The answer you’ve been searching for, in a nutshell.
This article will take you through a journey inside the mind of a person who has no/little experience when it comes to this topic before gaining a (somewhat) comprehensive understanding of ‘Artificial Intelligence’ (AI).
This journey will be broken down into 3 different parts:
Part (1): How will AI intelligence supersede human intelligence
Part (2): When will AI intelligence overtake the human race
Part(3): What if a digital superintelligence has successfully been created and for some reason, it wants to take over the world. Will it be able to do so?
FYI: This article focuses on Part (1).
Before we start, I would like to clear a general misconception about the concept of AI. As nicely summarised by Tim urban:
Stop thinking of robots. A robot is a container for AI, sometimes mimicking the human form, sometimes not — but the AI itself is the computer inside the robot. AI is the brain, and the robot is its body — if it even has a body. For example, the software and data behind Siri are AI, the woman’s voice we hear is a personification of that AI, and there’s no robot involved at all.
How did it all start?
The term “artificial intelligence” was coined in 1956 by John McCarthy, a researcher who later founded AI labs at MIT and Stanford.
In the early 1950s, the study of “thinking machines” had various names like cybernetics, automata theory, and information processing. McCarthy wanted a new, neutral umbrella term that could collect and organize these disparate research efforts into a single field focused on developing machines that could simulate every aspect of intelligence.
During the early days, the pioneers of AI did not believe that machines could behave intelligently and definitely did not consider the possibility that machines will eventually far surpass all the intellectual activities of any man!
However, when the results proved to be astonishing– computers being able to solve numerical problems, invent mathematical proofs that were more elegant than the original, follow instructions and answer questions in English, organizations like DARPA poured tens of millions of dollars into AI projects at MIT, Carnegie Mellon, Stanford!
Over the years, as the cost of computing started declining and processing power got more powerful, AI is now able to run a more complex algorithm on more data than ever!
What are the implications?
AI can now outperform human intelligence in many domains! You may be familiar with some of these ‘AI beats human champions’ events, which marked a significant advancement that happened far sooner than experts expected!
& the hype goes on. I.e. In 2019, AI triumphs against the world’s top pro team in strategy game Dota 2…..
What hasn’t AI taken over the human race yet?
In this article, I will attempt to break down this million-dollar question by unveiling the different factors mentioned in Bostrom’s Superintelligence book:
- AI’s technical limitations
- Different paths that lead to Superintelligence (Part 1)
This will allow you to form your final opinions/conclusion about the future of AI.
AI’s technical challenges
While AI has managed to beat human champions in multiple domains (Chess, dota), one might have thought that AI has mastered a high level of general intelligence to learn abstract concepts, think cleverly about strategies, compose flexible plans and make a wide range of logical deductions.
You are wrong.
While AI succeeded in doing essentially everything that requires thinking, AI still lacks what most 10-year-olds possess: Ordinary Common Sense.
However, AI can only perform well in a specific well-defined task and that is why IBM’s Deep Blue built around a chess-specific algorithm won’t be able to defeat the ‘GO’ champions because it is not programmed to do so!
HEY AI, IS THE MILK CARTON FULL? IF I PUT MY SWEETS IN A JAR, WILL IT STILL BE THERE TOMORROW?
So unless someone were to succeed in creating an AI that could understand common sense as well as a human being, they would have succeeded in creating an AI that could do everything that human intelligence can do.
When will human-level machine learning be attained?
No one has a clue!
Well, Bostrom did cite a survey conducted with a bunch of AI researches, and these were their opinions:
10% chance of happening in 2030
50% chance of happening in 2050
90% chance of happening in 2100
It could take another decade or next hundred years, no one knows! Regardless, now you know your knowledge of AI progression is not outdated.
2.0. Different paths that lead to Superintelligence
2.1. Artificial Intelligence
Traditional computers relied on human beings to tell them what to do and how to react. ‘Artificial Intelligence’ means equipping machines with the power to make its own decision like human beings.
How to give them power?
Just like how human beings store the information in their brain and learn from their patterns, scientists have also been able to use the stored machine information to make machines learn from them. This method of training computers is famously known as ‘Machine Learning’.
When our human brain sees this animal, we can immediately absorb these complex information and label it as a ‘Seal’. How about if we wanted a computer to do the same task, to classify (label) photos as seal/not-seal?
It’s not that simple (As explained in the earlier part of my article).
Essentially, machine learning trains the AI to function as an “Object Labeler “ by showing the computer a bunch of cute seals pictures so that it can figure out what is a seal/not a seal.
If humans took decades of learning to train our human brain, what about machines?
Currently, AI is not able to adapt to new situations without human assistance. I.e. An AI trained in chess is not able to win a dota game.
Food for thought: SEED AI
In 1950, Alan Turning came up with the notion of “child machine” where instead of trying to produce a program to stimulate the adult mind, why not train an AI to stimulate a child’s mind?
This brings us to the concept of ‘Seed AI.’ To put it simply, Seed AI would have to possess ‘recursive self-improvement’ capabilities where it would be able to iteratively improves itself by recursively rewriting its own source code without human intervention.
A Seed AI could start off with a relatively low level of intelligence. However, if it is intelligent enough to rewrite its source code to become more intelligent (that is, to become better at achieving its goal), as a result, it could become even better at rewriting its own source code to become even more intelligent. This could lead to an intelligence explosion, where the AI rapidly becomes superintelligent!
As far as we know, nobody has built a Seed AI yet. I’m not going to dive into the full discussion of the debates, but this concept is already interesting to think about.
Instead of training the AI from scratch….What if we downloaded our human brain into a thumb drive and plug it into our computer?
This brings us to the next possible path to superintelligence…
2.2. Whole brain emulation
Whole brain emulation involves scanning and closely modeling the computational structure of a biological brain to produce intelligent software.
This method totally reminds me of the show ‘Altered Carbon’ where the body no longer matters. As one character quipped: “You shed it like a snake sheds its skin.” That’s because the human consciousness has been digitized, and can be moved between bodies — both real and synthetic.
Just showing this diagram to illustrate the complexity of this method. Guess my order ain’t coming anytime soon.
How far are we currently from achieving a whole human brain emulation?
Well, no brains have been emulated yet. However, when the Bostrom wrote his Superintelligence book in 2014, the emulation path was believed to take another 10–15 years to gain some traction as there are several challenging technologies (Algorithm & Supercomputers) that have yet to be developed!
Algorithm? Scientist cracked it!
In 2018, an amazing breakthrough happened as Scientists successfully created an algorithm capable of performing a complete human brain simulation by simulating the brain’s one billion connections between individual neurons and synapses!
What’s the problem then?
Currently, even the most powerful supercomputers today such as the “K computer” at the Advanced Institute for Computational Science in Kobe, Japan can only tackle at most 10% of human brain simulation.
Will this method be truly feasible in the future?
Maybe? When future exascale supercomputers hit the scene — projected to be 10 to 100 times more powerful than today’s top performing computers — the algorithm can immediately run on those computing beasts and researchers hope to reach 100 percent simulation.
Nevertheless, compared with the AI path to machine intelligence, brain emulation seems to be more feasible as it relies more on concrete observable technologies (with traction) and not wholly based on theoretical insights.
2.3. Biological Cognition
If I can’t train machines to make decisions like humans or download a human brain, why not pop a pill to enhance my mental abilities?
Unfortunately, the use of drugs to improve memory and concentration already exist in the market and getting that special ‘pill’ to spark a dramatic rise of intelligence are generally dubious and shady (Be wary of the marketing gimmicks you see on eBay/Amazon).
This path extends beyond taking drugs to include ‘Manipulation of genetics’ as a way to achieve substantial improvement in cognition.
Far-fetched? I think not.
Dum dum dum………..
Late last year (Nov 2018), a team of scientists led by Southern University of Science and Technology, Shenzhen researcher, He Jiankui, claim to use a gene-editing tool known as “CRISPR” to tailor the genes of twin girls to make them resistant to HIV.
This incident is one most significant experiment in the history of human genetics and led to a renewed debate about whether designer babies are going to become a reality in the very near future.
Ah, but it’s not that simple…
WAIT BUT WHY?
(1) Genetics research has yet to progress to the point where scientists can pinpoint the genes related to intelligence.
(2) Maturational lag- What happens when the selected embryos grow into an adult human being? What if genetically modifying a specific gene (intelligence) leads to the creation of new rare and nasty diseases?
(3)Genetically modifying 1 child will affect its successive generation. We are talking about an impact that will spread over multiple generations!
(4)Social implications- Imagine a nation filled with smart babies only because the nation can afford the financial and technological resources. What are the implications?
On an individual level: How are non-genetically modified humans going to deal with a population of genetically enhanced humans?
On a global scale: Other nations will lose out in economic, scientific, military, etc. Imagine how this phenomenon will affect global equality?
& the list of constraints goes on…..
Will this method be truly feasible in the future?
As you can see, the biological path is clearly technologically feasible and cognitive enhancement could accelerate science and technology, including progress towards Machine Learning and Brain Emulation. Imagine a world where Average Joe had the brains of Einstein/ Alan Turning.
BUT (there is always a but), there are significant consequences to adopting this approach as mentioned earlier.
- 3 potential paths to achieving Superintelligence:
- Machine learning– Inputting codes into a machine for it to learn. Develop SEED AI: A machine that will iteratively improve itself w/o human intervention until it becomes smarter than humans
- Brain Emulation– Inserting contents of a human brain into a thumb drive and plugging it into a supercomputer
- Biological Cognition– Undertaking less-computer-centric approaches: infant nutrition, better education, and even selective breeding
2. Weak forms of superintelligence can be achieved by means of biotechnological enhancements (Education/infant nutrition/cognitive related drugs to improve memory/concentration).
3. If all the ethical/scientific concerns of creating ‘perfect babies’ are resolved, this adds to the plausibility that advanced forms of machine intelligence will become more feasible.
Hope Part (1) answered the “How” question and provided you with good insights on the various ways AI can potentially reach Superintelligence.
Stay tuned for Part (2) on “When” (in terms of the timescale) AI can potentially supersede human intelligence!