Thank you for being so accommodating with the timing.
I'm an economist in the future of work unit at the OECD. I'm going to use my five minutes to describe, first, what I think makes AI different from previous technologies; second, what impact AI is already having on the labour market; and third, where policy makers should really be focusing their efforts.
First, on what makes AI different from previous technologies, from a labour market perspective, I think we can all agree that the sheer speed and scale of progress is quite interesting. Because AI can essentially learn and iterate, and because it has applications in practically every industry and practically every occupation, I think that in 20 or 30 years, AI will be so deeply embedded in our society and in our work that it will be difficult to imagine life or work before it. We can think of it in the same league as technologies like the Internet or electricity. Unlike previous technologies, AI can perform non-routine cognitive tasks, which means that many high-skill occupations, for instance engineers and scientists, are particularly exposed to AI. These are jobs that have been traditionally more sheltered from automation. They did things that technology couldn't.
I'm not saying that these occupations will disappear, but certainly I think these occupations will be transformed by AI. That's something interesting about this technology.
Of particular interest about AI as well are the many applications it has in hiring and management. This brings new opportunities but also new challenges to the work environment.
I'll now move to the impact on the labour market, which the OECD has been assessing through its own data-collection exercises. We tend to see things through the framework of job quantity, job quality and inclusiveness.
In terms of job quantity, we don’t really see big signs that AI has impacted aggregate employment, at least not so far. When economists have done empirical studies looking at aggregate employment statistics, there's not really any strong evidence of mass displacement due to AI.
In a survey that the OECD conducted last year, which Canada actually participated in, over half of the firms that use AI—that we talked to—told us that it had had no impact on employment in their firms. Among those who reported that there was a change, they were relatively evenly split between those saying that AI had increased employment and those saying that AI had decreased employment.
Why is it that AI doesn't seem to have had a massive effect on employment, or hasn't reduced employment? Firms told us that AI mostly tends to automate tasks, rather than jobs. They tell us that the AI just isn’t quite there yet.
Where AI does automate a job, firms said that they tend to manage this through reallocating workers to other business areas, through relying on slowing hiring, and through attrition and retirement.
Taking all of this together with the fact that employment levels are currently high in most OECD countries, and given what we expect to see in terms of an aging population in many OECD countries as well in the next couple of decades, I don't think that we are so concerned that AI is leading to the end of work. However, I think there is certainly a lot of potential for disruption as workers have to adapt to changing skill needs.
The same OECD survey found positive results for job quality, but also some risks. Workers who use AI were overwhelmingly positive about its impact on, for example, job satisfaction and health and safety. A large part of this, I think, is that AI tends to automate a lot of dangerous and tedious tasks. Also, workers told us that they appreciated when AI assisted them in decision-making as well.
At the same time, most workers who use AI said that AI increased the pace at which they work. Now, this could be AI enhancing their productivity, which I guess would be, in a way, logical. At the same time, we know that increased work intensity can also induce psychosocial risks such as increased stress and anxiety.
Many workers also expressed concern that data collection in the workplace could infringe upon their privacy and lead to decisions biased against them. Many workers supported banning or restricting the use of AI in processes around the hiring and firing of workers.
Then there are some implications for inclusiveness too. Even if it is the case that the highly skilled workers are more exposed to AI than they were to technologies of the past, I think it still remains a big concern whether those with lower skills have the ability and the resources to adapt.
These people might be in more precarious positions. They may have less bargaining power, and they may find it more difficult to re-skill or upskill.
When the OECD—