AI, Philosophy, Sci-Fi & Start-Ups: A Conversation with Simon Hegelich

Posted on 15 November 2022

​I had the pleasure of speaking with Dr. Simon Hegelich - a professor at the Technical University of Munich. Who is also a co-founder of the Portuguese based start-up ddductr.

The first time we spoke our conversation was heavily influenced by our shared love for Science Fiction. Along with the AI Simon is building at his start-up company, the Hegelmaschine. After hearing about his company and what he was building I had to learn more. Especially because his AI has been heavily influenced by the world of Fiction and Philosophy.

We discussed how the field of Natural Language Processing has changed over time, the challenges one faces when building a start-up and – naturally – the influence of Science Fiction on AI.

What sparked your interest in AI?

Well, I think I was always interested in AI because of Science Fiction literature. I’m a big fan of Sci-Fi and suddenly in the 2000’s I realised that there were many things you could do with technology, that were not so far away from what we imagined in Sci-Fi literature.

Why did you choose the field of NLP?

What really drives me is that I would like to build a machine that can think in a logical way. But not just in mathematical binary logic, but like a human would think. Of course, humans do a lot of other things as well like feeling and having a consciousness. But thinking is based on language.

There is this old thesis of Herbert Simon and Alex Newell’s. That any symbolic system that is physically grounded would be a sufficient condition for general artificial intelligence. With language we have symbolic systems that are physically grounded. So, based off this thesis, I think this is the best place to start when building a logical, thinking machine.

How has the field of NLP changed since you joined it?

Natural Language Processing is probably one of the fastest developing fields in AI. When I started, it was still based on everything that could be done with word counts and back of word approaches like topic modelling. And this is also very interesting, but it was far away from getting any kind of semantic meaning. Even simple tasks in real NLP like identifying a verb or identifying a subject predicate object were still very demanding. But then came Deep Learning and with Deep Learning we now have very clever algorithms that can process huge amounts of data. In this data find patterns that we could not even spot as humans. This is a very strong tool and has really impacted the field.

What inspired you to start building the Hegelmaschine, which is what you are working on at your start-up, along with what drove you to embark on your start-up journey?

Well, I have several interests besides computers one of them being philosophy. Hegel’s The Science of Logic really changed my life, because it encourages you to start thinking about thinking and this is very interesting.

You start recognising what kind of mistakes you make in your own thoughts, when you get into a loop and come to a further level. But most importantly I learnt that everything is contradictory in the end. We are always looking for the truth, whatever that is. But truth is a very dialectic and complicated concept.

Truth is not in the word; we can’t just take the truth in the outside world and have it in our minds. We must create it in a dialectic process. Then at the same time I was thinking about artificial general intelligence. One of the biggest problems that we have today, is that people working in this field are not really thinking about what thinking is.

They are thinking about putting more layers in their networks, or they are thinking about what the whole human is. Like this theory that you cannot have a brain in a tube but you have to have a humanoid android walking around and living with us in order to reach a state of AI general intelligence.

I think we should focus on a sub-problem and the sub-problem is real thought and the thinking process. This was all described in detail I’d never read before in ‘The Science of Logic’.

What is the purpose of the Hegelmaschine?

There are several goals that we tried to achieve with the Hegelmaschine. The final one is to have a computer that can create new objective thoughts. For example, writing scientific articles that find something new in the world based off new research. Rather than just something that sounds like a scientific article.

This is a long way to go, but on the way there the whole process of Hegels logic, dialectics and negation will be very very useful. Especially in any situation where humans have to make decisions based on a lot of information.

Because every decision maker now has the problem where there is so much information out there. What AI is doing today is reducing the complexity so that you just get a summary. But you don’t want to make your decisions based on a summary. Because when you make really hard decisions you should be very well informed. You cannot reduce information more and more. I think we should go the other way, where we increase the complexity for difficult decisions.

This can be done with the Hegelmaschine. So that you can have an artificial partner who is saying “no, your idea is totally wrong” it’s the other way round, its contradictory. And this might lead to new ideas and stimulate you to start thinking about your own ideas. So yeah, this is something that we are close to achieving in our company. A kind of artificial discussion agent that is not saying what you want to hear but is developing new thoughts that go in a new direction.

So, the AI will be challenging you to think of something fresh and bring that extra perspective?

Yes.

Another question that people would love to hear from you is what advice do you give to people who want to found their own start-up?

Don’t do it in Germany…. But seriously, I work with many people who work in start-ups or fund their own start up and I think the times have changed. Let’s say five to ten years ago everyone was thinking: “oh I have this great idea I’ll make a start-up and then Google will buy it and I will become a millionaire”. This is not the way it works any more, or perhaps it never worked this way at all. It’s very important that you think about slow growth strategies, what you can do and achieve on your own in your start-up. Where do you need help? What do you need money for? You don’t need millions or billions of dollars to run an AI start-up, because hardware is not so expensive anymore. Of course, this is all depending on what you are doing.

So, I think my advice would be stop thinking from the end point. About how brilliant everything could be and how much money you can get from Google. But rather, think about what your next steps are, what can you do and how can you build it up slowly. Because I made so many mistakes with my start-up and because it was so small it was not so dangerous. But if I had millions in the beginning, I probably would have ruined everything.

What are some of the challenges you have faced getting your start-up off the ground?

One big challenge was that I normally trust people a lot. Then we started to do our first things in the start-up. Selling algorithms and what not. The atmosphere with our partners was always very nice, they would always say that “we are one big family”. But then they didn’t pay our bills. Lots of the things were just handshake contracts and not real contracts and this made it difficult. So, my advice would be to integrate a good lawyer from the beginning. And even if it is annoying someone stick to the rules, ask for them to sign an NDA. Make contracts that are binding, wait till the money is there before you start working.

I had this question as an optional one and you sort of opened it up a little bit at the start. The last time we spoke we discussed literature and Science Fiction in the world of AI. So, I wanted to ask how the world of fiction has influenced the product you are building today?

First, from a historical perspective the first science fiction novels that were published in German, were edited by the philosopher Gotthard Gunther who worked in the McCulloch AI institute in the USA. He was probably the best Hegel expert living on the planet at this time. He thought about science fiction as a new meta-physics.

Of course, since then many things have happened and many things in AI literature – science fiction literature about AI – are not really about AI. But more religious beliefs of a being that is beyond all our imaginations.

But on the other hand, I like the idea of how an AI would react to us if it’s superior in our thinking? An AI might just simply ignore us because, to it, we are boring. So, this is how our product has been influenced by fiction.

I feel like books and Science fiction have been a big theme here, so I was wondering if you could recommend a book or a film that the reader of this article should check out?

I’m a big fan of William Gibson. His latest novels are more about the concept of time travelling and several futures, but he is always brilliant in picking up trends in technology in his novels. Especially the book Pattern Recognition. In another one of his novels (Zero History) you have this concept of the ugliest t-shirt in the world which cuts you automatically out of every facial recognition algorithm. It’s really cool, and I think that today we already have this technology. I saw an article on this recently, but the book was written ten years ago.

It’s interesting how people writing these books decades ago often end up predicting technologies that are being created nowadays.

A big Thank You to Simon -

Simon and his brother Kolja co-founded ddductr in 2014 and have been steadily building their Hegelmaschine since. Their company is based in Portugal and you can find out more about them via their website here.

Simon has also created a short movie with all the pictures are made by AI. The story has been written by GPT-2 and you can watch it here.

Simon lives in Munich with his dog Trotsky (who joined our chat towards the end).

Share this article