Artificial Intelligence: The Journey into the Unknown - PART I

OP-ED: Vidya Munde-Müller and Sascha Lambert

We are in the middle of serious debate about human history approaching ‘singularity’. This is not science fiction anymore. AI technology has the potential to reshape our society within the next few decades. The reshaping already started with more and more automation that finds way into our workday life, e.g. voice assistants always available inside our smartphones. The authors of this op-ed intend to give their understanding of singularity and what kind of future may befall on humanity. Rather than choosing one or the other future, the authors want to show different possibilities of AI evolution using the lessons of human evolution.

From Natural to Artificial Intelligence

Human Evolution

It has been a human dream since ancient times to create machines that are equal to us or even surpass us. Literature references from ancient Egypt around 1100 BC, are filled with ideas about artificial humans. In greek mythology there is a bronze man named Talos, a giant automaton or self-operating machine. The human evolution itself offers some clues about this subject. Nature succeeded in creating and sustaining life on Earth. This remarkable journey forms the basis of our thesis on the way AI can develop in future.

Life began around 3.8 billion years ago on Earth with the appearance of microorganisms found in hydrothermal waters. On the evolutionary timescale the complex life began with some major milestones: eukaryotes, vertebrates, primates and homo sapiens. The real ‘human’ evolution started around 70.000 years ago with a ‘cognitive revolution’ which brought us fictive language or the imagined order. Imagined order meant that humans could not only perceive, act or think about their everyday world but they could imagine abstract concepts like money, law and countries. Transfer of what has been learned from biology and behavioral science into the language of mathematics, paved the way for the digital world. Biological nerve cells (neurons) form something like the smallest functioning unit of our consciousness. Each neuron works on its own like a filter and receives chemical and electrical signals and passes them on or off depending on a stimulus threshold. Through the interconnection of infinitely many such units or neurons, our complex brain structures developed in the course of evolution, which led to a kind of self-esteem, to dreams, to morality, to complex societies, and so on. Our brains evolved to plan and carry out complex actions and cooperate with others on a huge scale in form of companies and social groups. This is a criteria that many anthropologists see as the major difference between humans and animals and why we made the evolutionary race to the pride of creation.

The result is therefore apparently much more than just the sum of the individual components. What exactly is the magic behind it? In fact, we still do not know it today. What is consciousness, do we have a soul? How come? What is the big plan in the universe? Everything just pure coincidence or is there more behind it? A great riddle that leads people time and again to new spiritual masterpieces. But is that always good? We shed some light on it in the next chapters.

Comparing AI vs. Human Evolution

An important note to the readers: The following metaphors of comparison are only used to make the different steps in AI evolution clear to the reader. This comparison is in no way an exact science and only meant for better comprehension of the topic. The evolutionary milestones are used in broader terms and are not based on any detailed inspection. For simplicity we will only compare these milestones to the AI evolution: eukaryotes, vertebrates, primates and homo sapiens.

Image Credits: Shutterstock

First Milestone: Birth of AI

AI can be dated back 60+ years ago starting with the Dartmouth Summer Research Project on AI in 1956. There was a workshop held on the campus of the Dartmouth college and those who attended eventually became leaders of AI research. Although hyped at the beginning, the AI cycle was followed by long periods of disillusionment summarized as a period ‘AI winter’. This AI milestone comes close to the evolutionary milestone of Eukaryotes. Eukaryota were cells with a real nucleus and came into existence around 2.100 million years ago. An eukaryotic life-form was much better prepared to be organized in different organs and specialized tissue. Its level of organization is much higher than that of a prokaryotic life-form. With that as a turning point in evolution it took life on earth to a much more complex forms.

Image Credits: Dartmouth Researchers via Forbes.com, Eukaryotes Cells via Wikipedia.org

Second Milestone: Handcrafted Knowledge

The first wave of AI brought handcrafted knowledge or rules-based approach. Chess playing algorithms, customer self-service platforms and automated logistics scheduling programs are examples of this wave. These systems are in fixed domains and are good at reasoning and a little perceiving their environment but have no learning and abstracting component at all. We could take an example of IBM Deep Blue which defeated Gary Kasparov in chess. All it could do was to play chess. Even though the algorithm may be a digital grand master in the example of chess, it is completely lost in any other area, like cooking coffee.

This sort of AI evolution comes closer to the Vertebrates evolution in terms of the milestones. Vertebrates come around 505 million years ago. These are animals with a spine like the fire salamander. All vertebrates share a vertebral column or what we know as a spinal cord for humans.

Image Credits: IBM Blue via Wikimedia Commons, Salamander via nwf.org

Third Milestone: Deep Learning ‘Specialized’ Algorithms

Today we see weak AI. This type of AI (in most cases) uses huge datasets and compute power to achieve impressive outcomes like pattern-matching or computer vision. Ultimately, everything is just simulations. This works amazingly well in certain areas, but has nothing to do with real thinking and feeling, as we humans do. Examples are Facebook feed, Amazon Alexa, Netflix, Self-driving cars. But even if these algorithms or machines are let’s say digital zombies, with a type of brain structure like we have it, they can outperform humans. It’s true for chess, AlphaGo and other domains, not just in the gaming sector. The performance of algorithms has already achieved a tie in many disciplines. Algorithms are equally good when compared to human capabilities in recognizing faces. They are already on ‘eye’ level when it comes to recognition of written and spoken text. Of course, the algorithms do not really know that they are playing chess or are reading some document but to solve problems or to increase the level of automation in a given process it’s absolutely enough. The mystery to unravel is to find the magic component that makes these Zombies really think. Today we don’t know if this magic component even exists and can be found at all.

This third milestone of Deep Learning can be compared to the evolution of primates in the nature. Humans developed on earth from now extinct primates. Even though Primates are of course smarter than any AI today!

Image Credits: Amazon Alexa via Amazon, Primates via Discovermagazine

Fourth Milestone: Deep Learning ‘Cognitive’ Algorithms

So if we want to understand how the future of AI may look like, the best way is to try to compare the development to our own evolution from Neanderthals. The Neanderthals though very remarkable in many ways could not compare to the brain power of homo sapiens. They behaved not very different to the animals in the animal kingdom. But they did develop tools for hunting and showed signs of symbolic behavior like creating art.

This is similar to the third wave of AI (also called mixed approach) that can construct explanatory models of real world phenomena. In broad terms it could signify animal-like intelligence. If Yoshua Bengio, a deep learning pioneer is to be believed, our interaction with computers will become more conversational in nature as computers comprehend and contextualize information, picking on gestures and emotions. This will enable new smart personal assistants or personalized health care in the relatively short-term future. Not only this, machines can also start to do planning which is really a difficult skill. This has been shown in the labs of Juergen Schmidhuber, who is sometimes called the ‘Father of AI’ due to his work in the field. Machines can cook meals or deliver parcels showing the necessary planning skills. Also we will have self-driving cars as a result of this kind of cognition - where AI can not only perceive and reason but also learn and build abstract concepts. We would not call a self-driving car on its own to be very cognitive compared to human thinking but imagine a totally interlinked network of vehicles, traffic infrastructure and individuals wearing sensors on their body. Take all this data and learn from it to make traffic more efficient and most important safer than today. The whole system acts as one and learns to deal with new situations as well and adapts for future occurrences.

Arts and creativity are thought of a human only domains as well. Researchers showed recently that this hypothesis may not hold in the future. Algorithms can draw paintings and compose music that is hardly distinguishable from human work.

Image Credits: Self-driving car via Shutterstock, Neanderthal Museum via Bahn.de

Future of AI

Thus far, we have compared the AI evolution to human evolution and have come till the current status-quo i.e. the Artificial Narrow Intelligence. This is broadly also called the first wave of AI and as shown with some examples like a chess playing computer or Amazon Alexa, it is based on specialized algorithms. This wave is still ongoing and is in process of giving us self-driving cars and more intelligent personal assistants. In the next chapter we will go deeper into the next milestones of AI evolution. The second wave which is based in future is the complete game-changer for AI but it also the most unpredictable. That is why the next chapter is called the ‘Journey into the Unknown’. We just don’t know what kind of world we will be living in few decades.

Just imagine the big difference in time when comparing natural evolution to the digital one. It took millions of years until our modern societies evolved from some single-celled organisms. When we take the Dartmouth event as the birth of AI it is only about 70 years from then. That is nothing compared to the evolution according to Darwin. In this tiny period of time we created algorithms that led to Alexa, Alpha-Go and self-driving cars. Lo and behold! Will it take another 70 years to reveal something incredibly unbelievable or will it take like in our evolution thousands of years to create really thinking digital beings…or is it even impossible?

Image Credits: Alexa via Amazon, Robot via Unsplash | Alex Knight, Ex-Machina motion picture (Universal Pictures International Germany GmbH). Pictures should just symbolize potential evolutionary steps of AI.

In the next article we will go deeper into the 'Unknown'. Stay tuned!

------------------------------------------------------------------------------------

About the Authors:

Vidya Munde-Müller is the Founder of Givetastic.org (Giving. Made Fantastic) and Women in AI Ambassador for Frankfurt, Germany

Sascha Lambert is the Business Owner of Artificial Intelligence at Deutsche Telekom IT and Co-lead of AI Community at Deutsche Telekom