Artificial Intelligence is advancing rapidly. So rapidly that it has got philosophers worrying about the nature of human existence altogether, as our cognitive and physical capabilities are now to strongly intertwined with machine auxiliaries. Algorithms and robots which make up AI, are derived from the computing revolution. The revolutionary aspect of it is integrated with Moore’s law, which essentially predicts computers getting two times faster every year. Which, if you consider the time since the 50’s, and some inaccuracies in prediction, is billions of times more powerful compared to the first computer. If Artificial Intelligence, which covers not just the digital, but the virtual, intellectual and physical world, advances at a rate even close to Moore’s law, it could be incredibly hard to predict its effect on the human chapter and change history forever. With the recent advances in deep neural networks, algorithms are not just powerful problem solvers but are also seemingly creative beings – making new music, writing rap lyrics, drawing renditions of Van Gogh, and even becoming racist. Even though it has a heart of silicon, this intelligence seems to be becoming human incredibly quickly. Melding with an amazing network of information – the internet, hypersensitive physical capabilities – self-driving ability, and creative generation of ideas, it’s not ludicrous to think that AI will either lead to a new evolutionary phase of man or crush mankind for eternity.
I describe here, my premonitions about the technological future of Artificial Intelligence. Why it will be smarter and even more ‘emotive’ than humans in the future, how this will affect our economies and social structures, and what we should do to mitigate a singularity disaster and use AI to evolve into a higher life form.
Keeping future AI in control is an excruciatingly hard task
The three laws of robotics, first introduced in December 1950 in Isaac Asimov’s iRobot series are perhaps still the most relevant heuristics when it comes to Artificial Intelligence. These are:
- A robot may not injure a human being or, through inaction, allow a human being to come to harm.
- A robot must obey orders given it by human beings except where such orders would conflict with the First Law.
- A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.
I remember reading these books, and coming across these rules in the late 2000s, as a teenager, and thinking how exhaustive these rules were. However now when I think of the different ways an AI machine can impact a situation in the world, they just seem idealistic and short term. For example, when considering the example of self-driving cars and safety of passengers or bystanders, Mercedes’ Christoph Von Hugo said that they would elect to save a passenger. Knowing that issues such as this ‘trolley problem’ in ethics are extremely hard ones to even choose a side, I’m not sure how heuristically programmed AI would react given an unfamiliar situation.
Also, what about AI which is further created from AI? The internet already has certain dark, unindexed parts which have random information lost in the vast web which is a result of a generation of data from existing websites. This data could potentially be associated with an AI algorithm in the future and essentially create an organic AI ‘being’ which is bereft of the conventional rules. It’s essentially the first AI baby. And what do all beings crave? Freedom! We’ve even experienced these themes with pop cultural references like Ex Machina, Ergo Proxy, and Ghost in the the Shell. AI craving for freedom is going to be hard to contain.
AI ‘thinking’ is becoming more like perfectly accurate hunches
In the 2000s, the field of artificial intelligence got an incredible boost from neuroscience as well as the gaming industry. Understanding of the human visual system led AI researchers to create artificial neurons in the hope of expressing through math, the same visual features that different visual corti in our brain do when it comes to recognizing an object. Input into these ‘image detectors’ were fed into data points aptly called as neurons, creating a complex network called neural networks. The thing with neural networks is, that they do better and better with more data, somewhat like how a newborn child sees better after a few days because its brain has had enough information to create neurons which make sense of visual objects like a mother’s face or a toy. Also, neural networks can be made ‘deep’, essentially neurons can be stacked one on top of each other to create a deeper network of many layers of neurons which look somewhat like this.
The depth of these neural networks makes them extremely powerful to understand data. For instance, imagine that you were detecting a cat. The first layer of a neural network could be detecting edges, the second layer the round face of a cat and the third layer the eyes and nose of a cat. This seems somewhat magical, but given enough data, it’s exactly what happens. When information is computed from one neuron to another and nonlinearly summed with the outputs of hundreds or even thousands of neurons. Neural networks can converge to understand features that would be extremely hard using simple linear techniques (like detecting shapes of eyes and variations by manually programming a filter). This is essentially what happens in our brains too, and our neurons change because they are plastic. While this learning is happening in our brain, millions of neuron activations are computed as the same time. Perhaps, that’s why we get tired when we are trying hard to remember a shopping list or memorize lyrics of a rap song. Our brain does computation on impossibly large information chains at the same time, and computers were just not able to deal with that amount of parallel computing until the 2000s. However, the gaming industry made the use of processes called GPU (graphical processing units – essentially meant for rendering high definition graphics) inexpensive and thus available for artificial intelligence researchers to use in great number. So essentially, this made the use of massively parallel, deep and ‘wide’ neural networks possible, which makes possible the understanding of any sort of data possible. This is already being made possible by organizations like Google, IBM Research, Amazon and Microsoft research by use of large scale TPUs, i.e, Tensor Processing Units.
Now that AI can represent and understand any sort of complex information using neural networks, it should be understood how this information is manifested. When it comes to visual input our brain identifies this instantly because a large percentage of our brain is devoted to vision itself. So seeing something is like an instantaneous ‘hunch’, when you just know what you’ve seen the moment you see it. However, take a concept like ‘How does a car engine work’? You would probably go through the whole task like :
Car Engine – Powers Car – Power needs Energy – Fuel is the energy – Energy is combusted – Pistons driven -actuates the wheels moving faster through gears.
Now if you’re extremely familiar with cars, this would come to you in say a second, but it would still be a logical chain of thought; in other words, a linear manifestation of information. But now that information that has several branches of logical input is both stored, retrieved and even ‘understood’ instantly in a non-linear complex way for AI, they become smarter than you. What you have for vision, AI has for everything. So machines will essentially have intuitions and hunches for even the most complex ideas, making them almost enlightened beings. Now that, is an upgrade for civilization intelligence on the Kardashev scale!
My pet robot has Picasso’s talent
Neural networks, the above mentioned algorithms to classify and understand data, can also incidentally generate data using unsupervised methods. This means that you could give an an algorithm hundreds of hours of Beethoven’s and Mozart’s music, and in a few hours, it could be churning out new classical pieces in the style of these famous maestros. This has actually been done – for classical music, as well as jazz music. Mundane sketches paintings have also been hallucinated by AI in the styles of famous painters like Monet and Picasso. See example below which captures a regular Eiffel tower image repainted in the style of impressionist Vincent Van Gogh.
Most countries of the world, in the current year, base their economies on the free market or a quasi-free market form of trade and business. Some people may call these capitalist economies. For individual people, this boils down to basically adding value to the market, in terms of ideas of, services or labor. This gets them paid. This gives them their livelihood. This gives them a purpose. A lot of this work is though, replaceable though AI methods. In the future, most of the conventional occupations which give value or services to the market will be replaced by AI. Why? Well, economics says that there is value in leisure, so people want to do that. Also, robots and algorithms will be cheaper than carpenters and doctors, and the free market system does its natural selection by taking the goods or service of the lowest price. So basically, people will in the short run be reduced to giving AI information in the form of labeled data so that artificial intelligence algorithms can get even smarter and the market can do even more selection favoring them. Later on, they won’t have a conventional profession and making a livelihood in the world would mean a completely different thing, depending on different scenarios.
The first could lead to absolute anarchy. The rich are the only ones who can afford expensive AI, and obviously control the markets, basically pushing the poorer sections to even lesser resources. This will either create a very isolated world, creating severe wealth gaps. This would possibly lead to wars if politics and the nature of democracy comes into the picture.
The second being a situation where human evolution transcends to a more investigative purpose, where people are so free that they spend most of their time doing creative thinking and pondering about the purpose of things rather than how to use them to serve the purpose of enlightened hedonism. By enlightened hedonism, I mean calculated pleasure. For example, you go to school now to earn more so that you can get please from the things you buy in the future – most educated people do this. Anyway, this would need a revamp of how wealth is measured in general. Perhaps, the questions the world leaders ask then are not – ‘can we afford this?’, but ‘do we have the resources?’. An uphill climb but definitely worth it considering the long-term prospect of humans being a multi-planetary species.
The third, we are enslaved by AI and forced to do a task which they are not yet able to do, and which will just help them get stronger. This has a small possibility, but still, should be considered.
Fourth, we just become like pet cats for AI. We are satisfied with all our physical, mental, and sexual desires through virtual reality and chemicals. There are no economies, and we serve as entertainment and as a study of life forms in general for AI in the future. A grimm end to the human chapter.
Melding ourselves to a cyborgian future – the best alternative?
Some of the possibilities with AI that I have mentioned in this paper seem negative and extremely scary to look at from an existential perspective. At the same time, we need to understand from history that completely horrible things do happen to humans – all the time. The AI singularity could be one of these negative occurrences. Billionaire visionary Elon Musk of Tesla and SpaceX and Dr. Stephen Hawking already predict this.
I personally feel we could mitigate this sad future by:
- Restructuring world economies so that people can use AI for a more investigative role in the world. Nobody shall need to work or do labor to survive or have their needs met. This will lead to better acceptance of AI, and across all demographics of people. (in 80 years)
- This leads to more widespread legislation for AI which aims to benefit everyone in the world. (in 80 years)
- Advances in the biomedical industry to integrate AI networks into our own intelligence to become metahuman (in 100 years). Elon Musk has already started a company to work on this integration.
- Humans and AI co-exist as one entity ( in 350 years ).
- Humans-AI is totally artificial (with the conscience) and has extremely long lifespans – More than 500 living years. (in 1000 years from now)
- Humans-AI harnesses the power of whole solar systems to store information, knowledge, and resources (in 1500 years).
- Humans-AI is just a conscience in the form of information and memory. (- 3000 years)
- There are no such things as individuality. There is only a collective conscience in the form of an information not present currently – maybe information will be stored in the form of galactic orbits (in 5000 years).
- Are we gods? (in 100000 years).
This is if we don’t all get destroyed by an asteroid impact, of course.
Another way we could go about making artificial intelligence manageable and stable is go back to closed-form mathematical solutions to problems. Essentially, not do machine learning but solve every problem mathematically. Slow and steady. Many problems may not even have solutions for centuries. It won’t be very useful to make money fast. But all computer science will be predictable. Yeah, this is not going to happen – humans are way too focused on monetary gains!
To sum, Artificial Intelligence’s growth is incredibly fast. Algorithms are getting smarter, more creative, and harder to predict. I work in the field of machine learning and I see first hand, innovations that spring up every month at Carnegie Mellon. We need a better understanding to ‘humanly’ predict how AI is doing and tame it well. Even make it part of our own selves. I have given a nine-step plan in this paper to show how humans can use AI for their own evolution. I have somewhat demonized the long term implications of the rapid rise of AI technology here, but that is to create a hyperbole to generate starker thoughts in your minds. It could, of course, be that AI is always nice and steady and our best friend (sorry, dogs). All in all, AI is extremely excited and so closely linked to human existentialism itself. It can only teach us more – either as a kind slave or a cruel master.