“Decision-makers have a vision of AI tainted by newspapers, Star Trek and Star Wars”

Axel Angeli, Cybernetician and Digital Future Evangelist

In an interview with Bizweek, world-renowned German cybernetician and “digital future evangelist” Axel Angeli explained why AI should not be blindly trusted, but used as a tool to be combined with reliable data, good human judgement and practical use cases. He also delved on the cognitive industry, on ocean intelligence, and on Mauritius’ potential role as a regional hub for AI-driven blue economy research, which is an initiative of Mauritian businessman and entrepreneur Georges Chung Tick Kan.  

Klyven Veeramundar

You are often introduced as a cybernetician and digital future evangelist. For readers who are not familiar with the term, how would you explain cybernetics, and why does it matter in today’s digital economy?

You have probably all heard about cybercrime, cybersecurity, cyberspace… everything cyber. The word cybernetician was there before there was informatics. It was the original word for everything that was around computers. And since we are now going back to the initial thoughts of computer sciences, which is artificial intelligence, I thought I would choose the word cybernetician, because we go into cybernetics, which brings AI and classical things together again.

 

“What is the problem with politicians? The problem is that they are not aware.”

 

As for “evangelist,” in America, it is simply something like a professor who is not working in a university. It is somebody working in a company who teaches the latest technology, or the digital future, to their customers. That is my topic, and I evangelise it. I bring the good message to the people about what it can do.

Your work looks at complex systems, technology strategy and transformation. What do companies and governments usually misunderstand when they try to adopt AI or digital tools?

It is an interesting question, because what they do not understand is that they lack a lot of basic knowledge.

What they have – especially governments, politicians or decision-makers, and I do not blame everything on governments, as it is also in companies – is that the decision-makers, if they are lawyers, if they are business administrators, they have a certain vision of artificial intelligence which is tainted by reading the newspaper, by what they hear in the news, but also by some science fiction coming from Star Trek, Star Wars, or from some trashy science fiction like Alien, where they heard about things.

Good science fiction includes things like Total Recall, with Arnold Schwarzenegger. Although this does not really look like the future, a lot of things were predicted there, like self-driving cars, like a robot that can disassemble and reassemble itself, robot wars… so there are a lot of things there.

 

“90% of the data that would be helpful is kept under lock by governments. And from this 90%, I would say 80% is unnecessarily locked.”

 

I would say that the biggest issue at the moment for decision-makers in general is a lack of education. What I would love to give, especially to governments, is regular updates – not from the fancy, fashion-driven side – but with real facts.

You must know one thing: artificial intelligence is a lot of artificial. Intelligence is not so much built there. It is just big data processing at lightning speed.

You said that you would like to explain this to governments. Have you done it in Mauritius?

Together with George Chung, we tried this in 2018. There was a conference, and I was even invited by the government at the time. The idea was also to teach people in government what is possible, and to be an independent alien advisor, not somebody who has a political interest in the country itself, but somebody who comes from outside and who looks at it more or less like a professor, like a scientist.

Unfortunately, this did not continue. But I would be really interested in doing some teaching for decision-makers in Mauritius, in the form of guest lectures or whatever is convenient here. I would look at the audience as government decision-makers, as well as local businesspeople.

In the countries where you have given lectures, how satisfied were you with the results obtained?

Honestly speaking, I was not satisfied. With all this information that comes through the internet and through the news, I am just another voice. Often, people do not take it too seriously. It is like some kids saying, “Tomorrow I have to go to school.” They listen to what the teacher says, and by the afternoon, they have forgotten what was said.

 

“ChatGPT is often biased. If the same information comes too often, it believes it is true.”

 

So, in terms of the effect and impact of what I teach, I would wish for more. Internationally, anywhere in the world, there is no difference. But I must say, in what are called the more developed countries, like in Europe or in America, the problem is much deeper, because there, everybody thinks they know everything better, even if they do not know.

When you say this phenomenon is because of certain influencers who are doing something around AI, do you think that affects the public’s view of what you are teaching? 

The main reason is certainly information overflow, information affluence. You get information from everywhere, and the normal listener has difficulty making a distinction between what is quality information and what is just somebody saying something.

There is also a lot of what some would call fake information or tainted information, where somebody, in the end, wants to sell something rather than teach you.

Whoever they are, most influencers are nothing else than internet salespeople. These people are not giving you the right information.

Science channels are relatively rare, or they are not promoted, because either the studios believe they will not have the audience for them, or the audience itself finds them too boring. Maybe we just need to make them more entertaining.

You have spoken internationally on the future of technology. Are we entering a phase where AI will mainly improve existing systems, or fundamentally redesign how economies work?

We need to make clear that currently, we have two kinds of AI ecosystems. One is public AI, the one we know with ChatGPT, Anthropic, Claude, Gemini and all these things. This is for the masses. The place where AI has much more impact on industries and on reality is what we call industrial AI. And there is a big difference.

With public AI, it does not really matter what ChatGPT gives you as a result. It gives you something. If it is true or false, okay, maybe it hallucinated. But in industry, if you have a self-driving car and an AI, the decision this AI makes has to be safe. You cannot rely on saying, “Oh, maybe I got the wrong data. I was hallucinating. I saw an elephant on the street, but there was no elephant.”

So, you must have a completely different way of treating the data.

Secondly, industrial AI works on a far smaller amount of data, typically only on the data that is available in the company or in the company network. It cannot just take data from everywhere, or steal data from everywhere, as the big ones really do.

It prevents this problem of garbage in, garbage out…

Garbage in, garbage out. Yes, exactly. On the other side, industrial AI is currently changing the world. I have coined the term for this: cognitive industries. For me, this is just Industry 4.0. That is already a kind of standard that brings smart industries, with IoT devices controlled by computers, together with industrial AI capabilities.

Both together, I call it the cognitive industry, because it makes machines think in a certain sense, to make decisions by themselves, to improve themselves, to find flaws themselves.

If you have seen Hannover Messe, the biggest industry fair in the world, this year, the whole topic was about industrial AI and cognitive industry. That is where the industry is going. And that will change a lot. It will also allow us to have production that is unmanned. There will be no humans on the shop floor, which is already the case in big companies like Toyota.

But in small and mid-sized businesses, you will also have robot-driven manufacturing. This does not mean that we get rid of people. Robots can do jobs that humans cannot do. They can lift heavier weights, but they can also work in atmospheres where there is no oxygen.

If you build a hydrogen car and you work with hydrogen, you do not want to have oxygen in the same factory. So, you try to get an oxygen-free shop floor, and that is why only robots can work.

But what will happen is that you, as the operator, will just move to a cockpit. You sit there and control your shop floor from a kind of computer game-looking cockpit. You do everything from your cockpit.

We also know that remote surgery is currently becoming a trend. So, you are sitting in India and doing the surgery, but the patient is in Mauritius.

This is also something that happens together with cognitive industries and the digital twin. We first take the real data, put it into electronic data – which we call the digital twin – and based on the digital twin, we can do all the machine learning, all the AI, and also the visualisation in the form of a real-looking computer game.

Would you say that public AI is not really that useful compared to cognitive AI? If, as you argue, there are some things that Google itself can do, what is the actual purpose of public AI?

Let us start with the GPT hype. It is a bit like a magic trick, because the results look awesome, but technically, it is quite simple. The big AI in ChatGPT is not in GPT. The big AI is already in the search engines like Google and Bing. They already get the data from everywhere and put the results in a form that is somehow useful. They find the real data for you.

ChatGPT adds the feature that it can read 1,000 results in two seconds, make a summary, and give you a summary of what the search engine has produced before.

On the other hand, it works on unstructured data. In industry, you typically have structured data. Everything is somehow in tables. Even before, it was on paper and in paper spreadsheets. 

Even the table structure itself gives you information. If an entry comes later in the table, it is probably newer than the one that was there before, and maybe it is outdating the older entry. So, this is a different way of treating data.

Basically, both are nothing else than big data processing. But the focus of industrial AI is on reliability, reproducibility and safety. Safety not only in the sense of security. Security means that hackers are not coming to steal from me. That is, of course, important. But safety means that my car is not steering left when it should steer right.

While most organisations now speak about AI, not all of them are ready for it. What are the basic conditions that need to be in place before AI can create real value?

We are aware of this situation. Most small and mid-sized businesses are not ready for it, because it costs money and they do not have money like a Toyota, which can develop its own digital twin environment and do everything like this.

What is currently missing is a kind of standardised toolbox that allows companies to easily implement cognitive factory capabilities.

It is like at the end of the 1980s, when ERP systems became popular. Before that, everybody had their own accounting and shop management systems. It was all self-written, and it was not very good.

Then companies like SAP came. They developed a standard for an ERP system, and in the end, it was much cheaper to implement. Everybody could have such a standard, and because it was a standard, it was also good for interoperability. If you have a standard, you can better exchange data with other systems that work with the same standard.

The same thing is now required in the cognitive industry area. We need a platform that can collect the data easily, that you can customise and not develop. It should do all these standard things like security, safety, checking and everything that a small company would not even think of doing, and that would cost a lot in development.

So, we develop one platform for a thousand companies, and then we share the development costs among them. That is something my team is also working on, together with some Indians. They have a plan to build a system that goes by the name Karuna X. I believe that such a platform is what will be needed in the future. Otherwise, small companies will not be able to pay for what is needed.

As someone who studies systems and feedback loops, how do you see the relationship between human judgement and machine intelligence evolving over the next decade?

Currently, artificial intelligence relies very much on human judgement, because if you train a machine through machine learning, you need humans who teach it. You are the daddy, and it is the baby. It gets a lot of unstructured data, and when it assumes things incorrectly, you correct it and say, “This is right, this is wrong.”

So, you show it an apple, you show it an orange, you show it a banana, and in the beginning, it is just guessing. “This is a banana.” Then you tell it, “No, the banana is this,” or “No, this is an orange.” After a while, the machine can make the distinction and say, “Okay, there are certain characteristics, for example, the colour.” So, it finds out that every time you say orange, the colour is orange, and every time you say banana, the colour is yellow…

It finds this is a good way to discriminate. This is also like a baby. The baby, first, will discriminate between the apple, the orange and the banana by the colour it sees. The shapes are also different, but you need to teach that. You show it 10,000 pictures of apples, bananas and oranges, and every time you tell it, “This is an orange, this is a banana.” After a while, the machine gets enough criteria to make the discrimination between the three different elements. So human judgement is needed.

I think you are also referring to how AI can influence human judgement in reverse. I think it is already doing that, but it is the same as what influencers do. They permanently tell us that eating certain proteins is very good for you, and then everybody believes that eating proteins is good for them. But the reality is that it may be good for certain types of people, while for some people, maybe it is damaging.

AI is already shaping our recognition because people go to the search engine, to ChatGPT, and then they believe the result it gives them. I think the biggest change is that it makes us lazy in thinking. Very lazy.

I have a similar example from my youth. When we were young, we did not have a pocket calculator. So, if somebody said 17 times 35, we immediately knew the answer because we could calculate this in our head. Since the pocket calculator came, nobody can make this calculation. It takes several minutes to find out. A similar thing is happening now with these ChatGPT engines. 

A second problem, together with laziness, is that people are too willing to trust the result the computer gives them instead of verifying whether it is really possible. If somebody says there is a man living on the moon, and ChatGPT tells you this 10 times, you believe there is a man living on the moon, and you do not verify it. Instead of developing the capability to search, deep dive and look for the truth yourself, you just trust the answer ChatGPT gives you.

Would you say that the real problem is that people are not aware of what was searched beforehand by the AI? 

I would say, first, that ChatGPT is reproducing something. If ChatGPT has not read the same book that you have read, then it will not take this information in. If, for instance, it has only read books about vegetarians, then all results would favour vegetarian food. If it read only books about meat, it would only get results that talk about the best steaks in the world. So, it is also a matter of balancing.

It all comes back to the laziness of people. Most people who go to ChatGPT are not willing to think. They go to ChatGPT just to get an answer. They do not even say, “Okay, the answer could also have been in the encyclopaedia behind me on the shelf.” They do not even go to Wikipedia anymore. That is already a fact: Wikipedia usage is going down because people immediately ask ChatGPT. Wikipedia would be a much better source to get a reliable answer. It is not 100% reliable – nothing is 100% reliable – but it is much better than what ChatGPT is.

ChatGPT is often biased. If the same information comes too often, it believes it is true. That is why people like Elon Musk sometimes get advantages, because they are deliberately filling the internet space with information which is not false, but which is definitely not true, and definitely not the focus you expect.

Ocean Economy AI Lab says that large volumes of ocean data already exist, including satellite and environmental data, but they are not easily usable in daily operations. How can AI help turn this complex data into practical decisions for fishermen and vessel operators in Mauritius?

What the Ocean Economy AI Lab does is classical machine learning. If you have reliable data, you implement statistical methods to predict the future. In this case, the future is where the fish schools will be the next day, so that fishermen can target this area and be where the fish will be. Before, it was quite aleatoric – George would say a miracle. But in this case, it will be very deterministic.

So, Ocean Economy AI Lab does a very conservative, classical artificial intelligence application with a very clear and very beneficial use case. You can see the advantage for the fishermen. You can be sure that the results are as good as the data is.

Again: garbage in, garbage out, of course. In terms of Ocean Economy AI Lab, they are currently working on reported data, historical data, data that fishers report from yesterday. They try to predict, they do time series, and predict the future based on this.

That is the technical background. It is nothing new. It is something that has existed for centuries, going back to Leibniz in the 17th and 18th centuries.

What probably will be the next step for Ocean Economy AI Lab is to get real-time data. That is now the bridge to the physical world. You must tag the fish. You can take some mock fish, which swims with the other fish, and then you can control this.

Something like this will be the next step. But for now, Ocean Economy AI Lab does very conservative, very classical AI. The use case is there, and that is what industry needs most. We do not always need the fancy front. We need to stabilise. We do not need a flying taxi; it is good if we have any taxi at all.

The Lab’s first product is a potential fishing ground mapping tool, designed to support small-scale fishermen and larger fishing vessels in identifying areas where fishing conditions are more likely to be favourable. What could be the real operational value of such a tool for the Mauritian fishing sector?

The operational, or in this context commercial value, is saving money. Saving costs means more net gain. If fishermen go to an area where the fish are not coming, they will not harvest much fish. If the likelihood of going to a place where the fish are present is higher, then they will have a better harvest, so they will earn more money. If they go to a place in vain because there is no fish, they will also waste petrol, which costs money. So, the biggest commercial value is saving costs, or, as we would say in economics, higher efficiency.

The first phase focuses on tuna species, while future models will also look at demersal species such as Berry, Capitaine, Corne and Vieille. Why is it important for AI systems to be adapted to specific species, seabed conditions and local fishing realities?

Let us come back to the example of how machine learning works with apples, bananas and oranges. I think that explains it.

If I train the AI system on bananas, the system will know everything about bananas. The next day, oranges come in, and the AI says, “Orange, strange banana. The colour is not good; the shape is different.” It will try to apply the same results that it had on bananas to oranges. But the way you harvest a banana and the way you harvest an orange are slightly different. That is the general reason why you always must train on the specifics of the target system, to get as much detailed information as possible.

In terms of fish, if you train the system on bluefin tuna, that is nice. But if you now have some salmon, the system is always looking for the bluefin because it thinks this is the important significance. It will not find a bluefin on some salmon. It will not even find the salmon moving in the same way as a tuna. So, the main thing is that you have to teach the system the specifics of your target operation.

That also brings us back to what we spoke about before: the difference between big data and limited data models. We built limited data models. The limitation is on tuna, the limitation is on salmon, the limitation is on cod, the limitation is on shrimps…

They are all somehow different. They are all in the water, and the way fishing happens is very similar, but there are differences. And the differences make the efficiency.

Ocean Economy AI Lab wants to combine satellite data with field data collected from fishing activities, while also engaging with the local fishing community. How important is this feedback from fishermen in building a reliable Mauritian ocean data ecosystem?

Every data is important. Every feedback is important. So, the more data I have, the better I can be. The more data I have, also, the less important the reliability of each individual piece of data becomes. If I have 10,000 fishermen reporting back, it does not matter if 10 of them are talking nonsense, lying, or just not seeing the right data.

There is the law of large numbers, linked to Bernoulli, which says that if you make a measurement more than 10 times in the same environment, it does not really matter if you make some mistakes in the measurements, because everything over 10 measurements on the same spot already gives you statistical reliability that is very high. That is something that helps.

But I think you also mentioned combining satellite data. That brings us again to some political things. Satellite data is good if it is freely available. I cannot speak specifically about Mauritius, but I guess the situation is not different from other places in the world, where 90% of the data that would be helpful is kept under lock by governments. And from this 90%, I would say 80% is unnecessarily locked. It is just locked because they are either too lazy to publish it, or one person sits there saying, “No, no, we cannot disclose this,” and the others are not saying, “Actually, we do not care.”

Here, I would say that if the consortium between Mauritius and Comoros decided to take government-collected data and make it public, or limited public, for such uses in an easy and free-of-charge manner, that would be extremely helpful.

We can build our own sensors and make our own things, but we build them side by side. I know that ocean ministries and militaries collect this data in masses, and they just pile it up somewhere, and it will never, ever be used. It is something we used to call a write-only memory. You take data, you write it down, and there is never anybody who looks at this data.

We would earnestly urge access to this data, because it would be useful for us. It may not be useful for the military. It may just be garbage, white noise. But for us, it is precious data.

So here, my wish would be that governments engage a little bit proactively and say, “Every data that we do not need, we give to you for recycling.”

A bit like tuna. For Europeans, it is a delicacy. But for Mauritians, it is a poor man’s dish…

Yes, exactly. And in Scandinavia, salmon is the poor people’s dish, while in other places it is regarded as a delicacy.

Beyond fisheries, the Lab’s long-term ambition is to build broader ocean intelligence capabilities and help position Mauritius as a regional reference point in ocean data and AI-driven innovation. Where do you see the strongest opportunities for Mauritius in this field?

That brings me back to 2018, when I was here. After speaking with the government, I developed a concept. I gave it the name Mauritius Digital Twin Island. The big advantage of Mauritius, or of the whole area including Seychelles and Comoros, is that you have a unique capability. You are attached to the ocean.

If Mauritius builds a scientific research forum for the blue economy, for the economy around the ocean, around water, everybody will say, “Yes, of course. Where else would you do it?”

You would do it in Mauritius, Bali, Fiji, Tahiti… in these places where you have a lot of water.

I suggested that there could be an initiative in Mauritius, together with scientific institutes, with the support of the government and international organisations, to build here a centre for research and development around the blue economy.

This would include having a permanent forum researching this topic, and a summit every year where you gather all the experts of the world for the blue economy. Something like what happens in Davos for politics. You could have such an ocean summit in Mauritius, where politicians are invited and made aware.

This brings us back to the initial question: what is the problem with politicians? The problem is that they are not aware. If you bring them to such a big summit, they are all there, and all a sudden they say, “Oh yes, that is something we need to talk about. That is something we need.”

I think Mauritius, along with a small number of similar locations, is predestined to have such a research centre.

So just be the first. There is none of them yet.