Before the industrial revolution in the 18th century began, more than half of all Europeans earned their money working on fields. The steam machine by James Watt changed this dramatically in the 18th and 19th century. A cheaper and more efficient way to cultivate land was found and replaced the workforce. As a response, Humans invested in their education. They went to school longer, specialized in other areas such as construction, handcrafting and much later digital jobs. Assuming an artificial intelligence is cheaper and faster in solving complex tasks, where would we invest in next?
The Cognitive revolution and Artificial Intelligence
70,000 to 35,000 years ago the human brain evolved further which is now called the cognitive revolution. Humans were then able to talk about theoretical constructs and things that could not be directly observed. This made it possible to create a stable society including, rules, a cultural heritage, religion and a social life. Since this time, it has been proven that every generation becomes smarter and smarter, because of better nutrition in early years and of course better schooling. However, the world became and still becomes more specialized. Meaning humans learned specific knowledge and rely on others to produce their food, control their water quality, electricity supply etc. So, from a farmer family having broad knowledge to supply themselves in many regards, people become dependent on others.
What’s the result of specialization in terms of competiveness to an artificial intelligence? Well, we might perform poor. AI’s are actually much better in very specific tasks. They recognize brest cancer, understand texts better than humans and are capable of writing new texts. Bad news for us, right? But what are the restrictions AI algos such as deep neural nets face?
- A huge drawback of such specialized AI’s may be their constraint to the “smaller picture”. The world is changing fast, adopting to every new constraint is an advantage humans have over artificial intelligence. The algos behind AI’s use target functions which need to take into consideration changed objectives and therefore need to be trained again every time.
- The next restriction derives from a main premise for learning machines: Data points. A last resort for humans can be jobs where no data points are produced, such as very specialized handcrafting, individualized for every customer.
- Until now there is no universal AI, but Facebook and other big players introduce new algos every day, becoming broader in their tasks. Google just announced a developer tool helping to self-lern deep neural nets in only a few hours. However, as stated above this does not link the dots in a way humans can. A possible future may be AI’s deriving the insights business analysts did before and the management grounds their decisions on these insights — or machines will also decide and no managers are needed anymore.
- AI’s are not on their own creative. You can teach a neural net the style and technique of a van Gogh, but it can only reproduce this style. On the other hand, this is what most humans do in arts too. There is a new movement and people try to keep up and copy it. In business, this can be a drawback. This plays hand in hand with the first restriction of AI’s. They do not see the bigger picture and can only decide on the data points they have. Some solutions however need further insights.
Additional to the constrains a machine faces, big firms adopt very slowly to a changing environment. The spreading of self-learning machines in companies may take years, or even decades. Especially, if the availability of experts training these models will be kept scarce as it is. At this point no one knows how long these transitions will take, but one is for sure, there will be losers and winners. Google replaced all their old backend optimizations with deep neural nets in the field of search, translation and Youtube tagging.
A last point on AI is it’s capabilities to create new knowledge. As it turns out, many people try to define knowledge, so it is no easy subject to talk about, but I think that AI’s can create knowledge to some degree. But they learn on existing examples and find the optimal way to response to things, seeing the past. Can AI’s find new areas in science itself? I hardly believe that, because they are build using the existing tools. So fields like graph-theory would not have been discovered by an learning machine, but on the other hand, they already developed new languages. No one understood though the decision process for word creation etc. This is impressive and scary at the same time.
Conclusion and the Distribution of Rents
Are we all doomed (or blessed) to become artists and do whatever we like? No, I hardly believe that there will be an even distribution of rents generated by (smart) machines. Meaning only a very few people will participate in higher share prices, or innovation rents made by a very few big players. The run for data sets already began, be it in the health industry or using microphones like Siri, Alexa or Google Assistant to gather voice examples. In the end, there will be people losing their repetitive job. We have seen it already in the past. The industrial revolution left millions unemployed. We merged to new fields but those are threatened by new technologies again. Which profitable fields are left? How will rents be distributed? What about ethical constrains, will decisions always be made by humans? No, algos in cars decide who will live and die…