AI is different from humans, it does not have a psychological weakness-Lee Sedol, 9-Dan “Go” Player and 18 times world champion

Towards the end of the Tokugawa Shogunate, feudalism in Japan was breathing its last few breaths. The Meiji restoration ( Taisei Hōkan) of imperial rule symbolized the end of an era where Shoguns (masters) ruled their territories and the Samurai (warriors) would give up their lives to protect the Shogun. Today, the owners of algorithms (platforms) are the modern Shogun, the AI algorithms are the equivalent of the Samurai and the data set that is used in the algorithm is like the sword wielded by the Samurai- the Katana. The better the data set, the sharper the edge. Using the analogy of feudal Japan prior to the Meiji restoration, this article takes a futuristic look at a world engaged in an AI arms race where geopolitical boundaries become irrelevant and everyday life is reduced to algorithms.

Setting The Stage

If you want to imagine the future of Artificial Intelligence, start by watching the documentary “AlphaGo” on Netflix. The documentary showcases how a research endeavor based out of London called DeepMind created an AI software that defeated a world champion in the ancient Chinese game “Go” -5–0. The next version of AlphaGo i.e. AlphaGo zero beat its former counterpart 100 games to zero. Go is perhaps the most complex game humans have every played. Go is played on a traditional 19x19 square board by placing a black or a white pieces on each intersection. One player uses the white pieces and the opponent, black. The objective of the game is to surround your enemy and have as many of your pieces on the board as you can. The game of Go is believed to be the oldest board game still in play and it originated in ancient China 2,500 years ago. As per wikipedia, the word “Go” is derived from the full Japanese name igo, which is derived from its Chinese name weiqi (Middle Chinese “hjwɨj-gi”), which roughly translates as “board game of surrounding” or “encircling game”. There are roughly 40 million “Go” players in the world. To give you an idea of the complexity of the game, there are at least 10⁴⁸ moves that can be legally made on a “Go” board which means there are at least 10^(10⁴⁸) possible “Go” game plays that can be legally played. At the opening, in Chess, a player can have 20 different moves. In “Go”, there are 361 such possibilities. Although the rules of “Go” are less complex than those of Chess, Go has many more possible moves than Chess.

The reason I mention the documentary and the game of “Go” is because many people thought that it was impossible for AI to beat humans at “Go” so early. Everyone thought it would take decades because it would require AI to possess human level “intuition” which was hard to program. Demis Hassabis and his team at DeepMind proved otherwise.

The moral of the story is that it is impossible to predict the speed with which AI will achieve superhuman status. We only know we are in the middle of a renaissance as far as research and funding into AI is concerned which makes exponential advancement of AI a high probability. One such example of unpredictability comes from the AlphaGo software itself. When it was playing the world champion Lee Sedol in a game, AlphaGo made a move (move 37 in game 2) that most humans would never have made. Everyone thought it was a crazy move. The fact remains that it wasn’t crazy at all. In fact, it was calculated and appeared foolish. To use a devious analogy, a wolf in sheep’s clothing so to speak.

A Glimpse Into Feudal Japan

Now that I have explained the fact that AI is progressing very fast to reach an Artificial General Intelligence (AGI) stage or a level of intelligence that is equal to humans, let me explain why I chose feudal Japan to be a template for a dystopian vision of a world where AI and humans coexist.

For one, feudal Japan was extremely unequal with a set hierarchy very similar to the possibility that AI algorithms may be unequal or biased. In feudal Japan, the emperor was just a titular head with little power. The power was concentrated in the hands of the Shoguns and Daimyos. A Daimyo was a warlord and owner of land whose power was directly proportional to the power of his private army. If a Daimyo was powerful enough to keep the other Daimyo’s in check, he was given the title of “Shogun” (military leader).

A Samurai was a warrior who protected the Shogun and was willing to lay down his life (commit Seppuku) if he wasn’t able to discharge his duty honorably. The Samurai, however, had a very strict code of conduct called “Bushido” or code of the warrior. To understand the Samurai code, read the book “ Book of The Five Rings” by Miyamoto Musashi.

Secondly, a lot of power was concentrated at the top. Again, this is analogous to the power of platforms today. Some platforms wield out sized power and arguably, more power than nation states. Their power stems from the access to personal data that they have. In my example, the Shogun is the owner of the platform, the Samurai represents the AI algorithm that powers the platform and the Samurai’s weapon i.e. the Katana (sword) represents the data underlying the algorithm.

Thirdly, the people at the bottom were numerous but all they actually did was support the Shoguns and Daimyo’s. This would be very similar to the common man of the future who would not own or have access to algorithms but will support the modern warlords (platforms) with their data.

In a dystopian world of the future, nation states may be like the emperor i.e. titular but with very little power over the platforms. The platforms will be powered by a private army of patented AI algorithms and the weapons of choice in modern combat would be users’ data sets.

Programmed To Be Unequal

Human beings, by nature, are competitive. We end up creating an unequal world. Although there is a distinction between unfair and unequal, today the words are being used interchangeably. Since AI programs are a human creation, it is highly likely that there will be a built in bias and discrimination. A classic example is an article titled “ Biased Algorithms Are Everywhere, and No One Seems to Care” published by the MIT technology review on July 12, 2017.

Many popular books such as “Weapons of Math Destruction” or “Algorithms of Oppression” are already alluding to the fact that biased algorithms can soon turn on its masters and have drastic unintended consequences.

As Lee Sedol — the world champion that was defeated by DeepMind’s AlphaGo program says, AI doesn’t have a psychological weakness. In other words, it may not get bogged down by repeated defeats at the hands of it’s opponent. It will simply do what it is programmed to do. If the algorithm does not have ethics or empathy built in, it is not hard to imagine a scenario where a machine doesn’t hesitate to take human lives. The other possibility is the nature of the algorithm i.e. if the algorithm evolves by learning on it’s own, it can become unpredictable and dangerous.

Platforms Become Feudal Kingdoms

Machine Learning (ML) is a subset of Artificial Intelligence whereas Deep learning breakthroughs are now driving the new AI boom. For lack of a better model to emulate, most competing AI algorithms are using the human brain as a template. It doesn’t mean that future AGI would resemble the human brain but it could be superior to the human brain.

While neuroscience itself doesn’t have all the answers to how the human brain operates, it is working together with other disciplines to sharpen the AI edge. There are many different ways in which machines are being taught. For example, in case of DeepMind, the algorithm behind AlphaGo uses a technique called “Reinforcement Learning”. Under this technique, the computer program learns from its own experience and keeps getting better because of a rewards incentive system built into it. This technique is very similar to teaching kids by giving them candy as an incentive.

Other techniques in play include regression, back propogation, generative adversarial networks etc . broadly classified under supervised and unsupervised machine learning.

As sensors and actuators proliferate and every waking moment of everyone’s lives gets recorded, there will be quintillion bytes of data being generated continuously. This data will feed the AI algorithms which will progressively get better with experience.

For instance, as the video above explains, AlphaGo Zero adopted a blank slate (“tabula rasa”) approach to learning without any human input and quickly discovered things that the human mind could not. Irrespective of the techniques or combinations of techniques used, we are already witnessing a global AI arms race between corporations in China and the USA.

The key to the success of an algorithm is that it should have access to a wide array of global and diverse data which means corporations will have to become transnational or forge alliances to share data sets to hone their algorithms. Ultimately, this requirement would mean the blurring of nation state boundaries and a reversion to looking at humans at a specie level. Data will need to be representative of homo sapiens and not just Indians or Germans or Russians etc.

Looking Outward For Answers

AI would spawn a global arms race where multinational corporations that own the technology behind their platforms would compete for dominance over the lives of the common man making the mega-billionaire owners of these corporations the new overlords.

The more utility a platform offers i.e. the more things a platform offers e.g. banking, grocery delivery, music, payments, data, entertainment etc., the more users it will have.

Naturally, as with other human wars, there will be a battle for supremacy. The best algorithm with access to the most efficient user data set would be crowned king until the next king takes over. This cycle can play out over and over until there emerges a single overlord that keeps all the others in check thus forming the social hierarchy that resembled feudal Japan.

On July 8, 1853, American Commodore Matthew Perry led his four ships into the harbor at Tokyo Bay, seeking to re-establish for the first time in over 200 years regular trade and discourse between Japan and the western world (source: history.state.gov). When Commodore Perry landed in Japan, the Japanese quickly realized that their swords and bows and arrows are no match for the Howitzers or machine guns that could spew bullets in seconds and massacre an entire army very fast. In other words, the western world stood on the twin legs of science and technology with unmatched superiority. These revelations led to Japan opening up to the outside world, its industrialization and emergence as a global superpower using western science and technology.

The Meiji Emperor announced in his 1868 Charter Oath that “Knowledge shall be sought all over the world, and thereby the foundations of imperial rule shall be strengthened.”

In our story, this is a bit like the human race exploring other planets and seeking out alien races when it runs out of answers. If an evil overlord in the form of a platform with the most powerful AI takes over, homo sapiens would need to either leave planet Earth by finding an inhabitable planet which could very well be Mars. Else, our world will quickly resemble Hollywood’s version of Skynet or a post apocalyptic world. From the ashes of this world, a new order will emerge.

Epilogue

All of my arguments above can be pieced together to form an image of a very dark and dystopian future which resembles feudal Japan. However, I remain sanguine. The purpose behind writing this story and using feudal Japan as an analogue was to warn readers about how likely dystopia is and how far we can be from reaching an ideal world. My hope is that common sense will prevail and AI will force us to band together as a species leaving aside our petty differences of race, religion, caste, color and sex. When all else is done, we should all be guided by these beautiful words from Bertrand Russell:

The good life is one inspired by love and guided by knowledge.