4 min read
Artificial Intelligence is growing rapidly, and so is the need for understanding how it works. In addition to launching our first growth programme for companies which develop it, Tech Nation also produced a Guide to AI for anyone who wants to learn about the impact AI is having on society today, translate the jargon and cut through the hype.
How to: Understand the impact it's having on society
Will AI take my job from me? Will it surpass human intelligence? And most importantly, will Alexa finally be able to play the song I request? AI is associated with a set of risks, rewards and warnings. We want to discuss both sides of the coin.
Jobs, skills and investment
AI in the UK is raising more investment than competing countries. This is reflected particularly in sectors such as fintech and healthtech, which have seen particularly strong growth.
**Some of the biggest fundraises of last year came from Graphcore, which raised $200m. Renalytix AI, which develops AI that can diagnose kidney disease, raised $29m along with Upscale alumni, Beamery, who raised $28m.
AI has had a rocky history so far - it's no secret that interest has peaked and troughed along the way. But we can safely say we're in the middle of a surge of interest, hype, investment and research around AI. So who set this all off?
Geoffrey Everest Hinton
Who are they: English Canadian cognitive psychologist and computer scientist.
How they revolutionised AI: Hinton's research discovered ways of using neural networks for machine learning, perception, memory and symbol processing. His work has earned him the title of "Godfather of Deep Learning."
Who are they: Chinese-American computer scientist, investor and global leader in AI.
How they revolutionised AI: Apart from his many contributions to artificial intelligence, deep learning, robotics, and machine learning, Ng co-founded and led Google Brain and is a leader in "democratising" Deep Learning.
Their work contributed hugely to a step-change in the effectiveness of AI, which enabled a series of new, more practical uses. It’s really why we’re talking about AI again today and why you can ask you're Siri for directions or news updates.
Why everyone needs to get involved
Democratising AI is a big part of the government's Office for Artificial Intelligence work - and there's a good reason for that. AI is shaping how we live our lives, and how we build it is vital for it to work well for everyone. Differences in people, unconscious bias or simply lack of data from a certain group can have detrimental effects for the finished product.
Investment and research in AI simply isn't the answer to improving it. Dame Wendy Hall, Professor of Computer Science at Southampton University helped write the government's inaugural report on Artificial Intelligence (AI) and she has several times championed diversity and inclusivity in AI.
I think AI is too important to be left to the AI experts, we’ve all got to be involved. I think interdisciplinary teams are the answer. I think we’ve got to encourage lots of people to come into AI from different backgrounds. And we’ve got to do that now.Dame Wendy Hall, Professor of Computer Science
Interest creates jobs
Over the past few years, there has been a surge of interest in AI. This has been sparked by both a positive impact and a negative outlook from a number of thought leaders. The spike in interest and investment has however led to an increase in jobs in AI.
One report suggested that 300,000 jobs exist in Data Science, although the figure is speculative. We can also see tech giants investing internationally to expand their talent pool.
This also suggests that 2 million vacancies for Data Scientists currently exist. The skills gap has been identified to be the largest barrier to implementing AI across business operations.
Industries that benefit from AI
Healthcare, medicine and geriatrics
Artificial Intelligence is just breaking through in the healthcare sector. Several cases of improved cost and efficiency of care in areas like geriatrics, development of medication, diagnostics of serious illnesses and even for veterinarians have been recorded. One particular mind-boggling and life-changing story is that of Dennis Aabo Sørensen, who has regained the ability to feel, using a robotic hand that interacts with nerves in his arm.
Possible issues with AI
AI is all about making systems that are autonomous and help us carry out tasks. But what if the power of AI falls in the hands of someone wanting to cause harm rather than help? It’s hard to imagine what could happen if AI starts making its way into weapons manufacturing, or how hackers and cyber criminals could use the technology in the future.
AI is becoming more and more prominent in Cyber Security, specifically technology that helps us detect and defend against attacks before they happen. Preventative measures in Cyber Security saves people money, time and data loss.
Bias & discrimination
It's becoming more apparent that AI is inheriting a lot of our issues. In a male-dominated industry, severely lacking input from minorities, finished AI products have been reported to be biased and inadequate when analysing data input from minorities.
The school system hasn’t exactly been revolutionised since it first became an established part of our society, but we are seeing more and more examples of AI being used to create more personalised ways of learning for all levels of students.
What happens when technically a machine earns you money? Businesses using AI in Silicon Valley have 10 times fewer employees than that of a manufacturing business, leaving a bigger piece of the pie for the business owners.
Infrastructure & city planning
AI is excellent at predicting new trends or potential risks, like where traffic collisions are more likely to happen. And it has recently found its way into city planning, predicting everything from new housing estates to public transport.
How we interact with robots on a daily basis is shaping how we, and specifically children, communicate. Think about it - when was the last time you thanked Alexa?
We see more potential for AI to be used in creating practical solutions than ever, but for society to see the benefits, we need founders and leaders who want to solve real-world problems. This is what Tech Nation is trying to achieve with our Applied AI programme. Angie Ma, COO of Faculty, who supports AI companies on their scaling journeys puts it this way:
AI is the transformational technology of our time, but for society to see the benefits we need to apply it in the real world - enhancing products, improving services and saving lives.Angie Ma, COO, Faculty
Tales of Glory & Caution
With big names like Bill Gates, Stephen Hawking and Elon Musk warning against the progress of AI, you can start to wonder “are we ready for this?” Eleni Vasilaki, Professor of Computational Neuroscience at the University of Sheffield explains that it is more a question of how AI is being developed and used that causes issues.
“Machine learning algorithms are often thought of as black boxes, and less effort is made pinpointing the specifics of the solution our algorithms found. This is an important and often neglected aspect as we are more obsessed with performance and less with understanding.”
The answer may be that a stronger sense of regulation, ethics and general knowledge about this technology is needed, and this is why consideration of ethics is a strong judging criterion in Tech Nation’s Applied AI programme. If you feel like you could use some more info on what AI is and learn more about the technology, keep on reading!
The term Artificial Intelligence (AI) refers to the development and creation of machines that think or act like humans or can carry out tasks traditionally seen as being exclusively solved by human intelligence or speciality movements. Basically, this means that AI can be anything from a simple calculator to a robot hand with opposable thumbs. AI has technically been around for decades but first became a term in the 1950s.
Machine Learning (ML) is a subcategory of AI and refers to machines that learn or develop new skills without being exclusively programmed to do so. This is often combined with Big Data (which is exactly what it sounds like; lots and lots of information). It started out in the early 1980s and made its way into many commercial and household brands.
Deep Learning (DL) is defined as the training of a machine which uses an algorithm that attempts to mimic or solve tasks like a human brain. An algorithm like this is called an ANN (Artificial Neuron Network) and Deep Learning occurs when the ANN consists of several layers of “neurons”, sending information around. It took off in the late 2000s.
**In addition to this, you can also divide AI into the following categories
- Weak: Weak AI focuses on completing one single task, like playing chess or answering a question.
- Strong: A machine that perfectly simulates a human brain
- Artificial Superintelligence (ASI): A machine that can outcompete a human brain in any task.
The last two do not exist. Yet.
Too. Many. Abbreviations. Understanding AI can be daunting when you have to look up every other word just to deal with an explanation of it. Our glossary puts it plain and simple:
- AI Winter
This is a term for a time period where interest, investment and research in AI drops. Until now, AI has experienced several winters and they are usually triggered by pessimism in the AI community, followed by bad press and then cuts in funding.
- Artificial Neural Network (ANN)
Any algorithm imitating the human brain. An ANN is built up of layers of connected synthetic neurons that send information to each other.
- Artificial General Intelligence (AGI)
This refers to the intelligence of a machine that would be able to perform a task in the same way as the human brain (or a different way of saying strong AI) and is a popular subject amongst science fiction writers and researchers.
- Big Data
Loads of information that is both structured and unstructured, and normally too complicated for standard data-processing software to work with.
- Black box algorithms
When the computer or the researcher can’t really explain a decision the algorithm has made.
- Computer Vision
This interdisciplinary field of AI is trying to build computers that will gain human-like understanding by looking at photos or videos.
- Convolutional Neural Networks
A very simplified way of describing this Deep Learning algorithm is that it can recognise images, assign importance to certain elements of it and recognise copies or fakes.
- Deep learning
Several layers of ANNs, connected together. Hence it’s “deep.”
- Embodied A.I
A very complicated way or saying robots that use or run on A.I.
- Explainable AI or XAI
This type of AI can tell human operators exactly how it reaches a solution or conclusion.
- Few-shot learning
Normally a computer vision system that will need to see thousands of examples to be able to imitate something. Few-shot learning is the creation of a system that needs much less training. Compare it to how a toddler will, for example, pick up several languages with ease, but an adult can struggle with the simplest of pronunciations.
- Generative Adversarial Networks (GANs)
Two neural networks (or ANNs) that are trained with the same set of data. The first one tries to replicate the data it has seen and the second one judges the results. If the second network recognises an unauthentic copy, it will force the first network to make improvements.
A technique within computer science. It aims for quick, optimal, solution-based problem-solving.
- Machine Learning or ML
A system that learns from a data set and improves one specific task. This area of AI is experiencing the biggest research and investment boom.
- Natural Language Processing
A specialisation within AI that manages verbal and written language.
- Recurrent Neural Networks (RNN)
Recurrent Neural Networks are built up by loops, allowing information to persist and be passed down. The goal is to imitate how humans naturally learn and use their experience to read situations.
- Reinforcement learning
A machine will learn how to do a task through a “reward and punishment” system.
- Supervised learning
Here, a human would divide data into categories and teach an algorithm to solve a specific task using the data.
This collection of software tools is open to the public, as developed by Google to use in Deep learning. Anyone can use or improve it.
- Training Data
Think of this as a textbook for AI - a set of data is used to train, adjust and improve it.
- Turing Test
Created by Alan Turing in the 1950s, it tests if machines could show intelligence equal to or identical to that of a human.
- Unsupervised learning
This approach tries the opposite of supervised learning, and feeds only unlabelled data to the algorithm and lets the machine “teach” itself.
Did you know your Alexa, Siri and Google Assistant are all great examples of Natural Language Processing? The goal is to make these machines speak as naturally as humans, and the more they listen to us, the more they understand what we are saying.
There is no doubt that AI has an amazing, unreal and almost mysterious aura around it. It’s difficult, if not impossible, to predict how fast it will develop, but let’s demystify with some examples of far AI has actually come.
Examples of areas in which AI is superior to humans
In 2002 a machine named CHINOOK solves the game by only making the best possible moves. However, it never officially beat the world champion.
Solving a Rubix Cube
This tiny bot made by Ben Katz and Jared Di Carlo was able to solve a Rubix cube in 0.38 seconds. They claim it "can most definitely go faster."
Computers are officially better than us at recognising patterns in data. As mentioned, this has seen a range of different benefits in several sectors
BKG, created by Hans Berliner, wins against the world champion in 1979, the first machine to do so. He confessed a bit of luck was involved.
Examples of areas in which humans still have the upper hand
Number of brain cells
The human brain is built up of over 86 billion neurons. Computer neural networks have way fewer. Not even close.
AI trying to be creative hasn't worked so far. Just check out this Inspirational quote generator.
AI tends to be better at solving the more complicated crosswords but is still stunned by the easier ones or relatively easy complications.
Easy! Check out some of these suggestions.
Superintelligence: Paths, Dangers, Strategies - Nick Bostrom
Homo Deus - Yuval Noah Harari
The AI Podcast by NVIDIA
What is the big fuss about Artificial Intelligence? by Getting Curious with Jonathan Van Ness
News & blogs
MMC Ventures State of AI Report 2019 also serves as a useful introduction
Ways to think about Machine Learning by Benedict Evans
AI Nationalism by Ian Hogath
Who Is Going To Make Money In AI? By Simon Greenman in Towards Data Science
Nathan Benaich’s newsletter is a summary of the latest happenings in the AI startup sector