‘Super threatening:’ Experts say adopting AI into business might not work how you think - Edmonton Journal

Business leaders in Edmonton heard this week that, in the rush to board the AI hype train, most aren’t leaving the station.

“The biggest misconception right now is that AI is being wildly successful with companies,” said Peter Bishop, the chief innovation officer with Alberta advertising agency ZGM.

He said ZGM did a recent study with market researcher Stone-Olafson on the use of AI in the Canadian marketing industry. It found that while AI was being adopted, many couldn’t prove they were getting any sort of financial benefit or efficiencies they could track.

In that study, only 21 per cent of marketers said AI was having a high impact on their day-to-day work, and only three per cent fully trusted the outputs AI gave them.

Around 43 per cent didn’t know if AI is increasing return on investment at all.

He said those findings followed other similar research, including one study from MIT, which found 95 per cent of companies initiating AI projects were failing.

Bishop joined Edmonton Chamber of Commerce president and CEO Doug Griffiths, Alberta’s technology and innovation minister Nate Glubish, and Trevor Bruintjes from CopperTeams to announce the launch of new education tools geared towards helping Edmonton companies adopt AI effectively.

At the event, Bishop told the audience gathered at the Edmonton Chamber of Commerce that there is a big disconnect between what’s working and what’s not.

Bishop said a novel aspect of AI adoption is who is currently the most excited about it.

For the first time, executives and leadership are jumping into a new technology “head first,” while the younger junior staff are more cautious.

He said that while leadership is excited about using AI to find efficiencies, the younger workers are scared because they’re worried it’s going to replace their jobs.

“There’s this cultural shift that you have to be really aware of when you’re trying to ram AI down the company’s throat,” Bishop said. “When you understand that this is super threatening, unless you have permission from the company to do so, you are really starting with one foot behind the other.”

Trevor Bruintjes said there are some jobs that are probably at risk, but AI will never be able to replace humans. 

“There’s going to be changes in roles, people are going to have to adjust, the winners are going to be the ones that have the AI skills plus the knowledge, the experience, the education,” Bruintjes said.

Bruintjes and Bishop agreed that what AI is good at and going to be the most successful at is automating menial, boring tasks that humans don’t typically want to do, and freeing up time and space for people to do the more creative and fulfilling parts of their job.

The chamber event coincided with the Upper Bound AI conference, one of the largest of its kind in Canada.

There, Canadian tech paragon and president of Mozilla, Mark Surman, said going from a 10 per cent benefit to a tenfold improvement is the nature of figuring out new technologies. But he pointed to how it’s often not the inventors of new technology or the first adapters who get to that level.

Industrialization hit England and Germany before the United States, Surman said, but it was the Americans who rebuilt their entire production from scratch around new industrial technology and became “the great industrial power.”

But Surman, who is an advocate for open-source technology and the open internet, said open-source AI models are a boon that companies can use to tweak very customized solutions.

When it comes to bringing in AI, he said many companies’ first mistake is gravitating towards very expensive “frontier” models like those from OpenAI and Anthropic.

He said a company might test a powerful model like Claude and find new efficiencies.

“And then, after a month, after, you know, they put it in production, they got this huge bill.”

Instead, Surman said it’s possible to fine-tune an open-source model to do the same task, and it would only cost a fraction of the frontier model.

“Open-source models are really good as raw material for fine tuning raw material for simple tasks,” he said.

Surman also recommends using the more powerful tool to help experiment and find the right way AI models can be used for a company, and then tweak a free open-source model to do that task when scaling it.

And often what works best, Surman said, isn’t just all open-source, but a hybrid of the two working in tandem.

Bookmark our website and support our journalism:Don’t miss the news you need to know — add EdmontonJournal.com and EdmontonSun.com to your bookmarks and

You can also support our journalism by becoming a digital subscriber. Subscribers gain unlimited access to The Edmonton Journal, Edmonton Sun, National Post and 13 other Canadian news sites. Support us by subscribing today: The Edmonton Journal | The Edmonton Sun.