AI may have its benefits, but not for everything, say employment lawyers highlighting the risks
More than three-quarters (78%) of professionals believe that AI is a force for good in their profession.
But when it comes to the question of responsible use of AI, some tasks may be “a step too far for AI use, ethically speaking,” according to the recently released Future of Professionals Report from Thomson Reuters,
The vast majority (90%) of respondents said they believe that allowing AI to represent clients in court or make final decisions on complex legal, tax, risk, fraud and compliance matters would be “a step too far.”
Similarly, when it comes to AI providing advice or strategic recommendations, 54% say using AI would be be unethical, found the survey of 2,205 people in several regions including the U.S., Canada, Australia and New Zealand and Latin America.
On the other hand, using AI to draft basic documents (18% say it’s a step too far), for research and analysis (11%) and for basic admin tasks is much more acceptable.
Source: Thomson Reuters
So, how does that translate when it comes to the HR space?
“Generating docs such as job descriptions, offer letters, training material, those are all very common and easily adopted uses of AI,” says Mary Alice Vuicic, CPO at Thomson Reuters in Toronto.
But they still require a human in the loop, she says.
“AI does hallucinate; the more data you have in, the better the data, the less hallucination, but that can easily be managed with the right oversight.”
Over-reliance on generative AI
The tool can be a huge cost saving measure and promote efficiency, but it can also create a lot of risk if employers aren't keeping an eye on it and have a good understanding of what they're using it for and what's appropriate, says Nicola Watson, a lawyer at Cox & Palmer in Halifax.
“The use of generative AI in most areas in employment isn't a problem in and of itself; it's the exclusive use and reliance on generative AI that becomes a problem,” she says.
“I would never rule it out completely, I think it can be very helpful… sometimes, though, it can cause more headaches than good. But, it's very fact-specific in terms of what your piece of generative AI is being relied on for.”
Using the tech can be a great starting point and a way to “get your bearings, get ideas more efficiently,” says Watson, but in drafting documents or policies, it’s important “to have the human in the loop.”
And it’s also key that the employer understands what their obligations are from a legal perspective, she says.
“This applies not only to what their obligations are in the employment law context, but also in the changing regulatory context of AI generally, because we're seeing this constant shift.”
Things like employment agreements and human resources policies are fundamentally still legal documents, says Christopher Sinal, a partner at Siskinds in London, Ont.
And if generative AI engages in a hallucination — by providing incorrect or misleading information as if it were a fact — or creates something that just is wrong on the law, he says, “the downside risk is significant enough that the juice likely isn't worth the squeeze in the resources and time that you save in using generative AI to begin with.”
AI only as good as data going in
When it comes to the factors that are vital for defining the responsible use of AI, “compulsory review of outputs by a human professional” and “transparency in which data source responses have been drawn from” were in the top five, found Thomson Reuters.
Source: Thomson Reuters
In many ways, generative AI is not designed to be a research tool, says Watson.
“And depending on what program you're using, AI’s answer is only going to be as good as the data that it's learned from. And it can pick up trends or it can often provide answers that it thinks you want to hear, rather than what's actually true.”
For example, if HR asks ChatGPT to write a termination clause, the tool is going to use the aggregate of what it finds on the internet, says Sinal.
“And a lot of the contents that will be on the internet about Ontario termination clauses will be discussing the ones that were wrong. And so ChatGPT can innocently pull from all of these data sources and build you the worst possible employment or termination clause you could want.”
Riskier areas of generative AI
As far as concerns around AI go, respondents to Thomson Reuters’ survey most-cited worry was that people will rely too much on AI-powered technology rather than using their own professional judgment, followed by fears of AI being used for malicious or fraudulent purposes.
Source: Thomson Reuters
“We have responsibility, and it’s a collective responsibility, to ensure that the responsible and ethical adoption of AI, that there’s appropriate education,” says Vuicic.
“[It’s about] having open dialogue about what are the risks, and how to mitigate the risks, and then having cross-industry collaboration, intergovernmental collaboration on ‘How do we collectively solve this and move forward in a responsible way?’”
Some key areas that require “a high amount of caution” — if not avoidance if you're not going to be very cautious and have a good understanding of your AI programs — are things like final decision-making and employment status issues, such as hiring, firing or leaves, says Watson.
“That can be a very risky area, to allow generative AI to be making decisions for you… especially in the recruitment world, which we've seen quite a bit of talk about, on a more broader basis, relying on AI programs that might have inherent discriminatory measures that they've built up from the data that they've learned from.”
Another area of concern is drafting, she says, with issues around privacy, for example, depending on what information you're feeding into the AI programs, and whether that's appropriate, along with concerns about accuracy.
There can be employment law or legislative differences by province or by employee, says Watson, so “using generative AI for drafting is great for an initial first draft but that process shouldn’t be completely relegated to AI.”
Those risks also can be found when it comes to using chatbots for HR information such as leave requests, harassment complaints or hiring concerns, she says.
“If this chatbot starts asking… an individual about things that are protected grounds — such as disabilities, such as age — that could end up putting the employer at risk… which an HR professional would be well aware of.”
The same is true for job descriptions, says Watson, because of “anything that could be seen as discriminatory or suggest a lack of an ability to accommodate and things like that, especially with individuals with disabilities.”
Best practices for combatting risks of AI
When it comes to using generative AI for things like drafting or HR letters, the key is keeping a human in the loop, says Watson, “and having that human feel confident in what they're doing, and also that they are able to override or make judgment calls.”
“Because sometimes what you'll see with employees, whether they're management or HR or whatever, is that when an AI provides suggestions or predictions, they might not feel that they have the ability to truly override them. And you want to avoid having just a rubber stamping of things.”
Sinal views generative AI less as a tool to make documents and more as a “thought partner” in coming up with ideas for wording or putting together a policy, for example.
“I find AI is most useful there than it is just necessarily [asking it to] ‘Make me a document’ because, inevitably, it's going to end up being a first draft,” he says. “I'm going to use it as almost another member of my team to help me develop ideas. I think that's absolutely a good use of AI.”
One of the most challenging things about using AI as a tool is less what we're using it for, and more how to use it properly, says Sinal.
“There's an entire field around how to develop the appropriate prompts, how to actually speak to AI in a way that it will give you the types of results that you actually are working with. And it almost becomes like arcane magic spells, trying to get the correct product out of AI. And so that really is where I think the skill in using AI comes from.”
Generative AI may be better than a random person in writing a vacation pay policy, he says, “but it is not the equivalent of going to an HR professional that has dozens of years of experience saying, ‘Write me a vacation day policy.’
“So, I think as long as you're aware that you need to be circumspect in in the work product that you get, and treat it as a first draft ready for you to edit, I think it can be a great launching point for fantastic work.”
Transparency, communications
At Thomson Reuters, to best adopt and integrate AI, it’s about transparency and communicating openly among colleagues, along with giving hands-on experience with the new technology, says Vuicic.
“We are continuing to update our compliance, our standards and ethics around AI, and we’re giving people the opportunity to co-create new processes, different ways of working products in leveraging AI.”
It’s also important to make sure silos don’t rise up as different groups or departments explore the possibilities of AI. To that end, Thomson Reuters launched an enterprise AI adoption initiative, which Vuicic leads along with the head of technology and operations.
“We’re elevating these use cases so people can benefit from it, as well as the change management associated with adopting the technology using the new practices, sharing examples of creative ways to get people to use the technology, so hackathons, contests, and then we’re also using it to prioritize our investments… to get the greatest return for individuals, for the business.”
