Slides from an internal presentation I made for my colleagues.
Introduction
Unless you've been living under a rock with no wifi, chances are you've heard of these three terms in numerous interviews, articles and conferences.
Unfortunately, a lot of us (especially ones who're not directly working with product, data science or engineering) can do little more than vehemently agree each time someone says AI, Blockchain and IoT will change the world. Finding out more seems complicated, but if you dumb things down a lot, grasping the basic concepts doesn't require a PhD.
Artificial Intelligence (AI)
Let's start with Artificial Intelligence. AI deals with simulation of intelligent behavior in computers using mathematical or logical thinking. Put simply, it seeks to mimic human behavior. Examples of AI applications are all around you - voice assistants who try to mimic an assistant, driverless cars that mimic a chauffer and Netflix recommendations that mimic a friend who wants to turn you into a depressed couch potato.
Machine Learning (ML) and AI are used interchangeably by many. But ML is a special application of AI that gives systems the ability to automatically learn and improve from experience without being explicitly programmed. It works using huge volumes of data and progressively trains the machine into recognizing patterns.
You feed sets of inputs into a machine and let it figure out the process to get to a desired output. If correct outputs for a set of training inputs is provided, it's called supervised learning - it makes it easier for the machine to train itself. Once trained, the machine can give you the correct output for a new set of inputs.
Otherwise, you may simply ask the machine to keep learning on it's own, unsupervised. The second approach is usually used for grouping data points or to detect anomalies.
An associated term is Neural Networks. Without pretending I know much more, I can tell you that these are just mathematical functions that feed off each other, i.e. the output for one function becomes the input for another and so on. Deep learning is a specific case of ML that uses layers of Artificial Neural Networks.
Deep learning is smarter in a way, because different layers start specializing on their own over iterations and begin to extract different features. For example, you could use deep learning to extract different features from concert videos: faces, lighting, sounds, etc.
The current state of AI is reasonably advanced: we have audio assistants, self driving cars and bots playing chess. But the holy grail is to replicate the human brain.
While the advantages are numerous, an AI is only as good as the data training it. And if the data set has human biases, they creep into the AI system as well.
Case in point: Propublica, a nonprofit news organisation, had critically analysed risk assessment software powered by AI known as COMPAS. COMPAS has being used to forecast which criminals are most likely to reoffend. Guided by these risk assessments, judges in courtrooms throughout the United States would generate conclusions on the future of defendants and convicts, determining everything from bail amounts to sentences.
In short, as smart as AI is today, it's still not smart enough to correct for human stupidity.
Given the exponential boom in AI tech, how long before the machines take over you ask? Well don't hold your breath. The type of AI we have now is weak AI, also known as Narrow AI. This is because it focuses on a single task and is developed to a point in which it can perform this task better than humans.
When combined with many other Weak AI systems, Strong AI can be created. This second category of AI refers to the systems that are usually depicted in the movies. Strong AI refers to machines being able to think and perform tasks on their own, similar to a human. But as of now, it's a theoretical concept.
One of the most common applications of AI we find around us is in IoT.
Internet of Things (IoT)
Let's start by defining IoT. An IoT system essentially collects a lot of data through different devices , analyzes it and directs actions across the network ecosystem. It connects any device with an on and off switch to the Internet.
IoT devices are almost always special purpose devices and not just computers. This includes everything from cellphones, coffee makers, washing machines, headphones, lamps, wearable devices and almost anything else you can think of.
The %age of IoT devices is expected to double within the next 5 years.
This growth has been made possible by an ever increasing availability of internet bandwidth.
IoT systems always use some form of AI or ML. Some basic uses of IoT have started in travel. The Kayak Alexa skill can provide information such as where users can go for $300, hotels in Barcelona and a flight’s expected arrival time. Bookings still happen via mobile or desktop. But most travelers use it for weather information at their destinations (60%) and traffic updates (54%).
Smarter devices, better algorithms & faster internet - what could go wrong?
Security and data privacy are big issues that are often brought up. With billions of devices being connected together, what can people do to make sure that their information stays secure? An IoT toaster would probably not be the most secure device around, but what if someone hacks into it and then gets control of the entire house network - the heating, electricity, car?
On the supplier side, the issue is to figure out a way to store, track, analyze and make sense of the vast amounts of data that will be generated.
Speaking of security, there's a candidate that claims it can keep all data secure. Everywhere. And forever.
Blockchain
We've all heard of blockchain, but few know what it is or how it works.
In simple terms, blockchain is a shared database managed across a network. The data is encrypted and resistant to tampering. Every "block" in the chain has a unique digital signature assigned based on the data in the block. The signature is included in the data for the next block. Any change in data for the first block changes the signature and "de-links" from the next block, which has the original signature. Everyone connected to the blockchain has a copy of the entries. The data is encrypted but the entries are visible.
In theory, a blockchain can be altered but for a large blockchain like bitcoin, the cost of computing power required would be greater than the gains from hacking the chain. Basic concepts of trust & immutability apply to a truly decentralized system – a public blockchain. Private & permissioned blockchains don’t offer a fundamental shift, since control is still confined to a few.
Blockchain technology has the potential to revolutionize travel, by
Making it easy to enter the market: Capital guarantees will not be required. With lower barriers to entry, the airline industry will in theory become more innovative and competitive, and therefore better for customers too.
Improving fraud detection & cash flows: The IATA says that the airline industry loses as much as $1 billion per year due to fraud, plus many more billions on the costs of collecting payment. Blockchain would heavily reduce those payment acceptance costs and all but eliminate fraud. Introduction of smart contracts could automate settlements.
Reducing costs: Without intermediaries, travel industry players can make more money, and offer cheaper tickets and services to their customers too. True only for public blockchains. Permissioned & private ones would still cost fees.
But the same characteristics that make blockchain such a powerful tool also pose some challenges:
Scalability & speed: A test was done with four airports and one airline and up to 2 million changes recorded. The technology was considered immature for various reasons including ‘user friendliness’ and scalability. There is a lot of redundancy and junk data generated in the airline industry – storing each one of them is not essential and might hamper scalability and add to the cost. Purely decentralized Blockchains often have a structure that hinders them from achieving high transaction speed levels. the founder of Ethereum recently announced some tweaks (i.e. sharding and plasma) that are envisaged to boost the performance. Those tweaks could allow Ethereum to reach close to 100-150 transactions per second. To put that in perspective; Visa handles 24,000 transactions per second.
Revenue Management: Winding tree operates with a completely transparent ledger. Okay for hotels, but for flight seats which are perishable inventory, transparency can complicate revenue management.
Privacy concerns: GDPR people must be able to demand that their personal data is rectified or deleted. A blockchain on the other hand is a growing, shared record of past activities. A blockchain is distributed across many computers. And a key factor for its reliability is that this chain of blocks (essentially transactions) is in practice unchangeable. Not exactly two ambitions which match very well.
On top of that, there's the risk of volatility in cost. Winding Tree will offer a decentralized alternative to GDS and OTA distribution claiming to reduce cost of distribution and allow more flexibility. So, while a main attraction factor is that there is no fee charged to the supplier in order to provide their offer to the market place. The problem is that you are dependent and these fees are not ‘regulated’ and could theoretically end up higher than todays’ transaction fees – and once you are hooked in, it’s always hard to get out again – at a minimum you spend all your integration efforts for nothing.
For reference, the graph below shows the transaction fees for bitcoin mining over the last 12 months.
Superintelligence
(Adapted from a brilliant article here)
While all these technologies seem fabulous and futuristic, it's not easy to imagine they'll be commonplace in a few years and our lives will be changed forever. The Skynet doomsday scenario seems quite far away too, since it requires "Superintelligence" - an AI that has surpassed human intelligence and then goes on to further improve exponentially.
But historical trends suggest that the technology evolution trend has far outpaced any form of biological evolution.
At the moment, our internalized definition of intelligence intuitively makes us feel that AI has a long way to go before it can get anywhere near human intelligence. And that's a fair notion - we're still very far from having Strong AI. But we also think that once we have an AI that's close to primate level of intelligence, it'll still be far away from being as smart of the smartest humans.
But fact is that most processes we take for granted (combinations such as learning to cycle, swim AND whistle) are incredibly complicated to replicate through a common code base. But once these problems are solved and a machine is able to learn basic functions on it's own, learning the rest will be a piece of cake for the AI. From the AI's POV, the information we've stored through structured record keeping is nothing compared to the information stored in our genes. Which means, once an AI approaches anywhere near primate intelligence, it'll be smarter than the smartest person in no time.
On the bright side, if / when this happens...
...we'll at least know what to expect.
Comments