Skip to content Skip to footer
Search

Loading Results

TechTalk ep1: Are robots really taking over our jobs?

Paramita: You're listening to TechTalk, PwC Luxembourg's brand new podcast. I'm your host Paramita Chakraborty. We will talk with experts and try to tackle all things tech on these weekly conversations.

Artificial intelligence AI has been quite the talk of the town lately, particularly for its touted capacity to take over our jobs. For our very first episode, I sat with our A.I. expert Emilia Tantar to understand what's AI’s or automation’s impact on the service sector and on jobs in general.

P: Hi Emilia. How are you doing? Excited to be on the very first episode of our podcast?

Emilia: Wow thank you for the honour.

P: The honour is all mine.

So you know we're going to talk about artificial intelligence and automation. But can we talk about both the topics interchangeably. Is there any difference between A.I. and automation?

E: Thank you for this question. That's a much needed question. Artificial intelligence is a term that we use for different technologies. It's an umbrella term and automation is a bit different. It's not something new. We had automation for example when Leonardo Da Vinci created the first mechanical cavalier. That was in 1492. And then we had other types of automations which went more into the creative spectrum like for making music you can see it in the Museum of music instruments in Boston. So a lot of automated machines. We had it in the 19th century we had it for example to serve the tea. The karakuri in the Japanese culture. And that you can see also in the Tokyo National Science Museum.

P: So they are kind of a robot, is it?

E:  Yeah. They see them like robots but they do mechanical automation. So there was no computational power. Then you know things have evolved and we go in automation in tasks which are a little bit more challenging. Like we have automation which appears in game playing. You can create strategies and we had really interesting exploits in that regard. But at the beginning, in automation… even in chess we had the Mechanical Turk.

It was a mechanical machine that was simulating, it was an illusion. It was the illusion of the depth of the furniture. Well, you had this guy inside who was playing chess and was manipulating the muppet.

P: And so we are still talking about automation here not artificial intelligence.

E: Since ’57, since the conference at Dartmouth, we use the term artificial intelligence. That was the first time it was coined.

Different paradigms… you had neural networks, you had fuzzy, so it's not just one paradigm. What has changed and why we are again speaking about it is because of the new exploits which are enabled by the graphical processing units. So since 2000-2010 we have a lot of computational power which is not expensive. So we can win at chess because we can do really well number crunching. What's still keeping us, let's say, from moving forward is the fact that the machine can advance only based on whatever we are teaching it.

P: So, it all kind of depends on algorithms, what we put in, right, as inputs?

E: Yes, it depends on algorithms. On the algorithmic part, since the 50s we had incremental advancements, slow advancements. What has drastically changed is the way we access information. I'll give you an example: I'm sitting on a chair. I can understand that it has four legs. It has this seat and I'm sitting on it.

If I give 750 images of a chair to a machine and I associate the label, the word “chair” with the images, the machine will always be able to associate the word, just the word, you know this string of characters. But it will not know that it has four legs and that you can sit on it. So if you think how we perceive knowledge and we manipulate knowledge and information, as humans we are way more advanced. A machine needs that we put in the context, for example, in case of transactions. We need to say this is a fraud or this is not the fraud. The machine wouldn’t know it by itself. And the second part is that we need to correlate it. For example, the chair is in this room. The machine doesn't know that. We know that by experience, by perception. But the machine is quite limited. So what we are lacking now is this encyclopaedias of knowledge like we have dictionaries where we take our standards from. The machines do not have that. You have some exploits in image processing for example as I explained with the chair. If we label enough images we can recognize some shapes.

P: The machines need context.

E: Exactly. Data, as I always say, it is the new black. It's not a colour. It's everywhere. But without a context is meaningless. I always use this example: you say, “I went to Harvard,” and then you stop and then add, “I took the 5 o'clock train”. So my context is not an educational context anymore, it is a touristic context. The same happens, for example, with data which are domain specific. I have an exact science background. I will not be able to distinguish between a transaction being fraudulent or not. For me they look alike, you know, the semantics I don't have it there. But I need this kind of knowledge, structured knowledge in order to make sense of the information in order to put the algorithms providing improved results.

P: Excellent. So when we talk about let's say our work you know the kind of work that we are doing here and you know I spoke to you about it… the banner that we had in the Experience Center that said “Robots are not taking over our jobs”. So when are talking about our work where does all this fit into. Where I mean is it A.I.? Is it automation? Is it taking over our jobs or it's just overtaking us a little bit? What's happening?

E: That's an interesting perspective. I would say that we are in a service sector. And when we think about robots and about the way they manipulate information, robots they are only enablers like this table or the chair. If it has computational power, it has the same powers as a robot. If the robot has, for example, a humanoid like face, we’ll build empathy and we’ll think that it's intelligent that it's more intelligent than this chair, than the table. But it is not the case. It is the same intelligence, the same kind of machine intelligence that we have in all three of them. So robots they are not approaching human intelligence to take over the jobs. But robots, what we call now, we refer to the background services and not the shape that we put them in. They have access to information in fast ways. So if we provide them encyclopaedias of knowledge where the information is labelled and is correlated, they can move fast. They can do number crunching and information crunching faster than a human can. This is why, for example, they win at chess because they do all the possible combinations. We are not capable of doing it that fast. Of course we can do it and we have the champions but we are not able to do it that fast.

So in the service industry what's changed drastically is the fact that for the tools which are enabled by good encyclopaedias of knowledge, of domain knowledge, together with A.I. for navigating this information, we will have faster services, better in predicting, as good as the information we provided. So, we will need to adapt to this new type of access to information.

P: How do we adapt? Do we train? What do we do?

E: That's again a good question. We need to upskill ourselves. From our education, we all learned how to learn. So we are adaptable as human beings. How we adapt depends on us. When you build, for example, these solutions which are relating with A.I., you have two pillars. You have the domain knowledge. Domain knowledge can come from manufacturing, it can come from the carpenter, it can come from the financial industry, it can come from any domain. Any domain which has information is affected.

The second pillar is a more technical pillar. There you need experts to build this automated machine learning. So it builds, I would say, the shell. It builds the tool, the machine which crunches the information. So we have the two pillars, one being the technical one where you need technical skills like understanding the data but also how to navigate the data. So that's the technical skill and it involves data science, machine learning and so on. Besides the technical skills, you need to still rely on all existing domain knowledge. So everybody has his/her place in this new evolution. Because you bring out domain knowledge, you build tools and you support in maintaining the tools.

The great difference between the software tools that we have now-a-days and the so-called A.I. enabled tools is that information becomes obsolete, at the most, in one year time. So you need to maintain it with human knowledge. Nowadays that part is quite ignored. So you just think there's magic and you put the algorithm sand you put some data which is relevant now in this moment for this environment. And you think it will do everything. We don't have generally artificial intelligence, so you need to build the tools and you can be part of the process of building that. You can be part of the maintenance and then you use it… to improve the services that you are currently offering.

P: OK. A very naive question here. Let's say, I work here and I see that my job is going to be automated by an algorithm. What advice would you give me? What would you tell me?

E: If we think about automation, it can be applied to normal situation, normal behaviour. So we can automate what we consider as normal. What we saw in the past is normal. What we cannot automate is an “abnormal” situation. The ones we didn't see in the past. And the ones that we can't predict because we didn't see it. And the presence of these systems will bring a lot of unpredictability. So it will shift from checking the normal behaviour to creating analysis for the abnormal behaviour. And there'll be more and more abnormal behaviour. Now it takes us long to determine this, let's say, false negatives when a fraud occurs. If we have more and more frauds, the job will shift from treating the normal to treating the abnormal and also adapting, creating tools to adapt to these new situations. So I would say no human being can be replaced by automation because our particularity is that we adapt, we are able to adapt ourselves and we always create knowledge.

P: And our particularity, I guess, is that we are human. I mean, still… we don't know what the future holds…

E: Yes. We have singularity… Where it’s said that artificial intelligence, synthetic intelligence would overpass human intelligence. But that needs to be controlled. We need to build tools but we also need to put controls in place. But not in a fully automated way. Why? Because we take risks, not all processes behave normally… they behave stochastically. They don’t behave normally, all the time. So, I would think human being has an important role in creating tools and putting controls where we think that the speed of computation or the memory capacity is over passing that of the human.

The decision is always with the human because we have the theory of computing where Vilfredo Pareto (Italian engineer, sociologist, economist), explained that whenever we are faced with a real problem we have several criteria which can be different by nature or contradictory and there is no ideal solution.

We have a set of best compromise solution. So should we let the machine decide which of these solutions from the set of best compromise solution is the best for us?

P: Or we let us “us” decide. Yeah okay. So we spoke about training a bit earlier. Are there any concrete examples that are going on? On European level or in the world that you know businesses or governments are providing to employees for them to adapt and to train for this change that's happening?

E: Yes, that's a good question because actually we were working in the national programme In Luxembourg called Skills Bridge which aims in preparing the workforce for the new skills. It is enabled by the Ministry of Labour and by the National Employment Agency ADEM and that creates the link between these profiles of the individuals and where they can be upskilled and where they can transition. It identifies what skills are necessary in order to transition to a new position, what new positions appeared and also what training is required.

P: So it’s kind of a matchmaking thing. And in the European level? Do you know if there is anything?

E: Yes, I can think of one particular programme: Skills for Europe, which is run by the European Commission and in which we are working together and identifying what are the new skills which will be required and how education can be adapted to prepare for these new skills.

P: So we do have access, apparently, now to these kind of training programmes. Last but not least, I don't I don't think we'll have much time today to go into a lot of details, but can we say I know that we have heard a lot of negatives about A.I. and automation and how, you know, they're overpowering humans, can we probably end our first episode on a positive note?

I mean can we say that A.I. or automation can actually be enablers and can actually be an ally to the future of workforce?

E: In fact, it is an enabler. It's like electricity. We have a background service which allows us to have fast access to information, like here with electricity, we have fast access to light. So what changes is the way we access information, the way we construct information. In the olden times, we relied on books and we had this linear access to information… from left to right, from right to left. You even had organs which went on the vertical that but it was basically linear access to information using different materials. You had wax, you had to paper, you had papyrus you had clay… you had many different material or medium. Now the medium is a machine what you call a computer. And what we are lacking now and we need to build in order to empower ourselves is the encyclopaedias of knowledge for the machine with correlation, explaining the context, domain by domain.

Why we need to do this is to be responsible like this we ensure that all people alike have access to information, that we have different perspectives. Because we all have a different perspective. The way dictionaries were created, it was using different perspectives. Each of our perspectives is valuable. So we all have valuable opinions on our specific areas and we can all contribute to these encyclopaedias of knowledge and put the control points in the encyclopaedias so that when it becomes operational and the machine crunches the data, we know that we can stop it at the critical moments or the waning moments. So we all have a place in this industrial evolution of artificial intelligence, access to information because we are all bearers of knowledge.

P: Thank you so much Emilia for being here. And I'm sure in the coming episodes, we'll have occasions to meet with you again. Thank you so much.

E: Thank you Paramita.

P: So that was my conversation with Emilia. Hope you enjoyed it. If you did or if you have anything whatsoever to say, please drop us a line using the #PwCTechTalk on Twitter and I'll see you next time.

Contact us

Communication contact

Commmunication, PwC Luxembourg

Tel: +352 49 48 48 5821

Follow us