They define generative AI, talk about how accountants can use it, and go over some of the ethical issues with using AI at work.

A basic explanation of generative AI

So, what’s an easy-to-understand definition of generative AI? Consider this tongue-in-cheek exercise: Explain generative AI to your grandma.

“You can think of it kind of like Google searching, except instead of just giving you a page of results, it's generating a result based off all that content that it would have learned,” Logan starts.

Dave adds more to that thought. “It's like [a] talking Google. It seems really smart. Sometimes it's totally wrong, but it doesn't know what it thinks isn’t right. And sometimes it can be pretty slow.”

“The really important feature here is that, hopefully, [people using AI] know how to do their jobs, and they're catching the mistakes,” Dave says.

Clearly, there is still a crucial human element in working with AI. While it can do a lot, the work has to be checked for accuracy.

Entering the chat

AI is officially out there and has demonstrated itself to be helpful. But how can it actually be used now?

For entry-level ChatGPT use, try asking it to write a client email. It’s easy enough to then check for quality and doesn’t involve toying with numbers. Jason offers a video tutorial on how to make the email match your personal tone (along with other useful tips for using ChatGPT as an accountant).

Another accounting application for ChatGPT is getting structured data out of unstructured text.

“It's literally as simple as dragging, copying, and pasting the transactions portion of a bank statement,” Jason says. “It doesn't show the account number, just merchant descriptions, which are the same for everybody. So that's, in my book, anonymized information. You can tell it ‘I want a CSV with this column, this column, this column’, and it will give you structured information back.”

And there’s the potential for more usefulness in transactions. Jason shares that GPT-4 can read images, so it could pull data from a bank statement or scanned receipt.

Accounting software vendors are also dipping their toes in leveraging generative AI. For example, with Karbon AI, you can summarize long email conversations (including internal comments on emails), compose email drafts, provide personalized updates to clients as their jobs progress, prioritize your inbox, and more.

Recommended reading: Can AI put a stop to the year-end crunch? A conversation with Simon Langlois

But what about the ethics of generative AI?

Dave raises two issues regarding AI ethics: bias and misuse.

Currently, when ChatGPT is asked to list groups of people—say, innovators of modern technology—it defaults to a very racially limited group. Machine learning only knows what it’s been taught, and the internet has certain natural biases, whether they’re fair or not.

And just as AI can be used for good, it can just as easily be used for evil. In fact, crime using AI is already flourishing. Some common schemes are phishing scams and voice faking for man-in-the-middle (MITM) attacks.

Logan questions liability and responsibility. Take the case of letting AI write a client email. What if the sender doesn’t catch something rude or inaccurate before it goes out? Who would be responsible for that blunder?

Rachel asks whether firms will have a responsibility to disclose when work is powered by AI. It could damage relationships if someone receives work they think is somewhat subpar, and potentially powered by AI. Ian adds to that idea, saying that companies like Karbon are under a special obligation to make sure an AI-powered feature functions properly. It shouldn’t feel like AI.

Still, Jason doesn’t think the ethical concerns are enough to keep someone from taking advantage of the technology. “It will be used for a lot of bad things, but I don't see that as a reason to not use it for positive things.”

Ian sums it up nicely: “To Jason's point, you should think about AI technology as being your most junior staff member on your team. You wouldn't have one of your junior staff communicate on behalf of the firm, you would always double check it.”

Where to next?

With everything developing so rapidly, Ian asks the panel what people should do next with the wealth of possibilities.

Dave encourages the audience to get familiar with what AI can do. That way, when software powered by AI comes along (and it’s already moving very quickly), it will be easier to evaluate whether it offers a helpful solution. Rachel agrees with just trying it, and accepting what works and letting go of what doesn’t.

“This whole space moves so fast,” Dave says. “It is exhausting to just keep up with everything that happens every day.”

So get in there, play around, and see how AI and GPT-powered technology can make work easier today.

• Karbon hosted a fireside chat with accounting and software industry leaders to discuss the status of ChatGPT and AI in accounting.

• AI has the potential to automate menial tasks for accountants, but it won’t eliminate accountants altogether. There is still a major human element needed in working with AI.

• There are certain ethical concerns with using AI in business, such as disclosure and liability. But this shouldn’t discourage someone from using the tools.