High-level conversations about AI ‘aren’t going to result in true productivity’

“We’re talking about something which is resembling an Industrial Revolution, as opposed to another new, exciting, innovative technology — it is way more profound than that.”

So says Jas Jaaj, managing partner of AI and data at Deloitte Canada, in talking about the impact of generative artificial intelligence (AI).

It was roughly 18 months ago that OpenAI launched a free preview of ChatGPT, a new AI chatbot, and as other platforms follow suit, the world is still figuring out the implications.

But for Canada and its employers, now is the time to act if they haven’t already, says Jaaj.

“Pausing and not doing anything is the one of the worst things you can do at this stage.”

It’s a valid concern considering only 15 per cent of Canadian organizations are using AI, while 19 per cent say they plan to adopt the technology in the next few years, he says.

“These numbers are way too low for us to be able to have the type of impact we need to increase the productivity of the nation.

“Considering the velocity with which this space is moving, and the impact it is having in every major industry, in every geography in the world, it's imperative that Canada keep up and play a leadership role in this space.”

To do that, employers must have a strategy in place on how to scale the deployment of AI solutions across the enterprise, says Jaaj.

But a big barrier to AI adoption is employee hesitation.

Employee concerns about AI

A recent survey of over 300 risk and compliance professionals worldwide by Riskonnect found that the top generative AI concerns include:

• data privacy and cyber issues (65%)
• employees making decisions based on inaccurate information (60%)
• employee misuse and ethical risks (55%)
• copyright and intellectual property risks (34%).

Similarly, a recent survey by Deloitte in Australia found that 87 per cent of employees are concerned about gen AI making factual errors — up from the 73% last year. Other concerns for employees include:

• misuse of personal, confidential, or sensitive information (89%)
• legal risk and copyright infringement (84%)
• lack of accountability (84%).

In the healthcare setting, privacy is a big concern when it comes to using generative AI, according to Muhammad Mamdani, vice president - data science and advanced analytics, at Unity Health Toronto.

“Putting in sensitive information… that may actually be then relayed up to a cloud environment, where we don't know exactly what the privacy considerations are, can be a bit concerning and problematic.”

The second big concern is hallucinations or confabulation, when a generative AI model provides a coherent answer with complete confidence “that is wholly or partially invalid,” according to Deloitte.

The tech can make things up in a very convincing manner, says Mamdani, who is also a professor at the University of Toronto, so “unless you actually do your due diligence, you may end up believing or trusting something that shouldn't be trusted… And especially in healthcare, when you're dealing with some critical decisions, that may not be ideal.”

Building trust in generative AI

When it comes to scaling the deployment of AI across an organization, there are two primary barriers, says Jaaj. And these involve not building trust and not engaging the workforce in the process.

For the former, it’s about two dimensions, he says.

“One is the trust in the technology itself in terms of being confident around the accuracy of the results that are being generated. And, secondly, the trust that the workforce actually needs in order to make sure that they are secure and confident about their role in, say, this new workforce of the future.

“So, when leaders are not proactively putting a strategy in place, which clearly has mitigation controls, to manage the risk of the hallucination that comes from AI solutions, that is a barrier for adoption.”

And there are newer approaches that can minimize any errors, says Mamdani, such as RAGs or retrieval-augmented generation.

“They're basically.... machine learning algorithms that then challenge falsities to minimize confabulations and hallucinations.”

But then people also make things up, so what's an acceptable level of confabulation or hallucination?

“To say that it has to be zero, I think is unreasonable, I don't know if it'll ever get to that point,” he says.

Change management: Engaging employees in new tech

If you're not engaging the workforce in this process — to convey the vision and the way in which AI is going to be infused in an organization to augment and amplify the workforce experience — it will mean a much slower pace of adoption, in order to realize those gains, says Jaaj.

“You have to be able to develop this talent strategy and evolve the workforce by clearly articulating, through your strategy, how humans and AI will work together, symbiotically, to complete tasks and activities that different job functions require, to define what the future of work will be in your organization.

“And we feel that there's more work to be done there in organizations considering in Canada, not many organizations have thoughtfully developed their talent strategy in this area.”

A whole change management strategy must be put in place with different components that include clear communication, in terms of what the vision is, and training, so hands-on learning to gain experience in using the technology, he says.

Thirdly, employers should use creative strategies, such as hackathons, so employees can come up with their own ideas about using the technology in their respective scenarios and workflows, to make them more productive, says Jaaj.

“Until that thoughtful process is activated, the high-level conversations are not going to result in true productivity. You have to be able to go all the way down to where the tasks and activities are being executed in the enterprise, and then use that as a vehicle to be able to unlock the productivity and essentially bring value into the organization by the investments that will be made in this area.”

It’s all about education, so people become more comfortable in understanding what this tech does well and what it doesn’t do well — and giving people an incentive, says Mamdani.

“And [it’s about] identifying use cases that will provide value to people… to say, ‘If you use this, here's the amount of time you're going to save’ or practically ‘This is what it can do for you…  these are the concerns or limitations when you're using it.’”

Overcoming worries about legal risks

As seen in the surveys, many employees are concerned about issues such as legal risks, copyright infringement and employee misuse and ethical risks.

There are two ways to allay these concerns, says Jaaj. One is to clearly understand the indemnity offered by the model providers, such as OpenAI or Google.

“Some of them have indemnity as a part of their offering, in that if there is any lawsuit that emanates from a copyright infringement perspective, the customers will be covered, is number one, so you have to be very thoughtful in terms of what are you really buying and what type of indemnity is associated with the license.”

Secondly, employers should be interrogating, to the best of their ability, the sources of the data used to “train” these large language models, he says.

But one of the best ways to take advantage of these generative AI tools is to get beyond the “out-of-the-box base version” and create an enterprise version, says Jaaj.

“By fine-tuning these models with your own data so that they start to understand your organization better — your policies, your values that you want the models to be able to take into account when they're giving you responses — that is where organizations can be very particular about exactly what data sources went into the fine-tuning process, to make them what we call private, large language models for the enterprise.”

Taking these steps will address this risk and demonstrate diligence around how you are inputting the mitigation controls from a legal risk perspective, he says.

Getting the word out about benefits of new tech

When it comes to AI, this needs to be a “team sport,” says Jaaj, and the entire C-suite needs to be engaged. So the communication around the deployment and benefits of the new tech needs to come from the very top, he says.

“The CEO needs to be putting the messaging out so that everyone in the organization understands that this is something which is a priority, it is strategic, and ‘We're going to use it for all the benefits and positive outcomes that it can generate while making sure we're mitigating against the negative consequences.’

“And then every C-suite leader needs to do their part to amplify it in their respective areas.”’

Ultimately, the risks and the areas that still need to be flushed out are very important to focus on — but the way to do that is not by waiting, but by experimentation, says Jaaj.

“And you don't have to do it alone. That's a very important point to take away. We have a thriving AI ecosystem in Canada. There are organizations — all the way from academic institutions to professional services organizations to software companies to startups — that are doing a lot of very interesting and valuable things,” he says.

“So, organizations need to curate those partnerships to be able to gain that confidence and address the anxiety if they're feeling it, to be able to move forward with confidence.”