As it turns out, AI is not universally great. Moreover, many companies have problems differentiating AI systems because they are not experts in this field, and may not have any AI expertise themselves, says Prof. Dr. Michael Feindt, Strategic Advisor at Blue Yonder.
In this exclusive interview, he talks about what AI can and cannot do, where the biggest hurdles lie in integrating it into business models, and how AI will change supply chains.
Thanks for talking to us Michael. The coronavirus has turned logistics upside down. What are the medium-term challenges for the world of logistics and what prospects will result from them? What trends do you see shaping logistics over the next five years?
Of course, coronavirus caused a stir, and not just last spring. Even now there are not enough containers in Europe and transport prices have gone up massively. At the moment, there are a lot of considerations, especially in Europe. One is to focus more on regionality and resilience. Personally, however, I believe that logistics will revert to its old behavioural ways.
On one hand, we see flexibility and adaptivity. On the other, automation of routine planning is important for companies. Especially during times such as these, when people are overwhelmed with individual decisions. This has now become evident to many of us. Up to now, predictability has meant doing things as it always had been, something that has not proven itself to be true in the last year.
At the moment, you have to be able to adaptively switch to better decisions. Artificial intelligence can help here, as it provides an overview of all relevant data, new information and the current status in the supply chain. This is extremely important in order to be able to reschedule more quickly and thereby at least prevent the worst.
Another trend that companies still act very cautious towards, at least in Europe – is cooperation. The aim is for companies to pass on planning along the supply chain to their partners in advance, so that all parties involved can plan as well as possible, thereby avoiding the scenario when everyone just optimizes for themselves and keeps their information secret to have a better negotiating position. We have been promoting a better exchange of information across company boundaries for a long time.
So you believe that the trend towards regionalization will not prevail in the long term? For reasons of cost effectiveness?
Yes and no. It depends on whether you think in the short or medium term.
The coronavirus pandemic has been unpredictable and that’s why there were massive losses for many companies. They relied on the Far East to work it out, and I believe that many companies have turned their thoughts to becoming more regional and thus more resilient to shocks. However, I suspect that it will return to the old system within the next two years, because companies will again want to have the cheapest available option at short notice.
How can companies prepare and react to these new developments?
By realizing what has happened now and trying to be more resilient.
We have now seen that automation and digitization are advantageous, especially when you have to re-plan. Right now we are dealing with a market shakeout, especially in retail. Many smaller firms will not survive and there will be market concentration. This is, of course, is not good for end users, as it is better for them if they have a wider range of products. But I suspect that everything will quickly go back to normal after the market shakeout.
The topic of AI has been discussed constantly for some time. Philipp Hartmann, Director of AI Strategy at applied AI, said in an interview that AI solutions are often developed separately from specific problems, or that AI solutions are used because they are supposedly „en vogue”. How would you comment on that? Is AI just another annoying buzzword?
At the moment, AI is absolutely a buzzword. Today all software companies make and use AI, but of course when you go behind the scenes, that’s not always true.
Artificial intelligence is currently a fashionable term that is being misused to an extent. On the one hand, it is good that there people and companies are buying into the hype, but on the other hand there is also a lot of hot air being spread around. We now have to repeatedly position ourselves as a company and show that we’ve been developing AI for over 20 years and are really deep into it.
In fact, not everything that has an AI label on it is great. It is difficult for many companies to differentiate here because they are not experts in the field and may not have any. Therefore, at least the larger companies should bring in one or two such experts who can assess this and have the trust of the management. Personally, however, I think the trend of developing everything oneself is wrong.
Should all small and medium-sized companies now deal with the topic? In your experience, where could the biggest hurdles be when integrating AI into your business model?
The question is not an easy one. These are of course new technologies, but they are not cheap.
The biggest hurdles today are data connection and the integration with your own system. Smaller companies definitely have a chance if they are technically well positioned, have a few good IT people who understand their own data, and can ensure that the connection to the system works. The new systems are all software-as-a-service systems. A huge initial investment is not necessary, as everything in the company is billed annually. The biggest investment is your own implementation and the change process in your own company.
Despite all appearances, change processes in large companies are not always easy. Although they have an advantage in terms of their scaling, they often suffer from being sluggish and are not necessarily innovative. There has to be an absolute willingness to change and, in any case, the commitment from C-Level.
Startups are important drivers of innovation. But these play a much smaller role in Europe compared to the USA, China or even Israel. Is there a lack of willingness to take risks in Europe? In your opinion, do people in Germany invest enough in young founders?
The willingness to take risks in Germany, but also in almost all of Europe, is significantly lower than in America and other parts of the world – not only from the founders themselves, but also from investors and from the large companies that could be customers.
Large retail chains and companies from the industrial sector are very picky about suppliers, and want them to have gone through all possible certification processes. A small start-up cannot do that. I therefore put the blame here primarily against industry in Europe.
In addition, we are an envious society. 97.5% of young founders do not survive the second year and are then stigmatized for having failed. The culture in America is different. If something doesn’t work out, you can try again. The situation in Europe is indeed more difficult.
In which European countries is the situation better?
Sometimes you hear that the Baltic states are further down the road. The UK is also a bit better than continental Europe, though there is not much spectacular happening there either.
Last spring, we saw that AI could not predict supply shortages and panic buying. So how intelligent is AI actually? What can AI not yet do, and what are its limits?
What AI cannot do is overcome the laws of nature. Such 'Black Swan' events cannot be predicted – even nature does not know what is to come. There are many things that cannot be calculated. They are only determined when they really happen.
We are talking about so-called chaotic systems here. The weather is one of them. Depending on the situation, we can predict the weather sometimes two, sometimes ten days ahead. However, after 14 days the chance of predicting an accurate local forecast is basically zero. Chaotic systems also include the economy, pandemics, politics and the EU. That means we can only make short-term forecasts, as these are always associated with uncertainty.
What we always do though is predict the uncertainty. At some point it will be so big that we can no longer individualize anything, and simply have to assume that in the long term, on average, people will behave as they always have done.
However, AI can help us to make these forecasts automatically and adaptively as best as possible, and to predict the uncertainty and probabilities as well as possible. Then we can make the best decisions about what to do with this knowledge. This is called mathematical optimization. But AI does not mean that we have a crystal ball and will know exactly what the future looks like? We don’t know how a soccer game will turn out, but thanks to AI we can calculate the probability of each outcome as individually as possible.
You have already mentioned that AI-based solutions are not exactly the cheapest. After what period of time does the investment pay off again for the customer?
It’s profitable extremely quickly. With some solutions, for example with automatic pricing in retail, the investment often pays for itself during the pilot. AI-based solutions are expensive, but for a large retail chain what they offer in savings is gigantic. Many solutions are therefore profitable in the first year and definitely in the second year.
In addition, many modern software companies now offer software as a service, which means you don’t have to pay millions in advance for the license like in the past. The initial investment consists only in connecting to the data in your own system and in your own change process. This is a fundamental decision and every company has to make it for itself and decide whether it can bear these costs. Today the hurdles are lower than ever.
How far is the use of AI in logistics and supply chains?
We’re on the right track, but AI is still has a long way to go. For most companies there is a big step yet to take. At the moment, a lot of work is being done on assistance systems so that they gradually become more independent. This is the way to go, and many people are naturally afraid of losing control and being replaced. In addition, there is a lack of confidence in the capabilities of machines. This is wrong though, because machines have access to the whole experience of humans and can better connect data because they work with the best mathematical methods.
In addition, freedom from prejudice also plays a major role. Our gut feeling is not right when it comes to many decisions. Sociologists once carried out an investigation into this on supply chains. The test subjects were divided into three groups: professors who knew the optimal solution but had to calculate it by hand, students in their first semester, and practitioners with more than twenty years of professional experience.
The exciting result was that all three groups performed equally badly and much worse than if the task had been done automatically. This is because such decisions are not made by a slow, rational brain, but rather by a fast, emotional “gut feeling”. People are therefore unable to objectively weigh up all the existing influences against one another. That’s all good and indeed important for survival in evolution, but it is no longer good enough for the tough competition in today’s information society.
How will AI transform supply chains over the next five years?
I believe that supply chains will become much more efficient in the future. One trend that I mentioned earlier is linking. So far, optimization has mostly been carried out per silo. What was missing was end-to-end optimization and horizontal integration. If hoarding and safety stocks are being held everywhere again, it is of course nonsense. If you worked with those in front of and behind me in the supply chain, I could manage with a lot fewer resources. The collaborative cooperation, i.e. the so-called horizontal integration, is becoming one of the most important topics in the supply chain.
Another trend is resilience. At this point I have to say that algorithms do not care what kind of cost function or optimization function we put in, the management has to decide whether it should contain short-term or medium-term profit or, for example, customer satisfaction or sustainability goals. The individual operational decisions are then made automatically in order to optimize the specified strategic target function.
What comes after Artificial Intelligence? Is AI the beginning or the end?
Artificial intelligence is still a long way from being conscious. Many have rather science fiction-esque ideas about it. I wouldn’t say that AI is really smart. Without the people who design and control such systems, it would be stupid. But once we have built it and a problem has been solved, then AI can also do it fully automatically for the next customer, even if everything is different for them. That’s because AI can adapt itself. So far, however, artificial intelligence has only been able to solve clearly defined problems and no more. We humans have a consciousness, but AI systems are miles away from it.
When it comes to the automation of processes, we can see a steady progress. As regards object recognition in images, algorithms are now better than most people in that they can separate wildcats from one another, for example. A few years ago that was very difficult. The same goes for automatic translations, which were once utterly disastrous, are now only moderately disastrous. Eventually they will come good too.
Are we at the beginning or the end? Experts say it will take another 30 to 60 years for a computer to have as much computing power as a human brain. Then computers can become conscious, and I personally believe they will be. Consciousness is not something that only we humans can have, but something that arises from emergence. Humans are also an emergent system. Individual cells can be explained biologically, but they all work together to form a human being who has organized states and cultures. Something like that cannot be predicted. You see in nature that something new suddenly arises that you wouldn’t even suspect. And then you can start talking about the end. But there is still a long way to go.