AI op de werkvloer: kansen en uitdagingen voor de arbeidsmarkt

Dit artikel is geschreven door SThree in het Engels.

Bhargav Srinivasa Desikan, Senior Research Fellow at the Institute for Public Policy Research joins us to discuss how AI is reshaping STEM skills and jobs, its impact on working models and its potential to fill labour shortages across critical industries.

Listen to the full episode of “Forces of STEM“, by SThree in partnership with FT Longitude.

The world of STEM is evolving fast. Economic, environmental, and technological forces are reshaping skills and sectors, giving today’s business leaders a critical opportunity to reimagine tomorrow’s world of STEM.

Welcome to Forces of STEM. I’m your host, Ben Bschor, Group Editor at FT Longitude. In this episode, we are taking a look at how AI is reshaping the labour force and its impact on working models.

Joining me to discuss this is Bhargav Srinivasa Desikan, Senior Research Fellow at the Institute for Public Policy Research or IPPR for short.  Bhargav, thank you for joining me today. 

Bhargav, could you begin by giving our listeners a brief overview? What is the IPPR and as a senior research fellow, what is your area of research?

IPPR is UK’s leading progressive policy think tank, and we’re working towards a fairer, greener and more prosperous society. So, we are a team of researchers, communicators, and policy experts turning bold ideas into common sense realities. As for me personally, I’m a senior research fellow working at the intersection of AI and economic and social policy.

And you’re the co-author of a recently published report that looks into the impact of AI on the workforce. Before we go into the details of your findings, can you give us the broader context? How common is the use of AI in the workplace today?

Sure. In our paper, we highlight four phases. Phase zero, we call experimentation and platform investment. Phase one, is low hanging fruit implementation cases. Phase two, is integrated AI systems, and phase three, you have processes getting built around AI. Now, in this specific context, I’m talking about generative AI and depending on the firm, of course, AI usage would be in different phases. With generative AI and large language models, I would say we are in phase zero or one. So, experimentation and low hanging fruit.  

Is it possible to put a figure behind this? Can you say, I don’t know, so-and-so common is AI use, or so-and-so many people are exposed to it at the current stage?

Maybe another way of we could think about this is also if you ask CEOs what they think is the immediate impact, and luckily we actually had. There was a survey being done which showed that in terms of near-term impact, immediately in the coming years, only 5% of CEOs of, I believe it was Fortune 500, said that they see AI being used extensively in the workplace. 65% said that they see it being used in the next three to five years.

So you mentioned already what sort of Fortune 500 CEOs think, but what does your research show, is the direction of travel? What pickup will we see over a specific period of time? Might be five years, might be three years. What has your research unearthed?

What we looked at was basically, in the UK, if you have a list of all the kinds of tasks being done by different jobs. So, we got a database of 22,000 tasks across about 500 jobs, and we basically studied how are all of these tasks exposed to AI, right. And what we found was, with here and now, so something like phase zero, about 11% of tasks are currently exposed, which means that they can be done by AI, and going forward, if we build other workflows around AI, up to 60% of tasks can be solved. Tasks and responsibilities, let’s talk, can be taken care of by AI in some form or the other.

We’re not saying this will definitely happen. We’re saying that this is within the realm of possibilities. So, in terms of now getting back to your question on the direction, I would imagine if already we see, you know, our report shows that AI is capable of doing repetitive cognitive tasks quite well, I would imagine it would do these even better. It would be able to deal with audio and video as well, and it would also be able to work as teams of agents; analytical programming, mathematical tasks, will all be able to be done much better.

 So, I mean 59% or 60%, I think, is what you said. That’s quite a massive number. Does it mean the way we all work will be totally disrupted in the next couple of years?

If you look at the exact number, one of our headlines said 8 million jobs at risk in the UK. Now, of course those are definitely some eyebrow-raising numbers. One thing I would like to say before we dissect that number, is that these are all scenarios. Technology is not inevitable. It’s not simply like the big tech firms put out technology and we just accept it and use it, right? And this is where public policy comes into place. This is where legal infrastructure comes into place. This is where democratic, societal institutions come into place. Where we begin to decide, what kind of tasks do we want AI to help us with? How do we want AI to help us with tasks? And how can we make sure that AI benefits all of society and not just the firms making it?

So, I want to put that big, highly important disclaimer that we have some control over this. That being said, yes, I would imagine there would be some significant disruptions in certain sectors and domains. You can think of it also as another breakup, as augmentation and automation. We want AI to augment us, work with us, increase our productivity, and instead of replacing us or completely pushing us out of the way.

I think in your research you identify different types of workers or jobs, and you say some of them are more at risk than others from the increased use of gen AI. Talk us through that. What jobs are most at risk and why?

This is the question, isn’t it? So, when we look at all the tasks and then we just find on average which kinds of jobs have most of their tasks being able to be done in some form by generative AI, we find back office, administrative and secretarial tasks, most at risk. And then on top of that, we also find customer service tasks at a significant amount of risk. This isn’t just in terms of the tasks. In terms of the groups of people, we also find that right now women are most at risk, as they take up a disproportional number of these kinds of jobs. We also find that entry-level jobs are high at risk in this. So, those are the demographics you would imagine would be entry-level tasks, as well as tasks done by women. In terms of the jobs it would be basically these kinds of jobs. Now, another useful way to think of this is categories. We find that organisational and strategic tasks, as well as repetitive, cognitive and analytical tasks are most at risk. And we find that manual and operational tasks, as well as interpersonal and communication tasks are least at risk. 

And then you talk in your research about green jobs and you say those are less exposed to the impact of AI. Why is that? What makes those roles so unique?

We continued in our analysis to basically break up for green jobs, what is the normal distribution of tasks? And we find that on average, green jobs, 40%, as opposed to non-green jobs, this is 28%, 40% of the tasks done in green jobs are resistant to AI because they are manual, operational and technical tasks. You can think of in a more grounded example, if you are building heat pumps to insulate homes or if you are focusing on massive wind turbines, you can imagine AI is at least, especially the generative AI, which we’re talking about, are less likely to come in and scope that. So, we basically just saw, distribute the tasks, and then we just found on average green jobs are just generally more resistant. You can imagine why as the whole world is going through a time where we need to take our green transition extremely seriously, it helps in two cases. One is we can provide jobs that are resistant to AI, and we can also help us on other massive structural problems we’re facing.

One problem that in particular STEM industries face, is labour shortages, actually. There is basically not enough people around to have the right skills to fill all the roles that are out there and available. What does AI mean for these labour shortages? Could that be addressed? Could AI jump in there and replace workers or at least help them so that one worker can do more and maybe it’s easier to fill those roles?

From a research evidence perspective, it showed that especially for programming tasks—now programming is of course one kind of STEM job, but it’s a good way to give an example of how this works—it’s quite a structured task. It’s quite a structured task and can often have predictable outcomes, in terms of ins and outs. So, in this particular situation, large language models end up doing quite well. There’s also quite a few code-specific large language models. In a study that was done, it basically showed that if you are an entry level programmer, using a large language model to help you to assist in coding, helped you get to nearly the same proficiency as a semi-expert or expert programmer. Now, if you’re already an expert programmer, then ChatGPT also still gives you a small boost, but the percentage of that boost is far less than if you were a beginner level programmer.

So this basically means that you suddenly have a lot more capacity and capability. It also means that maybe in the past, when you needed to do a full university undergraduate course at the least, and then have a significant grounding, you could maybe now potentially get away with doing a coding camp or a crash course, for example. So in that case, yeah, it absolutely can. Now, with this comes certain caveats like, you would still want some expert in the room to be able to do a quality analysis. You would still want very, very thorough tests to be written. You wouldn’t want to ever only trust a model to do your job for you. You would really want a human in the loop. And that’s obvious in a lot of different contexts, especially in critical contexts. But certainly, I think in fact, that could be one way where I’m quite optimistic actually about how generative models can help, because you could maybe sense in our report, we’re quite cautious, in general, about how these tools should be used because for example, in healthcare, education, these are crucial societal functions. But in the context of STEM jobs, in the context of say, data analysis, programming, can be quite useful.

In my job as well. It helps me a lot and I benefit a lot in terms of my productivity. If my firm might need to hire two computer scientists right now, they can get away with just me. So, in this case, I think there is a very direct labour shortage solution. We still need more people who are able to understand and use these tools, and it won’t completely automate and come in the way. Basically beginners can now work as a semi-expert, or of course, all the caveats, and ifs and buts that come with that.

In a way, it’s not only knowing how to use it exactly, but it’s about understanding in what areas AI might be able to help me. If I can add from my own perspective, I just started last week playing around with using AI to analyse spreadsheets and it’s for me a steep learning curve and extremely interesting.

Final question, Bhargav. How can companies support the workforce during this period of change? Is there something specific they should do?

Yeah. At a time when there is a race, in a sense, for any firm that might be, or even around, somehow involved in, say, data in their pipelines or might want to use it as a customer service chatbot or whatever a firm might want to use it for, I would recommend pausing and thinking if 100%, if AI is necessary for that use case.

It would be very important to keep a human in the loop and to help empower your workers in knowing what the capabilities are, having a clear policy on how to use these tools, investment in training to make sure that workers can get the most out of these tools. So, I would say listen to your workers. Quite often, they know what helps them.

We’re also at IPPR discussing with the Trade Union Congress on how unions can also work together. Because for society and for firms, to get the most out of AI, workers need to feel empowered and comfortable. If they are feeling like they might be replaced or if they’re under surveillance, it’s not going to work for anybody. So I just want to put that as a very crucial thing: Firms must work with everybody who might be using these tools in a way that makes sense for both the managers or the decision makers as well as the workers.

And then of course, it really depends on your sector. See for your sector how is AI being used? Be able to see if it’s just a snake oil, if they’re just selling you something to just use a product. Be sure to be critical and grounded. It’s really important to make sure to not just use these tools because everybody else is using them, but to see what a potential use case is. See in a way where it’s fairly rolled out to your workers. Make sure that they’re educated well on it. Make sure that if there’s a productivity increase, there’s wage increases as well. So, all of these need to happen to make sure that this is being used in a way in society that really benefits everybody.

Bron: Forces of STEM – The ups and downs of an AI-augmented labour force

For more information about SThree, click here.

Lees ook :

• AI vraagt om een ethisch kader in organisaties
• Deze freelance opdrachten komen onder druk te staan door AI
• Hoe kan AI de STEM skills gap dichten?