The true sign of intelligence is not knowledge but imagination
— Albert Einstein
I know that I am intelligent, because I know that I know nothing
— Socrates
So how do you define intelligence? For some, intelligence depends on an IQ test. For others, it is a nebulous idea related to how you receive, process and understand information. However, Howard Gardner suggested that the traditional notion of intelligence based on IQ was too limited. He developed the multiple intelligences theory to explain different types of intelligence. One person excels at maths while another uses interpersonal skills in difficult situations.
I raise this point to show how hard it is to define human intelligence. Trying to define artificial intelligence (AI) is even tougher.
But let’s try. Let’s accept that intelligence includes the ability to perceive the world around us. And it allows us to interpret those perceptions and remember information. Then we can apply that knowledge within a specific context.
So, we see/hear something, understand it, learn from it, and use it to adapt or achieve a goal. Taken this way, AI would see a machine doing the same thing.
But not all independent machine actions are AI
In 1997 IBM’s Deep Blue computer stunned the world by beating Garry Kasparov. The machine defeated the reigning world chess champion in a six-game chess match.
Was this an example of AI in action?
No. It was not even machine learning at work. Instead, programmers worked with chess masters. They created a vast library of chess moves within Deep Blue’s code. They told the computer which moves were good, and which were bad.
The machine did not learn anything. It searched its memory for the best move out of those available. It shows how clever programmers and chess players can be, but not machines.
For this to have been AI in action, Deep Blue would have needed to study Kasparov’s game. Then it would have predicted moves based on their interactions.
Deep Blue would have learned to beat Kasparov.
What is the value of AI?
Computers first connected to one another in the 1990s with the birth of the internet. In 1995 it is estimated that there were 16 million connected users. The arrival of the Internet of Things in 2015 took this one step further. Gartner predicts that there will be 20.4 billion connected devices will be in use by 2020.
The sensors in these devices generate a staggering volume of data. AI offers a way to filter and process this data so that humans can use it for something meaningful. Using machine learning (part of AI), computers can spot patterns in data faster than humans.
One example is the smart city. In time, the city can tell drivers which routes to take to avoid traffic jams. Or it can tell them where to park based on information gathered from traffic or parking lot sensors. Humans can still glean some of this data from traffic cameras. But an individual needs to monitor the traffic feed and relaying messages.
AI frees people from these tedious tasks. So they can do more meaningful work elsewhere.
You can also use AI every day, from speech recognition to pattern processing. Asking Alexa the time becomes an AI interaction.
Neural Networks and AI
A detailed analysis of neural networks is outside the scope of this article. But in short, they use complex maths, statistics and probability to solve complex problems.
This is critical for two reasons. First, programmers could struggle to write the programs needed to model or solve many problems that AI techniques can solve more easily. Second, even if we knew how to write the programs, they could be too complex and impossible to get right.
If we combine machine learning and AI algorithms? They, and properly selected and prepared training data, can do it for us.
Deep learning sounds like a flashy concept. But it describes certain types of neural networks and related algorithms. They often consume raw data as input. They process this data through multiple layers of nonlinear transformations to calculate a target output.
This form of deep learning excels in unsupervised feature extraction. This happens when algorithms note important features within the data. In the past, the burden for feature extraction lay on the data scientist or programmer. They used the features for further learning, generalisation, and understanding.
Deep learning is probably one of the most exciting and innovative AI techniques. But it is not without flaws. It can reach an approximation of something, but it often cannot tell you how it reached that conclusion.
For example, say we want to teach it the difference between cats and dogs. We show the algorithm photos of each. While teaching it the difference, if we accidentally put a white dot on the photos of a cat but not on the photos of a dog the algorithm could think a cat is actually a white dot!
Uses of Artificial Intelligence
Despite the problems with algorithms, AI still offers lots of solutions. The recent headline-grabbing use of AI is in self-driving cars. UK Chancellor Philip Hammond announced he wanted self-driving cars on British streets by 2021.
Self-driving cars use a series of cameras and sensors to watch the world around them. They compare the incoming data with the programmed ‘rules of the road’ to navigate the streets.
In 2016, Joshua Brown’s self-driving Tesla car smashed into a truck that pulled out in front of him. The car’s AI did not recognise the side of a truck. Its programming only allowed it to recognise the front and back. Detractors point to the limitations of AI and its tendency to stick to its rules. It does not have the same freedom of thought as a human driver.
However, human error accounts for over 90% of car accidents. Food for thought?
But AI offers better solutions in the world of medical research. For Medical Futurist, using AI frees up time for overworked radiologists. That lets them focus on more difficult cases.
Even the art world has teamed up with AI. The Deep Dream Generator lets users see what neural networks ‘see’ through vivid artworks. It even learns art styles from one image to apply them to another. Google’s Quick, Draw project aims to teach neural networks to recognise doodles using crowd-sourced drawings. And InspiroBot is an AI inspirational quote generator, providing often hilarious results.
AI Successes
Netflix, Amazon, and Spotify rely on algorithms to recommend your next viewing experience.
And Google improved its search engine results using its RankBrain algorithm. Before, its engine relied on keywords to tell it what a web page was about. If a user searched for ‘best places to stay in Athens’, Google found websites containing that phrase. Now, the same search phrase can tell you where the best area is to stay. Or it can name the best hotels. It can provide a guide to the coolest neighbourhoods. Or find out where to stay that is close to tourist attractions.
All because RankBrain understands synonyms and context. It understands the nuances of what you mean by ‘best’.
And AlphaGo looks like one of Deep Blue’s AI descendants. It learned to play Go by analysing thousands of human games. Like Deep Blue, it even defeated a world champion.
But AlphaGo is far more advanced than Deep Blue. Go is a more complicated game than chess. The AI showed a human approach to the game by evaluating possible moves and choosing the best one.
AlphaGo Zero, the next version, removed human knowledge from its training. It relied on self-play and search. This version showed how an AI could achieve even better results and adapt to play similar games.
Does This Mean Humans Will Become Obsolete?
Not yet. As clever as AlphaGo is, it is still only focused on one task. It is impressive, but it is still a game. This focus is what is called ‘narrow’ AI.
By comparison, a human can master many tasks. And the failures of self-driving cars show what happens when you introduce AI into a world run by humans.
That said, AlphaGo still shows that AI can evaluate before it decides in the same way humans do. That might work for Go at present, but it could replicate across other uses.
Strong AI, or human-level AI, poses more of an issue. This branch of AI can understand, reason and learn its environment as a human would. The industrial revolution saw workers replaced by machines. So, humans found new ways to work that machines could not replicate. Then the assembly line began and automation forced employees out of work. Will AI cause a similar seismic shift in the way we live and work?
Some think the end of the world refers to Super-Intelligence. Machines become sentient and remove mankind from the planet. (See The Matrix or Terminator).
But I don’t think so
I don’t think survival will be the problem. Mankind has proven itself very adept at creating weapons and so far, we have survived. So, I think we will survive this as well.
And perhaps it is not the end of the world that we face at the hands of AI. Maybe it is the end of the world as we know it and it will bring something new.
Everything is now connected. Your devices share details of your life with one another and commercial organisations are collecting more and more data on central servers. The Cambridge Analytica scandal revealed the value of that data for marketing agencies.
But your devices demand so much more of your attention. Your notifications no longer tell you ‘you’ve got mail’. Now they tell you who is at the front door, how warm your bedroom is, and you need to buy more yoghurt.
Ask yourself, do you want so much of your life being shared and analysed?
The header image is reproduced curtesy of pixabay (Link) and is CC0 Public Domain.