Source: Deep Learning on Medium
The quest to replicate & mimic the human brain seems scary but incredibly fascinating at the same time. Are you interested in artificial intelligence? This article might help you get a better understanding of the technology and its recent developments.
We hand-curated this list of 5 TOP artificial intelligence videos for you. In case you don’t have time to watch the video — don’t worry, we took the liberty of writing summaries for each video!
Most of the topics spreading around the internet regarding artificial intelligence (AI) come from developments of deep learning. Deep learning involves the use of algorithms by using statistics to find patterns in data. It is becoming more powerful in mimicking humans ability to see and hear, while even going further to imitate our ability to reason/think. Think about Google’s search, Facebook’s/Instagram’s news feed, Amazon’s & Netflix’s recommendation engine — all powered by the vast amount of data and AI used to make sense of that data.
We have been used to machines doing physical work since the industrial revolution but AI is heading towards a future of doing the “thinking work — planning, strategizing and making decisions.” (“Marr, B.”)
A common misconception is to believe AI is solely about creating automation, because it extends further towards the augmentation of human decision making and interactions. AI is rapidly growing due to developments in algorithm design, improved computing power and of course the most significant of them all: advancements in big data.
So let’s dive into the top 5 videos about artificial intelligence — all easily compiled in one list — so you can enjoy a power learning/inspiration session on AI:
Comment below if you know more videos, which you would like to see added!
1. Artificial Intelligence is the New Electricity (1h:27min.)
Former Baidu chief scientist, Coursera co-founder, and Stanford adjunct professor, Andrew Ng, speaking at the Stanford MSx Future Forum. The analogy of AI being the new electricity: Andrew believes AI has a similar impact on how electricity transformed industries. He structured his speech in the following 4 sections:
- What are the trends of AI
- How can AI impact businesses
- The process of working with AI: practical advice
- Societal impact of AI
The IT industry is already being transformed by AI, including: web search & advertising, approving a consumer loan, ordering a take out, estimating delivery time. Fintech is on its way to be transformed by AI. Logistics, Healthcare, self-driving cars — all will be transformed by AI. For him, it’s hard to think of an industry that will not be transformed by AI in the next several years.
Despite all the hype, what can AI really do?
Its driving incredible economic value:
Input to response mappings: Most value of AI is driven by one type of AI: supervised learning — using AI to figure out a relatively simple A (input) to B (output) mapping — so for example giving an email and telling whether it is spam or not — or providing an audio clip and providing its transcripts. One of the most lucrative AI applications is the advertising industry, modeling whether an ad will be interesting for the specific user.
Many product managers find it difficult to figure out what AI can do for them. Product managers need to look at data, design features that users will like, sometimes even determine the marketing and pricing. Simplified: designing the features and deciding what the product is supposed to do.
The rule Andrew provides is that anything a typical human can do with 1 second of thought, we can probably now or soon automate with AI. If a human can do it, it must seem feasible, there must be data available for it & one can use human insights. Patterns Andrew discovered is that AI progress tends to be fastest if we try to program it to something humans can do. Diagnosing medical images for example or driving a car. However, predicting how the stock market will change for example seems more difficult for humans and therefore is more difficult for an AI to do that well. Over time AI will tend to make rapid progress to surpass human performance. However, Andrew mentions that AI will compete with human jobs.
One of the main reasons AI has really taken off is because of GPU computing that is able to absorb the vast amount of data being produced. Deep learning evolved from that. The graph above indicates how the performance of AI systems increase when the neural network becomes larger.
How do you build a defensible business with AI?
Andrew argues that the AI research community is quiet open and publishes the results freely. It’s difficult to keep algorithms secret anyway. There are 2 scare resources being explained: Data & Talent.
Data: For example, in speech recognition massive amounts of data is required or in face recognition (the research papers with largest training included around 15–200 million images). Obtaining this data required for AI is difficult.
Talent: AI needs to be customized for businesses — this AI talent that knows how to tune algorithms for businesses is scarce.
To create a positive feedback loop the circle includes the virtuous circle (as seen in the diagram below). Regarding the fear of AI, Andrew noticed a so-called “anti-virtuous circle of AI hype”, where the fear actually drives funding but the funding goes to anti-evil AI, leading to the results of that funding driving back to the evil AI hype. Andrew describes that this is an unhealthy cycle, especially when there is a financial incentive to drive fear. Job displacement is a major challenge that Andrew hopes people to focus more on.
Regarding product management, it becomes more important for product managers to work together with engineers in figuring out AI that is feasible, while at the same time being something users love.
2. Artificial Intelligence: Mankind’s Last Invention (20min.)
This video begins with an introduction to a board game named “Go”, arguably being the most complex board game in existence, being played since 2,500 years. In 2016, Google deepmind’s “Alphago” beat 18 time world champion Lee Sedol in ⅘ games. Why is this impressive? Go is such a complex game, which cannot simply be predicted and there are over 10¹⁷⁰ moves possible (in comparison we only have 10⁸⁰ atoms in the observable universe). Alphago was trained using human data from Go games and learned the techniques used by running through millions of games, while even creating new techniques that no one had ever seen. Even though this is already impressive enough, after 1 year of Alphago’s victory, a new AI named AlphagoZero had beat the original Alphago by 100/0 games in a row.
The fascinating part is that AlphagoZero had learned to play with zero human interaction and therefore was not restricted to human knowledge — surpassing 2,500 years of strategy/knowledge in only 40 days — simply playing against itself.
The thought being raised: If AlphagoZero learned how to play without any human interaction, made up strategies of its own and then beat us using those — that would imply that there is more non-human knowledge about Go then there is human knowledge. Leading to the conclusion that at some point there will be more non-human intelligence than human intelligence. There might be a point where humans represent the minority of intelligence.
The video tackles the question of what happens when we get stuck with AI that exponentially is becoming smarter than humans. Since it would be near to impossible to stop these developments unless someone could turn off the whole internet — AI cannot be stopped.
When people think about AI they often envision superintelligent AI that is set to destroy humanity — as captured by the entertainment industry. But will this really happen — AI creating malicious intent?
Currently there is a focus on narrow artificial intelligence, known as “weak AI”, AI that is being created for the sole purpose of handling one task. For example AlphaGo, speech recognition or facial recognition. In order to teach machines to think and learn like humans, a technique called machine learning, is used. The video further explains the concept of neural networks. It argues that as more and more data becomes available it becomes too difficult for a human to process this — which is where machine learning comes in handy. Machines don’t just analyse, but more importantly even learn from data. How can machines learn & analyse data so fast you may wonder? The biological neurons in a brain operate at about 200 hertz, while modern transistors (electrical neurons) operate at over 2 gigahertz. Neurons in a brain travel through axons, at about 100 m/s (pretty fast), but that’s only ⅓ as fast as the speed of sound. In contrast, computers can transmit information at the speed of light (300 million m/s). If you think about all these statistics it becomes evident that our brains capabilities and computers capabilities cannot be compared and will never be the same.
Starting in minute 8:01, the video addresses artificial general intelligence, known as AGI. AGI has more than a single purpose, working closer towards human intelligence. A big challenge towards reaching this point though is that humans don’t have to go through every option to reach a conclusion. So even though machines can transmit information faster, they still need to process so much more information, unrelated thoughts and observations.
It becomes clear that teaching machines to employ human level thinking is extremely difficult. Humans have the power to create, invent, build societies, play & laugh — all very difficult aspects to teach a computer. “How can you teach a computer to create something that doesn’t exist or hasn’t even been thought of? And what would be its incentive to do so?”
This where the video addresses HGI, meaning “strong AI”, as being one of the most significant AI’s — eventually leading to what is known as technological singularity, where artificial intelligence becomes so advanced to the point of an extreme explosion of new knowledge and information — some that humans might not be able to understand. Super intelligent AI would continuously improve upon itself, becoming smarter in a shorter amount of time and since this process is repeated, it will become faster each time too.
How is this all possible? Well, machine learning is exponential — starting slow but there will be a tipping point where everything starts to speed up. The difference between weak AI and strong AI is millions of times larger than the difference between strong AI and super intelligent AI. After reaching artificial general intelligence functioning like a human being, it may help us reach super intelligence level relatively fast.
Nowadays people tend to think we program something in the computer and the software will follow these rules — but this is not going to be the future. With highly advanced AI and machine learning, the AI teaches itself how and what life is, as well as how to improve it.
At some point there might not be a need for any human interaction — actually humans might slow down the process. Most importantly: what about wisdom? Wisdom & intelligence are not the same — so how can AI’s discover wisdom? If we were to ask an AI to solve world hunger — the easiest result presented would be to kill all life on the planet so that nothing would ever be hungry — but of course that’s not a solution! So we need to teach AI a moral code to follow by. But this is so subjective. Some interesting challenges being presented in this video! Check out the video — it’s worth it!
3. Can we build AI without losing control over it? (14min.)
Sam Harris begins this interesting TED talk by introducing our failure of intuition regarding AI. Currently, the media is entertaining people with the thought of machines taking over the world. He argues that we seem unable to marshal an appropriate emotional response to the dangers of AI that lie ahead. If you watch this video, you will find out more about the complexity of AI.
Harris explains that unless something destroys civilization — the most likely path for AI is that it will continue to be developed. Nothing will stop the development of AI, as we are improving the intelligent machines year after year — the result leading to machines becoming smarter than us. Once machines advance to this stage, they will begin to improve themselves, moving forward to a so-called “intelligence explosion” risking that the process could get away from us.
Rather than having robots attack humans, as often caricatured, the real concern is that machines become so much more competent than us — that the slightest divergence between their goals and our own could destroy us. Harris provides an interesting example about the way we interact with ants. We don’t hate ants but when their presence conflicts with our goals, let’s say when building a house — we crush them with no regards. What if machines, whether they’re conscious or not, could treat us with similarly?
AI thrives on data, processing the information to create its intelligence. Harris expresses that intelligence is a matter of information processing in physical systems. He explains that the rate of progress doesn’t matter, because any progress is enough to reach the point where machines outsmart us. And of course: we want to keep going. There are problems that we desperately need to solve, including curing diseases like Alzheimer’s or cancer, understanding economic systems and improving climate science.
“The train is already out of the station, and there’s no brake to pull”
When considering the spectrum of intelligence, there is yet so much more to grasp than human intelligence — so much more to conceive. Machines should be able to think about a million times faster than the minds that built these machines, as electronic circuits function about a million times faster than biochemical ones.
“How could we even understand, much less constrain, a mind making this sort of progress?” Harris addresses compelling thoughts about AI and the future, aiming to make the audience think about these questions because there is yet no absolute answer.
“We have to admit that we are in the process of building some sort of god. Now would be a good time to make sure it’s a god we can live with.”
4. Google Duplex: AI Assistant (4 min.)
This video shows a practical example of an AI Assistant, called Google Duplex. Google Duplex allows real-world tasks over the phone to be accomplished by the AI. In this video, you can see how the technology is focused on completing specific tasks, such as scheduling appointments (for a hairdresser or a restaurant). The conversational experience is kept as natural as possible, as if the person is speaking with another person, without thinking that a machine is behind all of this.
It is fascinating to see how the AI adapts to the whole conversation. Some calls don’t go as expected, but nevertheless the assistant understands the context to handle the interaction successfully. If you come to think about, there are so many challenges during phone calls, including: loud background noises, sound quality issues, accented speech, and even different meanings depending on the context. For example, when booking a reservation “Ok for 4” can indicate the time but also the number of people. In this case, the relevant context can be several sentences back — the AI needs to be able to detect this. If you want to read more about this in detail, check out this blog post: https://ai.googleblog.com/2018/05/duplex-ai-system-for-natural-conversation.html
5. How does IBM Watson work? (6min.)
IBM Watson is an artificial intelligence system, being among the most well known world-wide and also one of the most advanced systems. This video is great to learn more about real world use cases and practical methods of AI. Overall a great animated video to understand the current applications of AI in the business world. This is already happening as we speak — always helpful to be informed!
Starting with min.1:09, the video explains how AI works. AI is what gives machines the power to learn, adapt to new inputs and make better decisions. Machine learning, a subset of AI, uses computer algorithms to analyze data and make intelligent decisions based on what is learned. For example: when streaming services compare what one listener likes with others having similar tastes, allowing the system to recommend new music to the user.
Watson uses a sophisticated machine learning technique called deep learning. Deep learning layers algorithms to create an artificial neural network. This seems complex but essentially it allows the machine to continuously learn on the job — constantly improving and in turn this leads to greater accuracy of results. For example when helping insurance companies: Instead of analyzing the images of a damaged car, the AI can judge/understand the car model and detect the damage, even being as detailed as a broken exhaust.
Deep learning also facilitates Watson’s natural language understanding. The system learns by deconstructing sentences, then analyzing and identifying the concepts and relations. However, AI also needs to understand the specific terminology of different industries. Traditional AI models this would require too much additional data & computing power, taking weeks or months. Watson however, uses a technique called transfer learning to speed it all up.
Transfer learning is learning how to learn. Watson uses this technique to learn more from less, so that the AI is not trained from scratch. Transfer learning has enabled Watson to accelerate claims processing by 25% dramatically cutting operating costs.
Watson’s transfer learning is a 3-layered AI model:
For example for home insurance companies:
- The bottom layer is made up of out-of-the-box general knowledge: like Wikipedia telling Watson what a house is or what a tornado is
- The middle layer is prepackaged with knowledge tailored to specific industries: so for insurance it includes terms like coverage or beneficiaries
- The top layer is where the customer specific learning takes place: personalized to the company’s unique business needs to be fed with their own data and know-how, even understanding specific risk attributes and behavioral policy pricing models.
However, massive companies like IBM, that build AI systems, are of course doing it for their benefit to make money, therefore investing in R&D to be up-to-date on the cutting edge of new technologies. IBM Watson is not necessarily designed to help ordinary people or smaller businesses. “SingularityNET wants to address these issues by building the world’s first decentralized AI-as-a-Service open marketplace, where users can search for and implement AI services on the network. In this way, ordinary people and smaller businesses can now benefit from AI solutions which might not ordinarily be accessible to them.” (“faa.st”)
Don’t forget to check out Tanmay Bakshi (world’s youngest IBM Watson coder) explaining the IBM artificial intelligence: https://www.youtube.com/watch?v=8F0GRZhSBX4
We hope you enjoyed this article! Let us know your thought’s in the comments below.
Stay tuned and follow us on all our social channels for new updates!
Andrew, Ng. Artificial Intelligence is the New Electricity. Accessed on: 19.02.19, Available at:
Artificial Intelligence: Mankind’s Last Invention. Accessed on: 20.02.19, Available at: https://www.youtube.com/watch?v=Pls_q2aQzHg
Harris, S. Can we build AI without losing control over it? Accessed on: 18.02.19, Available at: https://www.ted.com/talks/sam_harris_can_we_build_ai_without_losing_control_over_it/transcript#t-459565
Google Duplex: A.I. Assistant Calls Local Businesses To Make Appointments. Accessed on: 19.02.19, Available at: https://www.youtube.com/watch?v=D5VN56jQMWM
Leviathan, Y., Matias, Y. Google Duplex: An AI System for Accomplishing Real-World Tasks Over the Phone. Accessed on: 19.02.19, Available at:
Marr, B. 5 Important Artificial Intelligence Predictions (For 2019) Everyone Should Read. Accessed on: 14.02.19, Available at: https://www.forbes.com/sites/bernardmarr/2018/12/03/5-important-artificial-intelligence-predictions-for-2019-everyone-should-read/#ae2f8fc319f5
How does IBM Watson work? Accessed on: 15.02.19, Available at: https://www.youtube.com/watch?v=r7E1TJ1HtM0
faa.st. AI Meets Blockchain: How SingularityNET (AGI) is Making AI Accessible to Everyone. Accessed on: 15.02.19, Available at: https://medium.com/faast/ai-meets-blockchain-how-singularitynet-agi-is-making-ai-accessible-to-everyone-47abf2d6d820