Karen Kharmandarian and Alexander Zilliox, Chief Investment Officer and Portfolio Manager at Thematics Asset Management, discuss what investors need to know about the AI-enabled future of work.
AI will be a huge question both for businesses and policymakers. It is very easy to encode unconscious biases in algorithms that may not be immediately apparent when they are deployed, and these can be magnified by data that represents biases already present in the context they emerge from. As the saying goes, if it’s garbage in, it’s garbage out.”
~ Karen Kharmandarian
Chief Investment Officer, Thematics Asset Management
AI is a very broad category that can mean different things to different people. How would you define it?
Very basically, AI is the ability of machines to perform tasks usually done by humans. In the early days, it was focused on very simple and mundane tasks, but nowadays it’s getting more and more complex and they are taking on increasingly complicated tasks.
There is this perception that AI is magical and something that we are working toward in the future, when really it is about the expansion of computing power thanks to increasingly sophisticated engineering and software.
A lot of software functions we take for granted today – Siri, for example – are very sophisticated AI products by the standards of the recent past. We have been conditioned to see AI through the lens of science fiction films, when the reality is that it represents a growth in power and expansion in capability of systems we largely already have, or at least are recognisably of the same lineage.
It’s fun but it’s not really practical to think in terms of general intelligence or quasi-sentient computers at this point, especially if we’re building real-world investment cases.
There is a huge amount of ‘noise’ around AI – how do you cut through and identify investment opportunities in promising companies?
We try and bring it down to fundamentals that can be quantified. For example, companies that offer their clients substantial savings by automating their back-office processes.
As Alex says, a lot of the real opportunities lie in the B2B side, rather than the more funky stuff like humanoid robots attending the front desks in Japanese hotels. That’s great fun, but not where the true opportunity lies. AI’s real potential is in the somewhat more mundane, but far more valuable area of improving the efficiency of almost any operation. It offers more productivity, fewer mistakes, better client service – things that really add to a firm’s value proposition.
Then again, we are seeing more and more ambitious applications. If we look at industrial design – in the beginning, as Karen says, AI was applied to find efficiencies.
The next generation, however, is generative design, where engineers input certain parameters around dimensions, materials and so on, and then software designs various iterations of the product optimised for certain outcomes. Software won’t create anything it’s not asked to, so there won’t be flashes of inspiration, but I think the opportunities here for AI to back up human creativity are clear.
If we look at the workplace, what kind of applications of AI do you think most white-collar workers will see by 2030?
There are a lot of use cases in HR that are beginning to come online. Absence management, expense management, background checking and so on. Taking that further, there are applications in performance management, where areas of weakness can be identified and training offered to compensate. However, it remains to be seen how comfortable professional workers will be with what amounts to machine surveillance and evaluation.
Also, biases are a notorious issue in recruitment. AI recruitment platforms can account for these and evaluate candidates in a far more equitable fashion than even the most idealistic HR professional.
This raises one of the perennial issues around AI that will be a huge question both for businesses and policymakers. It is very easy to encode unconscious biases in algorithms that may not be immediately apparent when they are deployed, and these can be magnified by data that represents biases already present in the context they emerge from. As the saying goes, if it’s garbage in, it’s garbage out.
Let’s turn to financial services. This would seem an area where AI has huge potential – first, where is it in the picture if we look at B2C?
Customer service is one area already being impacted. Chat-bots can process a lot of small-scale insurance claims very fast, for example. One company in our investment universe is already processing 30% of their claims that way, which is huge. This frees up staff to work on higher-value tasks, while also providing a more granular and reactive service to their clients.
We can take this further: the trend is clearly toward automating mortgage applications and larger-scale insurance policies. Credit checks are already largely free of human decision-making, but financial services firms with more powerful software and access to more data can have significantly better understandings of their prospective clients’ risk profiles. Again, this can sound almost Orwellian, but the outcome would be fewer people falling through the cracks, and more personalised pricing structures – benefits consumers will find attractive.
We should talk about our industry, asset management. Where does AI fit into the future of our jobs?
There are firms that use AI and huge data sets for stock selection. We’ve seen companies build AI capabilities in-house that run mock portfolios, and they’ve seen that they can sometimes outperform both industry benchmarks and their real fund managers. So, more and more portfolios are being managed at least in part by algorithms. This has the obvious attraction of not being affected by emotions or bias, which even the most cool-headed investor would admit can get in the way of clear decision-making.
Algorithmic investing solves some problems, but it’s not going to replace humans as you have the ‘explainability’ problem. They can be black boxes: it’s impossible to know precisely how they operate and the exact reasons why they do what they do. These aren’t abstractions, these can be decisions with serious real-world consequences, so you have to have accountability or you’ll have some very unhappy clients.
As AI has increased in capability, this has in turn increased the value of the data required to ‘train’ various solutions. Data is ‘the new oil’ in that it is now a commodity that underpins a huge range of other economic activity. Is that borne out in investment strategies?
Absolutely – think of how children learn. They ingest a vast amount of information, process it, and eventually are able to use it to draw relationships between things and form an understanding of the world around them. AI is no different: the more data you have, and the better quality it is, the higher the quality of the AI solution.
For example, if we look at the cybersecurity industry we see legacy players who operate in a highly siloed way. They have been extremely vulnerable to disruption from newcomers who are cloud native and are better able to consolidate and cross-reference all their data pools. This in turn allows them to build superior AI products, which is all based on their ability to draw from as deep a reservoir of data as possible.
We can take that further, because it’s not simply a question of quantity – quality is also hugely important. The nature of the data and how it’s organised, analysed and processed are critical. How you don’t just get energy by digging oil out of the ground and setting fire to it. But we don’t want to stretch the metaphor too far.
Also, there are new techniques that are arriving where you need much less data to train the algorithms, which will dent the competitive advantage in owning a large amount of proprietary data. For example, Google’s DeepMind has predicted the structure of proteins in DNA based not so much on crunching vast amounts of data, but on highly sophisticated techniques based on better understanding of the problem that faced it.
This is hugely important, as so many of the predictions of where the industry is headed have centred on the idea that more data equals more success. The future could in fact be quite different to how people have assumed.
Many people have anticipated AI cutting white-collar jobs in the same way robotics massively reduced the headcount needed for manufacturing. Do you think this is likely?
Yes, if you automate your back office, that you need in your back office. But on the other hand, there will be new jobs created, and often these will not be a huge leap from the skillsets of back office workers. Creating algorithms, monitoring them – they won’t be totally autonomous, and the idea that there won’t be any people at all is probably going to go the way of the predictions we used to hear about with the paperless office.
It’s true that there have been some academic studies suggesting truly alarming job losses. I think one of the misconceptions is that there is a difference between automating tasks and automating whole jobs. The picture is a lot more nuanced than some of the doomsayers maintain.
There are other ways for employees to create value than repetitive tasks that require little creativity. Soft skills, for example, will become ever more important – whether its negotiation skills or a bedside manner, there are things AI just won’t be able to do in any realistic timeframe. I think we have gotten used as a society to the idea of working for a longer period of our lives and reskilling for various portions of our career – I’m not sure the AI challenge is significantly different to this.
Down the line, questions do arise about whether a five-day working week will be necessary to generate the same purchasing power, and many younger people talk about universal basic income as a solution to AI job losses. There can be no doubt that they are only going to become more salient with time. But again, this is on such a long time horizon that certainty becomes impossible.
There seems to be so much consensus about the potential and capabilities of AI that it could be wise to consider counter-narratives. Do you think the consensus is accurate, or do you think we need to manage our expectations?
Clearly, expectations are high. There is proof that certain things are working, but it’s also not hard to find areas where we may need to be a little more realistic. Look at fully autonomous cars – they have been ‘five years’ away for a lot longer than five years, and that’s not even factoring in the enormous legal obstacles you’d face if you filled major cities with computer-controlled vehicles.
I think we can look to the past few decades as a reference – computers arrived in every industry and every office quite a long time ago now. While they radically changed how we work, ending things like typing pools, I don’t think there is a slam dunk case that in the developed world ICT alone has unlocked productivity on a grand scale.
There will be other factors of course like demographic trends and the macroeconomic picture, and with general purpose technologies it can be hard to draw causal links to specific outputs. But overall the digital economy has not been a machine to unlock enormous value and productivity across economies. It has changed things a great deal, of course – I don’t want to sound negative – but I think it’s hard to argue that the pace of development in, say, the last thirty years matches that between 1950 and 1980.
Bearing this in mind allows us to take a cool-headed approach to AI. I think people overestimate what AI can do in the short term, and underestimate what it can do in the long term. Not long ago, you could have guessed that software companies like Microsoft, hardware companies like Apple and a digital retailer like Amazon would become enormous – we could anticipate these opportunities. But not many people would have guessed that two of the most valuable companies in the world would be in search and social networking, and that’s not even going back that far.
Indeed, companies like Google and Facebook were able to build on the achievements and infrastructure of those who came before them, in an exponential curve of capability. That’s what I think AI will be like – we’ll see slow improvements at first, then the pace will be dizzying as innovation builds on innovation. The real excitement is going to be in seeing what companies and capabilities emerge that we can’t anticipate today.
THEMATICS ASSET MANAGEMENT - An affiliate of Natixis Investment Managers. A French SAS (Société par Actions Simplifiée) with a share capital of €150 000. RCS Paris: 843 939 992. Regulated by the AMF (Autorité des Marchés Financiers), under no GP 19000027. 43, avenue Pierre Mendès France 75013 Paris, France.