, businesses are not necessarily known for their sense of humor, nor do they look kindly at errors in their data. But that doesn’t mean they can’t enjoy the benefits of AI. It just needs to be done a bit differently.
Putting the user in the driver’s seat
In consumer apps, AI blunders rarely have serious consequences. Yeah, there was that one time Netflix recommended I watch Adam Sandler’s “The Ridiculous 6”, but that only had mild mental repercussions.
In B2B, an error can result in huge financial losses, damage to the company’s reputation and legal actions being taken by customers. A few years ago, I was working on a legal analysis product for lawyers. Imagine what would have happened had we presented them false information with that ChatGPT confidence. They wouldn’t be very happy about that.
These days, I am working on a creative asset management platform called Tagbox.io. We often get asked “why not just use Google Photos”? And while there are many reasons, the main one is still giving users control of their data.
You see, if you’re looking for images of yourself at the beach with a friend, it’s totally fine if the app finds 22 images out of the 27 photos you actually have. In a business context, it means you just lost 5 images. Images that might have been shot as part of a campaign that cost your company thousands of dollars. And now you don’t know where they are.
Or think, for example, of a fashion company who’s having a sale on all dresses. The designer goes to the system and finds a great image of a dress to put on the event’s cover, later to discover that it was actually a skirt that matched the top a little too well.
To bridge that gap, we decided to include a review process as part of the upload flow - showing you what the AI found, and allowing you to change it to make sure everything is 100% accurate. It’s a delicate balance - make the review process too complex or time-consuming, and no one would do it. Hide it or nudge the users too softly to do it, and they’ll just skip and move on to other, more important-looking tasks.
Giving the users more control in Wordtune
I loved seeing the release of Wordtune Spices a few weeks ago. Generative AI has huge potential, but too often people using it to create full posts or articles end up with something that is run-of-the-mill in some cases, and actively false or misleading in others.
AI21 understood the delicate balance between control and continuous flow. Have the machine complete the entire piece in one mouse click, the piece is likely to be boring, and probably also too difficult to edit. Have the completion be too frequent, say one sentence at a time (like the sentence completion feature in Gmail), and you’re just not getting the full value of what the AI is capable of.
Being able to instantly create a paragraph, and controlling the purpose and tone of voice of each one, gives the user just the right amount of control. It also makes editing easier, since it breaks the editing into chunks, so you never stray too far off course before you have a chance to adjust.
Being able to add facts and statistics is another key element in making sure the user understands where the text is coming from, allowing them to feel sure that the work they’re producing is worthy of being published.
Since 2022 was the “rise of the machines” year, there were lots of discussions on what AI is actually doing - is it helping us, or replacing us. Giving users control is taking a stand on the matter, and saying “we’re here to help you achieve your goals easier, faster, and better”.
While ChatGPT is all about asking the machine to write an article, Wordtune is about asking it to help you write your article, the way you wanted it to be written, but found it too difficult or time-consuming to do.
While Google Photos is about asking AI to automatically organize your media, Tagbox.io is about giving it a strategy for organization, and asking it to do the manual labor. More control, less grunt work.
As more developers reach this balance of human-controlled AI, and as we get more used to the AI’s statistical characteristics and learn to treat its output as “mostly reliable recommendations” instead of absolute truths, I hope to see it reaching other business applications, and helping us do better, more meaningful work.
Guy Barner is the CEO of Tagbox.io
This article was co-written with Wordtune. Wordtune didn’t write the whole piece. Instead, it contributed ideas, suggested rephrasing alternatives, maintained consistency in tone, and of course - made the process much more fun for the writer.