n a short time, AI has gone from creating freaky-fingered phony photos to a serious business tool. With that rapid change has come equal parts hype, critique, and alarm. To help small businesses leverage AI’s potential while avoiding the perils of security breaches and “slop,” Alaskan experts have been staying abreast of the march of progress to help businesses stay informed.

“Artificial intelligence is a game changer for a lot of small businesses in Alaska,” says Carlos Machuca, director of the AI resource program at the Alaska Small Business Development Center (SBDC).

While Machuca has been leading nonprofit consultations, the private sector is keeping apace with technology too. “We ourselves are small and growing. We’ve found it to be indispensable,” says Kyle Easterly, co-founder and CEO of the “AI deployment service” Delve Group in Anchorage.

Machuca says even businesses that seem low-tech—he cites the example of a landscaping contractor—could still benefit from AI-driven tools. However offline or hands-on a company’s services, that business or sole proprietor still must deal with tasks such as payroll, financial reports, or creating business policies. AI could help save time with such administrative tasks, Machuca says.

From Google to Adobe and Canva, many longstanding business software providers increasingly incorporate AI functionality. Most companies seeking to use AI in their business, however, will want to choose one of the four main standalone apps: OpenAI’s ChatGPT, Anthropic’s Claude, Microsoft’s Copilot, or Google’s Gemini.

“Stick with the big stuff,” says Matt Weber, owner of Kenai-based Weber and Co., a digital marketing agency that does technology consulting for AI. “I wouldn’t go chasing like these boutique small operations or people who are building layers on top of the big guys,” he adds. “The big guys are eventually going to win.”

The favorite of these experts is Claude, which Machuca thinks is especially good for working with numbers or writing code. Similarly, Easterly says both Claude and ChatGPT are “the most well-resourced” and have help manuals. “There’s a lot of resources out there about them.” Between the two apps, Claude is optimized for handling massive files while retaining context, whereas ChatGPT is designed for speed and versatility, trained by human reviewers to prioritize helpfulness.

Machuca compares ChatGPT to a Swiss Army knife and says you can use it for almost any business. Easterly says this AI can also generate images, like flow charts, which can be very useful.

Gemini and Copilot got more mixed reviews. Easterly calls Gemini a “distant third,” but Machuca’s a big fan of its Notebook LM. He says this tool “grounds your data,” allowing one to work only from that material. Machuca also uses Gemini’s podcast feature to create audio conversations about articles he wants to read, which he listens to during workouts.

Whatever tool one lands on, choosing a paid plan unlocks better tools while giving a business some protections. “I meet all kinds of people that use the free one, and they have all this trouble with it,” Easterly says. “It’s not the same product at all” as the paid versions.

Most paid or “pro” versions start around $20 a month. For help getting started, the Alaska SBDC offers lots of free resources, Machuca says. To get personalized advice, small businesses can sign up for a business advisor. The SBDC website also offers a mix of paid and free videos on topics such as marketing, financial projections, data analysis, and how to create an AI assistant.

To start using AI, Easterly recommends businesses begin in the background. “I would very, very strongly encourage people to not start with public-facing use of AI,” he says. “The last thing I would do is jump right into writing all of my social media content, producing images for it.”

Instead, he suggests starting by using AI to write an employee manual or to document certain business processes. “The analogy is, Hollywood is using a ton of AI right now, but you don’t see it on the screen,” he says. “They’re using it to hone down to a better finished product.”

Thoughtful, effective use of AI requires understanding its strengths and weaknesses, and understanding one’s own. “AI works best if you already have a deep understanding of your topic and what you want to do,” says Jennifer Carroll, a UAF associate professor in the Department of Alaska Native Studies and Rural Development, who’s led multiple initiatives on AI and curriculum. “You have to be able to catch all the different little things that aren’t going to go right. If you have a solid foundation, then AI can help you organize, brainstorm, and expand your learning. But if you use it as a crutch or as a shortcut, you’re vulnerable.”

Weber says AI tools are especially good at “simple, repeatable things or data ingestion.” As an example, he cites analysis of the 500 leads he got in a recent month. Rather than create an Excel script to go through them all, he could have an AI tool go through the leads list.

AI is also “very good at crossing multiple sets of data” and compiling it, Weber says. This includes compiling information spread across multiple software programs and systems. Easterly says one of the things AI is really good at is dealing with “organizational scar tissue.” By that he means exporting, importing, or compiling things from multiple systems. He gives the example: “I’m going to pull things out of these four systems and put it into Excel.”

Machuca says this cross-platform compilation could help a small company create the kind of business plan that banks often require to get a loan. An entrepreneur might have a bunch of financial documents from which to pull several different figures. Rather than spend hours combing through the documents to pull the figures needed, a business owner could tell their AI tool to abstract the data needed for the business plan.

AI has also allowed SBDC to offer services it didn’t have before, such as industry-specific templates for standard operating procedures. “Now, because of artificial intelligence, we can hyper-personalize,” Machuca says. “We can actually create templates for every industry that you can imagine,” such as a dog-grooming business.

Both Carroll and Easterly also like using AI as a search engine replacement. “The quality level now is… 100 times better probably, at least,” Easterly says, compared to hallucination-prone web searches when ChatGPT splashed onto the scene in 2022.

Tools like Claude or ChatGPT still have significant limits. AI tools struggle with tasks that require uniquely human traits such as creativity, imagination, or judgment.

“AI is built to give you an answer confidently, so it will tell you things with confidence that may or may not be true,” Weber says. For example, if one asks AI for a contract, it might cause a so-called “hallucination,” he says, where “it goes and invents an email that never existed and says, ‘Here’s the contract.’”

When “you’re asking it to bring things to the table” rather than to find or synthesize existing information, “that’s where you’ve got it to be really careful,” Easterly says.

Machuca agrees. “The way imagination works on AI is just [to] recycle a lot of ideas that are already out there,” he says. He and Carroll also note the models’ ignorance of Alaska life and culture, especially in rural contexts.

“The material that these systems have been trained on is not Alaskan material, is not Northern material,” says Carroll. “The solutions that it gives you, the answers it gives you… are geared toward a Lower 48 population.” For example, she says an Alaska Native image-generation request might yield something that uses “clothing from some well-known tribe of the Lower 48.”

Machuca says AI also might not understand the seasonality of certain industries or activities, like what time of year Alaskans might be fishing.

“People should understand that it’s trained on… what is common, right, within our society,” Carroll says. “Alaska is not common; Alaska Natives are not common… It’s not built to use our voice. I think that’s something that any of the… businesses that are looking to use it should be mindful of.”

Effective use of AI requires important safeguards such as employee training, data encryption, and clear policies. Machuca says businesses should ensure that employees understand what tools the company allows, as well as AI’s capabilities and limitations and the risks of improper use.

Depending on the setup, some AI models can access internal or personal documents, such as Google. “That could be a layer of risk if you don’t have the right education on how to access those documents,” Machuca says.

Security risks are “tricky,” Weber says. For one client, his firm resolved the document-access challenge by creating an account for their Google admin. By controlling what folders and Slack channels that account could access, the client limited what Claude could access. A cleaner solution might exist, but sometimes those take IT or other resources a small business doesn’t have. “Security almost becomes a risk analysis that you have to do for yourself every time you connect another tool,” Weber says.

The same applies to adding additional account users. “There’s risk when it’s just you and then there’s risk about… are the other people around me using it right?” Easterly says.

As an example, he cites the settings for whether an AI tool can use your data to help improve the model for everyone. This benign-sounding checkbox can have great consequences for data security, Easterly says. “If you uncheck that box, then it’s actually reasonably safe” to use AI tools. But he says companies should “verify that maybe once a week and make sure you didn’t accidentally put it back. Or, if you’re sharing an account with somebody, they didn’t change it.” Some team plans also offer the ability to control certain settings across the organization.

Businesses that face stricter regulation of their data protections, such as health insurance privacy, should probably get a professional’s help when using AI. “Nobody should be DIYing HIPAA with AI,” Easterly says.

Despite all the buzz around AI, business fundamentals haven’t changed. Weber says he gets lots of calls from people concerned about how AI affects search results. But just like Google before it, Weber says AI “scrapes the internet” for available information: social media posts, a company’s website, what people say about it, and other companies’ links. Search engine optimization feeds AI the same type of results.

“If you do all the same things you would normally do for Google to rank, guess what? They pretty much work for AI,” Weber says. “The best advice I could ever give you… that costs you nothing is, go get a lot of reviews. Make sure you’re engaging with people on social media when they engage with you.… And be interesting.… Make sure people are shouting your name from the rooftops when somebody goes into a forum and says, ‘Hey, who’s the best fishing guide?’”

AI’s growing capabilities also require diligent cybersecurity education, no matter how much or little a person or business uses the tool. “It’s like trying to avoid the internet,” Easterly says. “You can do it for a while, but pretty quickly here… it’s going to be impossible to avoid.”

AI education is everyone’s responsibility because of the power it puts in the hands of users, ethical and otherwise. Machuca says AI’s ability to create deepfake videos can make certain scams more plausible and persuasive. If someone supposedly contacts a business with a request that could put the company at risk, “make sure you call back to make sure that that’s the right person,” he says.

As AI’s use and influence grows, Carroll says keeping a sharp mind will be more important than ever. “The best analogy I’ve seen for this is to working out,” she says. Learning does for the brain what exercise does for muscles. “You need to put your brain under friction, under tension. You need to work it. And that’s how you see gains in your own learning and in your ability to handle material.”

Used effectively, AI becomes one more piece of equipment in the gym.