The new buzz-word: AI

The hype is real. Artificial intelligence, the holy grail of computer science, is receiving a lot attention. AI surely is a lovely buzz word. It has everything a buzz word requires: the underlying technology is great, the word itself highly overstates the technology and the majority of people do not really know what it is. AI based technology surely is very promising, however do not get carried away by its large promises. Calling this new technology is the same as calling Physics “ . The string theory is a comprehensive theory, which is able to explain all physical effects. It literally is the holy grail of physics. We are nowhere near solving intelligence, therefore the current developments are rather a form of . “Artificial Intelligence” String Theory” Advanced Informatics Theories of what involves “intelligence” aside, new technologies move very fast from revolutionary to standard informatics. Think about Snapchat filters, Siri or other revolutionary AI based technology, they do not keep us impressed for a long time. An intelligent being would be able to impress us, time and time again. Therefore, for me I see AI as technology made by really smart people who use mathematics and a lot of computer power. There are a couple other reasons why I think we are hugely overstating when using the buzz-word Artificial Intelligence. The first two are more from a technical perspective, the last one is from a business perspective. We are currently very dependent on supervised learning: we need to tag datasets to be able to learn those tags. Intelligent beings do not learn by being supervised, but they learn by natural feedback from a system. If we want to create really smart beings, unsupervised learning is the way to go. Supervised learning: Google is using our feedback to train their AI. Tables have turned! Google DeepMind has made some huge steps in unsupervised learning. They have shown that AlphaGo Zero is able to excel without human input (or rather human bias) in a game with a huge amount of situations. Being able to switch to unsupervised learning requires a huge change in mindset for humans. We are used to be in charge and think that our inputs are very valuable, but we might be very biased, narrowly focused and not able to think out of the box. Even in the hottest tech, people will always try to relate to human experience. They will hard code stop signs instead of looking at untold relations in the data between cars stopping and this huge red sign. We seem to be very bad at letting go. he data we currently use to train AI is extremely structured. Almost all our databases are relational. All pictures have to be resized to certain sizes to be trainable. We often deal with missing data by substituting them with data averages, to not make them stand out. We really seek for structuring data to be able to learn patterns. Structured data: T In the real world, data input is extremely unstructured (or sparse, which is low amount of active signals compared to all possible signals). Brains can easily catch these sparse signals and process only the related parts of the brain. Our current neural network structure does not allow for partial activation (and training) or missing data. The models we train are also very rigid. When a brain discovers a new signal, it can quickly adapt to grasp its implications and in its turn activate other neurons to establish this relationship. Neural networks have difficulty with new inputs and will need to train the whole network to deal with this. The gap between AI and regular business (non-tech) is still enormous. Businesses tend to have loads of data these days as they got the whole “we need big data” hype. They probably felt relieved, thinking they catched the big data train. However it hurts to see that most businesses are still ran from Excel spreadsheets, which is not able to capture relationships hidden in big data. Companies are not ready: If companies are not even able to reap the benefits of proper data analytics, how can they ever be ready to reap those which AI will bring them? Okay, probably you are working at a tech start-up which is ready for AI, but 95% of the companies are surely not able to reap the benefits. Even if the data is available, most of their data is read only, thus AI is not integrate-able with their software. Another problems is that people are often blinded by their own bias of working somewhere for 20 years or even worse: they do not really care about improvements. This is where the cockroaches come in: the consultants. These people love to earn some money by leveraging buzz-words. Often they are seen as the most capable people by businesses and are therefore the go-to option when a CTO is reading about AI. It allows them to say: we are really looking into AI. It gives a false sense of comfort and I see it a lot. When people are not able to grasp a concept or problem, they will calm themselves by saying: they will handle it for me. The problems is: these consultants do not really sell solutions, they sell your own biggest dreams. Artificial intelligence is great, but we are not there yet. The technology is not as scary as some people might think, as it is only able to excel at a small set of skills due to its supervised nature. Furthermore, the gap between businesses and AI is still very large, making it difficult to reap its benefits. So there is plenty room for great ideas, fellow AI enthusiasts! 👏 Thank you for reading my article. Clap it up if you like my articles: Winner take all markets and Threats of Neuralink . Feel free to contact me, I am always open for discussion! Some tips for businesses who want to go down the AI road: Get your data structured properly. Locate your data so it can be easily used in a wide variety of tools. Get the right mindset/people to leverage data to explain behaviors, test hypothesis and eventually improve business. Allow people to criticize and improve every part of your business. Pick software with an integration of AI engines (personal recommendations in e-mails) and allow for feedback loop in data (stock management). Do your data analysis in house. Get someone with experience in Python/R and enable them to be critical. If you really want to develop AI in-house. Look at how companies like Uber structures their pipeline to be able to feedback analytics into their eco-system.